chore: release 2.1.0; guard auto-bump against deliberate version changes

Set package.json to 2.1.0 to match the CHANGELOG entry for the upgrade tripwire (a [BREAKING] change warrants a minor bump). The startup tripwire reads package.json as the source of truth, so this is the version the gate will enforce. bump-version.yml previously ran `pnpm version patch` on every push to main, which would patch a deliberate 2.1.0 up to 2.1.1. It now skips the auto-bump when the pushed commits already changed package.json themselves. fetch-depth: 0 so the before/after diff has both tips. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
docs(changelog): release the upgrade-tripwire entry as 2.1.0
2026-06-18 18:29:35 +08:00 · 2026-06-07 17:03:02 +03:00 · 2026-06-07 16:59:30 +03:00 · 2026-06-07 16:57:13 +03:00 · 2026-06-06 13:02:12 +03:00 · 2026-06-05 08:04:24 +00:00
266 changed files with 22007 additions and 1719 deletions
@@ -0,0 +1,246 @@
+---
+name: add-atomic-chat-tool
+description: Add Atomic Chat MCP server so the container agent can call local models served by the Atomic Chat desktop app via its OpenAI-compatible API.
+---
+
+# Add Atomic Chat Integration
+
+This skill adds a stdio-based MCP server that exposes models running in the local [Atomic Chat](https://github.com/AtomicBot-ai/Atomic-Chat) desktop app as tools for the container agent. Claude remains the orchestrator but can offload work to local models served by Atomic Chat on `http://127.0.0.1:1337/v1` (OpenAI-compatible).
+
+Tools exposed:
+- `atomic_chat_list_models` — list models currently available in Atomic Chat (`GET /v1/models`)
+- `atomic_chat_generate` — send a prompt to a specified model and return the response (`POST /v1/chat/completions`)
+
+Model management (download, delete) is done through the **Atomic Chat desktop UI** — the app is a fork of Jan and manages its own model library.
+
+The skill ships the MCP server source in this folder and copies it into the agent-runner tree at install time, then wires it up with small edits to `index.ts`, `providers/claude.ts`, and `container-runner.ts`. No branch merge — all edits are additive and idempotent.
+
+## Phase 1: Pre-flight
+
+### Check if already applied
+
+Check if `container/agent-runner/src/atomic-chat-mcp-stdio.ts` exists. If it does, skip to Phase 3 (Configure).
+
+### Check prerequisites
+
+Verify Atomic Chat is installed and its local API server is running. On the host:
+
+```bash
+curl -s http://127.0.0.1:1337/v1/models | head
+```
+
+If the request fails:
+
+1. Install Atomic Chat from the [latest release](https://github.com/AtomicBot-ai/Atomic-Chat/releases) (macOS only for now — `atomic-chat.dmg`).
+2. Open the app.
+3. Open **Settings → Local API Server** and make sure it's enabled on port `1337`.
+4. Go to the **Hub** (or **Models**) tab and download at least one model (e.g. Llama 3.2 3B, Qwen 2.5 Coder 7B).
+5. Load the model once by sending any message in Atomic Chat's UI to warm it up.
+
+## Phase 2: Apply Code Changes
+
+### Copy the MCP server source
+
+```bash
+cp .claude/skills/add-atomic-chat-tool/atomic-chat-mcp-stdio.ts container/agent-runner/src/atomic-chat-mcp-stdio.ts
+```
+
+### Register the MCP server in the agent-runner
+
+Edit `container/agent-runner/src/index.ts`. Find the `mcpServers` object that currently looks like this:
+
+```ts
+  const mcpServers: Record<string, { command: string; args: string[]; env: Record<string, string> }> = {
+    nanoclaw: {
+      command: 'bun',
+      args: ['run', mcpServerPath],
+      env: {},
+    },
+  };
+```
+
+Add an `atomic_chat` entry alongside `nanoclaw`:
+
+```ts
+  const mcpServers: Record<string, { command: string; args: string[]; env: Record<string, string> }> = {
+    nanoclaw: {
+      command: 'bun',
+      args: ['run', mcpServerPath],
+      env: {},
+    },
+    atomic_chat: {
+      command: 'bun',
+      args: ['run', path.join(__dirname, 'atomic-chat-mcp-stdio.ts')],
+      env: {
+        ...(process.env.ATOMIC_CHAT_HOST ? { ATOMIC_CHAT_HOST: process.env.ATOMIC_CHAT_HOST } : {}),
+        ...(process.env.ATOMIC_CHAT_API_KEY ? { ATOMIC_CHAT_API_KEY: process.env.ATOMIC_CHAT_API_KEY } : {}),
+      },
+    },
+  };
+```
+
+### Add the tool glob to the allowlist
+
+Edit `container/agent-runner/src/providers/claude.ts`. Find `'mcp__nanoclaw__*',` in the `TOOL_ALLOWLIST` array and add `'mcp__atomic_chat__*',` on the following line:
+
+```ts
+  'mcp__nanoclaw__*',
+  'mcp__atomic_chat__*',
+];
+```
+
+### Forward host env vars into the container
+
+Edit `src/container-runner.ts` in `buildContainerArgs`. Find the `TZ` env line:
+
+```ts
+  args.push('-e', `TZ=${TIMEZONE}`);
+```
+
+Add ATOMIC_CHAT forwarding right after it:
+
+```ts
+  args.push('-e', `TZ=${TIMEZONE}`);
+
+  // Atomic Chat MCP tool: forward host overrides if set (default is host.docker.internal:1337).
+  if (process.env.ATOMIC_CHAT_HOST) {
+    args.push('-e', `ATOMIC_CHAT_HOST=${process.env.ATOMIC_CHAT_HOST}`);
+  }
+  if (process.env.ATOMIC_CHAT_API_KEY) {
+    args.push('-e', `ATOMIC_CHAT_API_KEY=${process.env.ATOMIC_CHAT_API_KEY}`);
+  }
+```
+
+### Surface `[ATOMIC]` log lines at info level
+
+In the same file, find the stderr logger:
+
+```ts
+  container.stderr?.on('data', (data) => {
+    for (const line of data.toString().trim().split('\n')) {
+      if (line) log.debug(line, { container: agentGroup.folder });
+    }
+  });
+```
+
+Replace it with:
+
+```ts
+  container.stderr?.on('data', (data) => {
+    for (const line of data.toString().trim().split('\n')) {
+      if (!line) continue;
+      if (line.includes('[ATOMIC]')) {
+        log.info(line, { container: agentGroup.folder });
+      } else {
+        log.debug(line, { container: agentGroup.folder });
+      }
+    }
+  });
+```
+
+### Add env-var stubs to `.env.example`
+
+Append to `.env.example`:
+
+```bash
+# Atomic Chat MCP tool (.claude/skills/add-atomic-chat-tool)
+# Override the host where Atomic Chat exposes its OpenAI-compatible API.
+# Default: http://host.docker.internal:1337 (with fallback to localhost)
+# ATOMIC_CHAT_HOST=http://host.docker.internal:1337
+
+# Optional API key. Leave unset for a local Atomic Chat install — it does not require auth.
+# ATOMIC_CHAT_API_KEY=
+```
+
+### Validate code changes
+
+```bash
+pnpm run build
+pnpm exec tsc -p container/agent-runner/tsconfig.json --noEmit
+./container/build.sh
+```
+
+All three must be clean before proceeding.
+
+## Phase 3: Configure
+
+### Set Atomic Chat host (optional)
+
+By default, the MCP server connects to `http://host.docker.internal:1337` (Docker Desktop) with a fallback to `localhost`. To use a custom host, add to `.env`:
+
+```bash
+ATOMIC_CHAT_HOST=http://your-atomic-chat-host:1337
+```
+
+### Set API key (optional)
+
+Atomic Chat does **not require authentication** when running locally — leave this unset. Only set it if you've put Atomic Chat behind a reverse proxy that enforces auth:
+
+```bash
+ATOMIC_CHAT_API_KEY=sk-...
+```
+
+### Restart the service
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+# Linux: systemctl --user restart $(systemd_unit)
+```
+
+## Phase 4: Verify
+
+### Test inference
+
+Tell the user:
+
+> Send a message like: "use atomic chat to tell me the capital of France"
+>
+> The agent should use `atomic_chat_list_models` to find available models, then `atomic_chat_generate` to get a response.
+
+### Check logs if needed
+
+```bash
+tail -f logs/nanoclaw.log | grep -i atomic
+```
+
+Look for:
+- `[ATOMIC] Listing models...` — list request started
+- `[ATOMIC] Found N models` — models discovered
+- `[ATOMIC] >>> Generating with <model>` — generation started
+- `[ATOMIC] <<< Done: <model> | Xs | N tokens | M chars` — generation completed
+
+## Troubleshooting
+
+### Agent says "Atomic Chat is not installed" or tries to run a CLI
+
+The agent is looking for a CLI that doesn't exist instead of using the MCP tools. This means:
+1. The MCP server wasn't copied — check `container/agent-runner/src/atomic-chat-mcp-stdio.ts` exists
+2. The MCP server wasn't registered — check `container/agent-runner/src/index.ts` has the `atomic_chat` entry in `mcpServers`
+3. The allowlist wasn't updated — check `container/agent-runner/src/providers/claude.ts` includes `mcp__atomic_chat__*` in `TOOL_ALLOWLIST`
+4. The container wasn't rebuilt — run `./container/build.sh`
+
+### "Failed to connect to Atomic Chat"
+
+1. Verify the host API is reachable: `curl http://127.0.0.1:1337/v1/models`
+2. Confirm the Local API Server is enabled in Atomic Chat's settings
+3. Check Docker can reach the host: `docker run --rm curlimages/curl curl -s http://host.docker.internal:1337/v1/models`
+4. If using a custom host, check `ATOMIC_CHAT_HOST` in `.env`
+
+### `model not found` / 404 on generate
+
+The model ID passed to `atomic_chat_generate` must exactly match one of the IDs returned by `atomic_chat_list_models`. Ask the agent to list models first, then pick one from that list.
+
+### Slow first response
+
+Atomic Chat lazy-loads models into memory on first use. The initial call may take longer while the model warms up. Subsequent calls against the same model are fast.
+
+### Agent doesn't use Atomic Chat tools
+
+The agent may not know about the tools. Try being explicit: "use the atomic_chat_generate tool with llama3.2-3b-instruct to answer: ..."
+
+### Context window or output size issues
+
+Atomic Chat respects each model's native context length. If you hit limits, pass `max_tokens` explicitly when calling `atomic_chat_generate`, or switch to a model with a larger context window in the Atomic Chat UI.
@@ -0,0 +1,229 @@
+/**
+ * Atomic Chat MCP Server for NanoClaw
+ * Exposes local Atomic Chat models (OpenAI-compatible, /v1) as tools for the container agent.
+ * Uses host.docker.internal to reach the host's Atomic Chat desktop app from Docker.
+ */
+
+import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
+import { z } from 'zod';
+
+import fs from 'fs';
+import path from 'path';
+
+const ATOMIC_CHAT_HOST =
+  process.env.ATOMIC_CHAT_HOST || 'http://host.docker.internal:1337';
+const ATOMIC_CHAT_API_KEY = process.env.ATOMIC_CHAT_API_KEY || '';
+const ATOMIC_CHAT_STATUS_FILE = '/workspace/ipc/atomic_chat_status.json';
+
+function log(msg: string): void {
+  console.error(`[ATOMIC] ${msg}`);
+}
+
+function writeStatus(status: string, detail?: string): void {
+  try {
+    const data = { status, detail, timestamp: new Date().toISOString() };
+    const tmpPath = `${ATOMIC_CHAT_STATUS_FILE}.tmp`;
+    fs.mkdirSync(path.dirname(ATOMIC_CHAT_STATUS_FILE), { recursive: true });
+    fs.writeFileSync(tmpPath, JSON.stringify(data));
+    fs.renameSync(tmpPath, ATOMIC_CHAT_STATUS_FILE);
+  } catch {
+    /* best-effort */
+  }
+}
+
+async function atomicFetch(
+  apiPath: string,
+  options?: RequestInit,
+): Promise<Response> {
+  const url = `${ATOMIC_CHAT_HOST}${apiPath}`;
+  const headers: Record<string, string> = {
+    ...((options?.headers as Record<string, string>) || {}),
+  };
+  if (ATOMIC_CHAT_API_KEY) {
+    headers.Authorization = `Bearer ${ATOMIC_CHAT_API_KEY}`;
+  }
+  const finalOptions: RequestInit = { ...options, headers };
+  try {
+    return await fetch(url, finalOptions);
+  } catch (err) {
+    // Fallback to localhost if host.docker.internal fails
+    if (ATOMIC_CHAT_HOST.includes('host.docker.internal')) {
+      const fallbackUrl = url.replace('host.docker.internal', 'localhost');
+      return await fetch(fallbackUrl, finalOptions);
+    }
+    throw err;
+  }
+}
+
+const server = new McpServer({
+  name: 'atomic_chat',
+  version: '1.0.0',
+});
+
+server.tool(
+  'atomic_chat_list_models',
+  'List all models available in the local Atomic Chat desktop app. Use this to see which models are loaded before calling atomic_chat_generate.',
+  {},
+  async () => {
+    log('Listing models...');
+    writeStatus('listing', 'Listing available models');
+    try {
+      const res = await atomicFetch('/v1/models');
+      if (!res.ok) {
+        return {
+          content: [
+            {
+              type: 'text' as const,
+              text: `Atomic Chat API error: ${res.status} ${res.statusText}`,
+            },
+          ],
+          isError: true,
+        };
+      }
+
+      const data = (await res.json()) as {
+        data?: Array<{ id: string; owned_by?: string }>;
+      };
+      const models = data.data || [];
+
+      if (models.length === 0) {
+        return {
+          content: [
+            {
+              type: 'text' as const,
+              text: 'No models available. Open Atomic Chat on the host and download a model from the Hub.',
+            },
+          ],
+        };
+      }
+
+      const list = models
+        .map((m) => `- ${m.id}${m.owned_by ? ` (${m.owned_by})` : ''}`)
+        .join('\n');
+
+      log(`Found ${models.length} models`);
+      return {
+        content: [
+          { type: 'text' as const, text: `Available models:\n${list}` },
+        ],
+      };
+    } catch (err) {
+      return {
+        content: [
+          {
+            type: 'text' as const,
+            text: `Failed to connect to Atomic Chat at ${ATOMIC_CHAT_HOST}: ${err instanceof Error ? err.message : String(err)}`,
+          },
+        ],
+        isError: true,
+      };
+    }
+  },
+);
+
+server.tool(
+  'atomic_chat_generate',
+  'Send a prompt to a local Atomic Chat model and get a response. Good for cheaper/faster tasks like summarization, translation, or general queries. Use atomic_chat_list_models first to see available models.',
+  {
+    model: z
+      .string()
+      .describe(
+        'The model ID as returned by atomic_chat_list_models (e.g. "llama3.2-3b-instruct")',
+      ),
+    prompt: z.string().describe('The prompt to send to the model'),
+    system: z
+      .string()
+      .optional()
+      .describe('Optional system prompt to set model behavior'),
+    temperature: z
+      .number()
+      .optional()
+      .describe('Sampling temperature (0.0–2.0). Defaults to model default.'),
+    max_tokens: z
+      .number()
+      .optional()
+      .describe('Maximum number of tokens to generate in the response.'),
+  },
+  async (args) => {
+    log(`>>> Generating with ${args.model} (${args.prompt.length} chars)...`);
+    writeStatus('generating', `Generating with ${args.model}`);
+    try {
+      const messages: Array<{ role: string; content: string }> = [];
+      if (args.system) {
+        messages.push({ role: 'system', content: args.system });
+      }
+      messages.push({ role: 'user', content: args.prompt });
+
+      const body: Record<string, unknown> = {
+        model: args.model,
+        messages,
+        stream: false,
+      };
+      if (args.temperature !== undefined) body.temperature = args.temperature;
+      if (args.max_tokens !== undefined) body.max_tokens = args.max_tokens;
+
+      const startedAt = Date.now();
+      const res = await atomicFetch('/v1/chat/completions', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify(body),
+      });
+
+      if (!res.ok) {
+        const errorText = await res.text();
+        return {
+          content: [
+            {
+              type: 'text' as const,
+              text: `Atomic Chat error (${res.status}): ${errorText}`,
+            },
+          ],
+          isError: true,
+        };
+      }
+
+      const data = (await res.json()) as {
+        choices?: Array<{ message?: { content?: string } }>;
+        usage?: {
+          prompt_tokens?: number;
+          completion_tokens?: number;
+          total_tokens?: number;
+        };
+      };
+
+      const response = data.choices?.[0]?.message?.content ?? '';
+      const elapsedSec = ((Date.now() - startedAt) / 1000).toFixed(1);
+      const completionTokens = data.usage?.completion_tokens;
+
+      const meta = `\n\n[${args.model} | ${elapsedSec}s${
+        completionTokens !== undefined ? ` | ${completionTokens} tokens` : ''
+      }]`;
+
+      log(
+        `<<< Done: ${args.model} | ${elapsedSec}s | ${
+          completionTokens ?? '?'
+        } tokens | ${response.length} chars`,
+      );
+      writeStatus(
+        'done',
+        `${args.model} | ${elapsedSec}s | ${completionTokens ?? '?'} tokens`,
+      );
+
+      return { content: [{ type: 'text' as const, text: response + meta }] };
+    } catch (err) {
+      return {
+        content: [
+          {
+            type: 'text' as const,
+            text: `Failed to call Atomic Chat: ${err instanceof Error ? err.message : String(err)}`,
+          },
+        ],
+        isError: true,
+      };
+    }
+  },
+);
+
+const transport = new StdioServerTransport();
+await server.connect(transport);
@@ -0,0 +1,161 @@
+---
+name: add-codex
+description: Use Codex (CLI + AppServer) as the full agent provider — planning, tool orchestration, native compaction, MCP tools, session resume — in place of the Claude Agent SDK. ChatGPT subscription or OPENAI_API_KEY. Per-group via agent_provider. Distinct from using OpenAI as an MCP tool (where Claude remains the planner).
+---
+
+# Codex agent provider
+
+NanoClaw runs agents in a long-lived **poll loop** inside the container. The backend is selected with **`AGENT_PROVIDER`** (`claude` | `opencode` | `codex` | `mock`).
+
+Trunk ships with only the `claude` provider baked in. This skill copies the Codex provider files in from the `providers` branch, wires them into the host and container barrels, updates the Dockerfile to install the Codex CLI, and rebuilds the image.
+
+The Codex provider runs `codex app-server` as a child process and speaks JSON-RPC over stdio. That gives it native session resume, streaming events, MCP tool access, and `thread/compact/start` compaction — same feature bar as the Claude Agent SDK, without the Anthropic-only lock-in.
+
+## Install
+
+### Pre-flight
+
+If all of the following are already present, skip to **Configuration**:
+
+- `src/providers/codex.ts`
+- `container/agent-runner/src/providers/codex.ts`
+- `container/agent-runner/src/providers/codex-app-server.ts`
+- `container/agent-runner/src/providers/codex.factory.test.ts`
+- `import './codex.js';` line in `src/providers/index.ts`
+- `import './codex.js';` line in `container/agent-runner/src/providers/index.ts`
+- `ARG CODEX_VERSION` and `"@openai/codex@${CODEX_VERSION}"` in the pnpm global-install block in `container/Dockerfile`
+
+Missing pieces — continue below. All steps are idempotent; re-running is safe.
+
+### 1. Fetch the providers branch
+
+```bash
+git fetch origin providers
+```
+
+### 2. Copy the Codex source files
+
+Wholesale copies (owned entirely by this skill — user edits to these files won't survive a re-run, as designed):
+
+```bash
+git show origin/providers:src/providers/codex.ts                                      > src/providers/codex.ts
+git show origin/providers:container/agent-runner/src/providers/codex.ts               > container/agent-runner/src/providers/codex.ts
+git show origin/providers:container/agent-runner/src/providers/codex-app-server.ts    > container/agent-runner/src/providers/codex-app-server.ts
+git show origin/providers:container/agent-runner/src/providers/codex.factory.test.ts  > container/agent-runner/src/providers/codex.factory.test.ts
+```
+
+### 3. Append the self-registration imports
+
+Each barrel gets one line — alphabetical placement keeps diffs small.
+
+`src/providers/index.ts`:
+
+```typescript
+import './codex.js';
+```
+
+`container/agent-runner/src/providers/index.ts`:
+
+```typescript
+import './codex.js';
+```
+
+### 4. Add the Codex CLI to the container Dockerfile
+
+Two edits to `container/Dockerfile`, both idempotent (skip if already present):
+
+**(a)** In the "Pin CLI versions" ARG block (around line 18), add after `ARG CLAUDE_CODE_VERSION=...`:
+
+```dockerfile
+ARG CODEX_VERSION=0.124.0
+```
+
+**(b)** Add a new standalone `RUN` block for the Codex CLI, after the existing per-CLI install blocks (around line 106, right after the `@anthropic-ai/claude-code` block). The Dockerfile splits each global CLI into its own layer for cache granularity — keep that pattern; do not collapse them into a single combined `pnpm install -g` call:
+
+```dockerfile
+RUN --mount=type=cache,target=/root/.cache/pnpm \
+    pnpm install -g "@openai/codex@${CODEX_VERSION}"
+```
+
+Note: **no agent-runner package dependency** — Codex is a CLI binary, not a library. Unlike OpenCode, there's nothing to add to `container/agent-runner/package.json`.
+
+### 5. Build
+
+```bash
+pnpm run build                                         # host
+pnpm exec tsc -p container/agent-runner/tsconfig.json --noEmit   # container typecheck
+./container/build.sh                                   # agent image
+```
+
+## Configuration
+
+Codex supports two primary auth paths and one experimental BYO-endpoint path. Pick the one that matches your setup.
+
+### Option A — ChatGPT subscription (recommended for individuals)
+
+On the host (not inside the container), run Codex's OAuth login:
+
+```bash
+codex login
+```
+
+This writes `~/.codex/auth.json` with a subscription token. The host-side Codex provider ([src/providers/codex.ts](../../../src/providers/codex.ts)) copies `auth.json` into a per-session `~/.codex` directory mounted into the container — your host's own Codex CLI is never touched.
+
+No `.env` variables required for this mode.
+
+### Option B — API key (recommended for CI or API billing)
+
+```env
+OPENAI_API_KEY=sk-...
+CODEX_MODEL=gpt-5.4-mini
+```
+
+The host forwards both variables into the container. If both subscription (`auth.json`) and `OPENAI_API_KEY` are present, Codex prefers the subscription.
+
+### Option C — BYO OpenAI-compatible endpoint (experimental)
+
+Codex's built-in `openai` provider honors the `OPENAI_BASE_URL` env var directly. Point it at any OpenAI-compatible endpoint — Groq, Together, self-hosted vLLM, an OpenAI proxy, etc.
+
+```env
+OPENAI_API_KEY=...
+OPENAI_BASE_URL=https://api.groq.com/openai/v1
+CODEX_MODEL=llama-3.3-70b-versatile
+```
+
+Codex also ships first-class local-runner flags — `codex --oss --local-provider ollama` or `--local-provider lmstudio` — that auto-detect a local server. To use those inside NanoClaw, set `CODEX_MODEL` to a model your local runner serves and add the corresponding base URL; see the Codex CLI docs for the full `model_provider = oss` configuration.
+
+**Experimental caveat:** tool-calling quality depends on the model and endpoint. Not every OpenAI-compat provider implements the full function-calling spec, and smaller models (< 30B) often struggle with multi-step tool orchestration. Test before committing.
+
+### Per group / per session
+
+Set `"provider": "codex"` in the group's **`container.json`** (`groups/<folder>/container.json`) — the in-container runner reads `provider` from there, not from the DB. The DB columns **`agent_groups.agent_provider`** and **`sessions.agent_provider`** (session overrides group) only drive host-side provider contribution — per-session `~/.codex` mount, `OPENAI_*` / `CODEX_MODEL` env passthrough — and do not propagate into `container.json` at spawn time. Set both, or just edit `container.json`; if they disagree, the runner uses `container.json` and the host-side resolver falls back through session → group → `container.json` → `'claude'`.
+
+`CODEX_MODEL` applies process-wide via `.env`; if you need different models for different groups, set them via `container_config.env` on the group.
+
+Extra MCP servers still come from **`NANOCLAW_MCP_SERVERS`** / `container_config.mcpServers` on the host. The runner merges them into the same `mcpServers` object passed to all providers.
+
+## Operational notes
+
+- **Spawn-per-query:** Codex's app-server is spawned fresh per query invocation, matching the OpenCode pattern. No long-lived daemon to keep healthy across sessions.
+- **Per-session `~/.codex` isolation:** each group gets its own copy of the host's `auth.json`. The container can rewrite `config.toml` freely on every wake without touching the host's Codex config.
+- **Native compaction:** kicks in automatically at 40K cumulative input tokens between turns, via `thread/compact/start`. If compaction fails, the provider logs and continues uncompacted — no fatal error.
+- **Approvals:** auto-accepted inside the container (the container is the sandbox; same posture as Claude/OpenCode).
+- **Mid-turn input:** Codex turns don't accept mid-turn messages. Follow-up `push()` calls queue and drain between turns, matching the OpenCode pattern. The poll-loop only pushes between turns anyway, so no messages are dropped.
+- **Stale thread recovery:** `isSessionInvalid` matches on stale-thread-ID errors (`thread not found`, `unknown thread`, etc.) so a cold-started app-server can recover cleanly when it sees a stored continuation it no longer has.
+
+## Verify
+
+```bash
+grep -q "./codex.js" container/agent-runner/src/providers/index.ts && echo "container barrel: OK"
+grep -q "./codex.js" src/providers/index.ts && echo "host barrel: OK"
+grep -q "@openai/codex@" container/Dockerfile && echo "Dockerfile install: OK"
+cd container/agent-runner && bun test src/providers/codex.factory.test.ts && cd -
+```
+
+After the image rebuild, set `agent_provider = 'codex'` on a test group and send a message. Successful round-trip looks like:
+
+- `init` event with a stable thread ID as continuation
+- One or more `activity` / `progress` events during the turn
+- `result` event with the model's reply
+
+If the agent hangs or errors, check `~/.codex/auth.json` exists on the host (Option A) or that `OPENAI_API_KEY` is forwarding correctly (Option B) — `docker exec` into a running container and `env | grep -i openai` to confirm.
@@ -93,10 +93,13 @@ Generate the secret: `node -e "console.log('nc-' + require('crypto').randomBytes

 ### 6. Build and restart

+Run from your NanoClaw project root:
+
 ```bash
 pnpm run build
-systemctl --user restart nanoclaw   # Linux
-# or: launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
+source setup/lib/install-slug.sh
+systemctl --user restart $(systemd_unit)              # Linux
+# or: launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
 ```

 ### 7. Verify
@@ -0,0 +1,65 @@
+# Remove DeltaChat
+
+## 1. Disable the adapter
+
+Comment out the import in `src/channels/index.ts`:
+
+```typescript
+// import './deltachat.js';
+```
+
+## 2. Remove credentials
+
+Remove the `DC_*` lines from `.env`:
+
+```bash
+DC_EMAIL
+DC_PASSWORD
+DC_IMAP_HOST
+DC_IMAP_PORT
+DC_SMTP_HOST
+DC_SMTP_PORT
+```
+
+## 3. Rebuild and restart
+
+Run from your NanoClaw project root:
+
+```bash
+pnpm run build
+source setup/lib/install-slug.sh
+
+# Linux
+systemctl --user restart $(systemd_unit)
+
+# macOS
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)
+```
+
+## 4. Remove account data (optional)
+
+To fully remove all account data including DeltaChat encryption keys:
+
+```bash
+rm -rf dc-account/
+```
+
+> **Warning:** This deletes the Autocrypt keys. Contacts who have verified your bot's key will need to re-verify if the same email address is re-used with a new account.
+
+To keep the account for later reinstall, leave `dc-account/` intact.
+
+## 5. Remove the package (optional)
+
+```bash
+pnpm remove @deltachat/stdio-rpc-server
+```
+
+## Verification
+
+After removal, confirm the adapter is no longer starting:
+
+```bash
+grep "deltachat" logs/nanoclaw.log | tail -5
+```
+
+Expected: no `Channel adapter started` entry after the last restart.
@@ -0,0 +1,258 @@
+---
+name: add-deltachat
+description: Add DeltaChat channel integration via @deltachat/stdio-rpc-server. Native adapter — no Chat SDK bridge. Email-based messaging with end-to-end encryption.
+---
+
+# Add DeltaChat Channel
+
+The adapter drives the `@deltachat/stdio-rpc-server` JSON-RPC subprocess directly — pure Node.js against the DeltaChat core library. Messages are delivered over email with Autocrypt/OpenPGP encryption.
+
+## Install
+
+### Pre-flight (idempotent)
+
+Skip to **Credentials** if all of these are already in place:
+
+- `src/channels/deltachat.ts` exists
+- `src/channels/index.ts` contains `import './deltachat.js';`
+- `@deltachat/stdio-rpc-server` is listed in `package.json` dependencies
+
+Otherwise continue. Every step below is safe to re-run.
+
+### 1. Fetch the channels branch
+
+```bash
+git fetch origin channels
+```
+
+### 2. Copy the adapter
+
+```bash
+git show origin/channels:src/channels/deltachat.ts > src/channels/deltachat.ts
+```
+
+### 3. Append the self-registration import
+
+Append to `src/channels/index.ts` (skip if already present):
+
+```typescript
+import './deltachat.js';
+```
+
+### 4. Install the adapter package (pinned)
+
+```bash
+pnpm install @deltachat/stdio-rpc-server@2.49.0
+```
+
+### 5. Build
+
+```bash
+pnpm run build
+```
+
+## Account Setup
+
+A dedicated email account is strongly recommended — it will accumulate DeltaChat-formatted messages and store encryption keys. Not all providers work well with DeltaChat; check https://providers.delta.chat/ before picking one.
+
+**Default security modes:** IMAP uses SSL/TLS (port 993), SMTP uses STARTTLS (port 587). Both are configurable via `.env` — see Credentials below.
+
+To find the correct hostnames for a domain:
+
+```bash
+node -e "require('dns').resolveMx('example.com', (e,r) => console.log(r))"
+```
+
+Most providers publish their IMAP/SMTP hostnames in their help docs under "manual setup" or "IMAP access."
+
+## Credentials
+
+Add to `.env`:
+
+```bash
+DC_EMAIL=bot@example.com
+DC_PASSWORD=your-app-password
+DC_IMAP_HOST=imap.example.com
+DC_IMAP_PORT=993
+DC_IMAP_SECURITY=1        # 1=SSL/TLS (default), 2=STARTTLS, 3=plain
+DC_SMTP_HOST=smtp.example.com
+DC_SMTP_PORT=587
+DC_SMTP_SECURITY=2        # 2=STARTTLS (default), 1=SSL/TLS, 3=plain
+```
+
+Security settings are applied on every startup, so changing them in `.env` and restarting takes effect without wiping the account.
+
+Sync to container: `mkdir -p data/env && cp .env data/env/env`
+
+### Optional settings
+
+The following are read from the process environment (not `.env`). To override them, add `Environment=` lines to the systemd service unit or your launchd plist:
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `DC_ACCOUNT_DIR` | `dc-account` | Directory for DeltaChat account data (IMAP state, keys, blobs) |
+| `DC_DISPLAY_NAME` | `NanoClaw` | Bot display name shown in DeltaChat |
+| `DC_AVATAR_PATH` | _(none)_ | Absolute path to avatar image; set at startup only |
+
+The `/set-avatar` command (send an image with that caption) is the easiest way to set the avatar at runtime without modifying the service file. Only users with `owner` or global `admin` role can use it.
+
+### Restart
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+
+# Linux
+systemctl --user restart $(systemd_unit)
+
+# macOS
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)
+```
+
+On first start the adapter configures the email account (IMAP/SMTP credentials, calls `configure()`). Subsequent starts skip straight to `startIo()`. Account data is stored in `dc-account/` in the project root (or your `DC_ACCOUNT_DIR`).
+
+## Wiring
+
+### DMs
+
+**DeltaChat contacts cannot be added by email alone** — to start a chat, the user must open the bot's invite link in their DeltaChat app or scan its QR code. This triggers the SecureJoin handshake.
+
+#### Step 1 — Get the invite link
+
+After the service starts, the adapter logs the invite URL and writes a QR SVG:
+
+```bash
+grep "invite link" logs/nanoclaw.log | tail -1
+# url field contains the https://i.delta.chat/... invite link
+# also written to dc-account/invite-qr.svg (or $DC_ACCOUNT_DIR/invite-qr.svg)
+```
+
+The invite URL is stable (tied to the bot's email and encryption keys) so it stays valid across restarts.
+
+#### Step 2 — Add the bot in DeltaChat
+
+Two options for the user to connect:
+
+- **Link**: Copy the `https://i.delta.chat/...` URL and open it on the device running DeltaChat. The app recognises it and shows a "Start chat" prompt.
+- **QR code**: Open `dc-account/invite-qr.svg` in a browser or image viewer, display it on screen, and scan it from the DeltaChat app using the QR-scan button on the new-chat screen.
+
+After accepting, DeltaChat exchanges keys and creates the chat automatically.
+
+#### Step 3 — Wire the chat to an agent
+
+Once the first message arrives the router auto-creates a `messaging_groups` row. Look up the chat ID:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "SELECT platform_id, name FROM messaging_groups WHERE channel_type='deltachat' AND is_group=0 ORDER BY created_at DESC LIMIT 5"
+```
+
+Then run `/init-first-agent` — it creates the agent group, grants the user owner access, and wires the messaging group in one step:
+
+```bash
+pnpm exec tsx scripts/init-first-agent.ts \
+  --channel deltachat \
+  --user-id deltachat:user@example.com \
+  --platform-id <platform_id from above> \
+  --display-name "Your Name"
+```
+
+### Groups
+
+Add the bot email to a DeltaChat group. When any member sends a message, the router creates a `messaging_groups` row with `is_group = 1`. Run `/manage-channels` to wire it to an agent group.
+
+## Next Steps
+
+If you're in the middle of `/setup`, return to the setup flow now.
+
+Otherwise, run `/init-first-agent` to create an agent and wire it to your DeltaChat DM (see Wiring above), or `/manage-channels` to wire this channel to an existing agent group.
+
+## Channel Info
+
+- **type**: `deltachat`
+- **terminology**: DeltaChat calls them "chats" (1:1 DMs) and "groups"
+- **supports-threads**: no — DeltaChat has no thread model
+- **platform-id-format**: numeric chat ID as a string (e.g. `"12"`) — the DeltaChat core's internal chat identifier
+- **user-id-format**: `deltachat:{email}` — the contact's email address
+- **how-to-find-id**: Send a message from DeltaChat to the bot email, then query `messaging_groups` as shown above
+- **typical-use**: Personal assistant over DeltaChat DMs; small groups where participants use DeltaChat
+- **default-isolation**: One agent per bot identity. Multiple chats with the same operator can share an agent group; groups with other people should typically use `isolated` session mode
+
+### Features
+
+- File attachments — inbound and outbound; inbound waits up to 30 seconds for large-message download to complete
+- Invite link logged on every startup — URL + QR SVG written to `dc-account/invite-qr.svg`; see Wiring for the bootstrap flow
+- `/set-avatar` — send an image with this caption to change the bot's DeltaChat avatar (admin/owner only)
+- Connectivity watchdog — restarts IO if IMAP goes quiet for 20 minutes or connectivity drops below threshold for two consecutive 5-minute checks
+- Network nudge — `maybeNetwork()` called every 10 minutes to recover from prolonged idle
+
+Not supported: DeltaChat reactions, message editing/deletion, read receipts.
+
+### Connectivity model
+
+`isConnected()` returns `true` when the internal connectivity value is ≥ 3000:
+
+| Range | Meaning |
+|-------|---------|
+| 1000–1999 | Not connected |
+| 2000–2999 | Connecting |
+| 3000–3999 | Working (IMAP fetching) |
+| ≥ 4000 | Fully connected (IMAP IDLE) |
+
+## Troubleshooting
+
+### Adapter not starting — credentials missing
+
+```bash
+grep "Channel credentials missing" logs/nanoclaw.log | grep deltachat
+```
+
+All six required vars (`DC_EMAIL`, `DC_PASSWORD`, `DC_IMAP_HOST`, `DC_IMAP_PORT`, `DC_SMTP_HOST`, `DC_SMTP_PORT`) must be present in `.env`.
+
+### Account configure fails
+
+```bash
+grep "DeltaChat" logs/nanoclaw.log | tail -20
+```
+
+Common causes:
+- Wrong IMAP/SMTP hostnames — double-check provider docs
+- App password not generated — Gmail and some others require this when 2FA is enabled
+- Port/security mismatch — defaults are port 993 + SSL/TLS for IMAP and port 587 + STARTTLS for SMTP; override with `DC_IMAP_PORT`/`DC_IMAP_SECURITY` or `DC_SMTP_PORT`/`DC_SMTP_SECURITY` in `.env`
+
+### Provider uses SMTP port 465 (SSL/TLS) instead of 587
+
+Set `DC_SMTP_SECURITY=1` and `DC_SMTP_PORT=465` in `.env`, then restart.
+
+### Messages not arriving
+
+1. Check the service is running and the adapter started: `grep "Channel adapter started.*deltachat" logs/nanoclaw.log`
+2. Check connectivity: `grep "DeltaChat: IO started" logs/nanoclaw.log`
+3. Check the sender has been granted access — run `/init-first-agent` to create their user record and wire the chat
+4. Verify the messaging group is wired: `pnpm exec tsx scripts/q.ts data/v2.db "SELECT mg.platform_id, mga.agent_group_id FROM messaging_groups mg JOIN messaging_group_agents mga ON mg.id = mga.messaging_group_id WHERE mg.channel_type='deltachat'"`
+
+### Stale lock file after crash
+
+```bash
+rm -f dc-account/accounts.lock
+systemctl --user restart "$(. setup/lib/install-slug.sh && systemd_unit)"
+```
+
+### Bot not responding after restart
+
+The account is already configured — IO restarts automatically on service start. If the RPC subprocess is stuck, restart the service. Check for errors:
+
+```bash
+grep "DeltaChat" logs/nanoclaw.error.log | tail -20
+```
+
+### Messages received but agent not responding
+
+The messaging group exists but may not be wired to an agent group. Run:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db "SELECT id, platform_id, name FROM messaging_groups WHERE channel_type='deltachat'"
+```
+
+If the group has no entry in `messaging_group_agents`, wire it with `/manage-channels`.
@@ -0,0 +1,54 @@
+# Verify DeltaChat
+
+## 1. Check the adapter started
+
+```bash
+grep "Channel adapter started.*deltachat" logs/nanoclaw.log | tail -1
+```
+
+Expected: `Channel adapter started { channel: 'deltachat', type: 'deltachat' }`
+
+## 2. Check IMAP/SMTP connectivity
+
+Replace with your provider's hostnames from `.env`:
+
+```bash
+DC_IMAP=$(grep '^DC_IMAP_HOST=' .env | cut -d= -f2)
+DC_SMTP=$(grep '^DC_SMTP_HOST=' .env | cut -d= -f2)
+
+bash -c "echo >/dev/tcp/$DC_IMAP/993" && echo "IMAP open" || echo "IMAP blocked"
+bash -c "echo >/dev/tcp/$DC_SMTP/587" && echo "SMTP open" || echo "SMTP blocked"
+```
+
+## 3. End-to-end message test
+
+1. Open DeltaChat on your device
+2. Add the bot email address as a contact
+3. Send a message
+4. The bot should respond within a few seconds
+
+If nothing arrives, check:
+
+```bash
+grep "DeltaChat" logs/nanoclaw.log | tail -20
+grep "DeltaChat" logs/nanoclaw.error.log | tail -10
+```
+
+## 4. Check messaging group was created
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "SELECT id, platform_id, name FROM messaging_groups WHERE channel_type='deltachat' ORDER BY created_at DESC LIMIT 5"
+```
+
+If a row appears, the inbound routing is working. If not, the adapter isn't receiving the message — check logs for `DeltaChat: error handling incoming message`.
+
+## 5. Verify user access
+
+If the message arrived but the agent didn't respond, the sender may not have access:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db "SELECT id, display_name FROM users WHERE id LIKE 'deltachat:%'"
+```
+
+Grant access as shown in the SKILL.md "Grant user access" section.
@@ -44,7 +44,7 @@ import './discord.js';
 ### 4. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/discord@4.26.0
+pnpm install @chat-adapter/discord@4.27.0
 ```

 ### 5. Build
@@ -162,10 +162,13 @@ If you changed `EMACS_CHANNEL_PORT` from the default:

 ## Restart NanoClaw

+Run from your NanoClaw project root:
+
 ```bash
 pnpm run build
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw   # macOS
-# systemctl --user restart nanoclaw                # Linux
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)   # macOS
+# systemctl --user restart $(systemd_unit)             # Linux
 ```

 ## Verify
@@ -240,8 +243,8 @@ grep -q "import './emacs.js'" src/channels/index.ts && echo "imported" || echo "

 ### No response from agent

-1. NanoClaw running: `launchctl list | grep nanoclaw` (macOS) / `systemctl --user status nanoclaw` (Linux)
-2. Messaging group wired: `sqlite3 data/v2.db "SELECT mg.platform_id, ag.folder FROM messaging_groups mg JOIN messaging_group_agents mga ON mg.id = mga.messaging_group_id JOIN agent_groups ag ON ag.id = mga.agent_group_id WHERE mg.channel_type = 'emacs'"`
+1. NanoClaw running: `launchctl list | grep "$(. setup/lib/install-slug.sh && launchd_label)"` (macOS) / `systemctl --user status "$(. setup/lib/install-slug.sh && systemd_unit)"` (Linux)
+2. Messaging group wired: `pnpm exec tsx scripts/q.ts data/v2.db "SELECT mg.platform_id, ag.folder FROM messaging_groups mg JOIN messaging_group_agents mga ON mg.id = mga.messaging_group_id JOIN agent_groups ag ON ag.id = mga.agent_group_id WHERE mg.channel_type = 'emacs'"`
 3. Logs show inbound: `grep 'channel_type=emacs\|Emacs' logs/nanoclaw.log | tail -20`

 If no messaging group row exists, run the `register` command above.
@@ -282,15 +285,18 @@ If an agent outputs org-mode directly, markers get double-converted and render i

 ## Removal

+Run from your NanoClaw project root:
+
 ```bash
 rm src/channels/emacs.ts src/channels/emacs.test.ts emacs/nanoclaw.el
 # Remove the `import './emacs.js';` line from src/channels/index.ts
 # Remove EMACS_* lines from .env
 pnpm run build
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw   # macOS
-# systemctl --user restart nanoclaw                # Linux
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)   # macOS
+# systemctl --user restart $(systemd_unit)             # Linux

 # Remove the NanoClaw block from your Emacs config
 # Optionally clean up the messaging group:
-sqlite3 data/v2.db "DELETE FROM messaging_group_agents WHERE messaging_group_id IN (SELECT id FROM messaging_groups WHERE channel_type='emacs'); DELETE FROM messaging_groups WHERE channel_type='emacs';"
+pnpm exec tsx scripts/q.ts data/v2.db "DELETE FROM messaging_group_agents WHERE messaging_group_id IN (SELECT id FROM messaging_groups WHERE channel_type='emacs'); DELETE FROM messaging_groups WHERE channel_type='emacs';"
 ```
@@ -0,0 +1,238 @@
+---
+name: add-gcal-tool
+description: Add Google Calendar as an MCP tool (list calendars, list/search/create events, free/busy queries) using OneCLI-managed OAuth. Multi-calendar and multi-account supported. Mirrors /add-gmail-tool's stub pattern — no raw credentials ever reach the container; OneCLI injects real tokens at request time.
+---
+
+# Add Google Calendar Tool (OneCLI-native)
+
+This skill wires [`@cocal/google-calendar-mcp`](https://github.com/cocal-com/google-calendar-mcp) into selected agent groups. The MCP server reads stub credentials containing the `onecli-managed` placeholder; the OneCLI gateway intercepts outbound calls to `calendar.googleapis.com` / `oauth2.googleapis.com` and swaps the bearer for the real OAuth token from its vault.
+
+**Why this package (and not gongrzhe's):** `@gongrzhe/server-calendar-autoauth-mcp` only supports the `primary` calendar and exposes 5 tools (no `list_calendars`). `@cocal/google-calendar-mcp` explicitly supports multi-calendar and multi-account, and is actively maintained.
+
+Tools exposed (surfaced as `mcp__calendar__<name>`, exact set depends on version — run `tools/list` against the MCP server to enumerate): `list-calendars`, `list-events`, `search-events`, `create-event`, `update-event`, `delete-event`, `get-event`, `list-colors`, `get-freebusy`, `get-current-time`, plus multi-account management tools.
+
+**Why this pattern:** v2's invariant is that containers never receive raw API keys (CHANGELOG 2.0.0). Same stub pattern `/add-gmail-tool` uses. This skill is deliberately a sibling, not a combined "Google Workspace" skill — installs independently and removes cleanly.
+
+## Phase 1: Pre-flight
+
+### Verify OneCLI has Google Calendar connected
+
+```bash
+onecli apps get --provider google-calendar
+```
+
+Expected: `"connection": { "status": "connected" }` with scopes including `calendar.readonly` and `calendar.events`.
+
+If not connected, tell the user:
+
+> Open the OneCLI web UI at http://127.0.0.1:10254, go to Apps → Google Calendar, and click Connect. Sign in with the Google account the agent should act as. `calendar.readonly` + `calendar.events` are the minimum useful scopes.
+
+### Verify stub credentials exist
+
+The stub lives at `~/.calendar-mcp/` by convention (shared with `/add-gmail-tool`'s sibling). cocal doesn't default to this path (it uses `~/.config/google-calendar-mcp/tokens.json`) — we override via env vars below so it reads our stubs instead.
+
+```bash
+ls -la ~/.calendar-mcp/gcp-oauth.keys.json ~/.calendar-mcp/credentials.json 2>&1
+```
+
+If both exist with `onecli-managed`:
+
+```bash
+grep -l onecli-managed ~/.calendar-mcp/gcp-oauth.keys.json ~/.calendar-mcp/credentials.json
+```
+
+...skip to Phase 2. If either file has real credentials (no `onecli-managed`), **STOP** — back up and delete before proceeding.
+
+If absent, write them:
+
+```bash
+mkdir -p ~/.calendar-mcp
+cat > ~/.calendar-mcp/gcp-oauth.keys.json <<'EOF'
+{
+  "installed": {
+    "client_id": "onecli-managed.apps.googleusercontent.com",
+    "client_secret": "onecli-managed",
+    "redirect_uris": ["http://localhost:3000/oauth2callback"]
+  }
+}
+EOF
+cat > ~/.calendar-mcp/credentials.json <<'EOF'
+{
+  "access_token": "onecli-managed",
+  "refresh_token": "onecli-managed",
+  "token_type": "Bearer",
+  "expiry_date": 99999999999999,
+  "scope": "https://www.googleapis.com/auth/calendar.readonly https://www.googleapis.com/auth/calendar.events"
+}
+EOF
+chmod 600 ~/.calendar-mcp/*.json
+```
+
+### Verify mount allowlist covers the path
+
+```bash
+cat ~/.config/nanoclaw/mount-allowlist.json
+```
+
+`~/.calendar-mcp` must sit under an `allowedRoots` entry.
+
+### Check agent secret-mode
+
+For each target agent group, confirm OneCLI will inject the Google Calendar token:
+
+```bash
+onecli agents list
+```
+
+`secretMode: all` is sufficient. If `selective`, explicitly assign the Calendar secret.
+
+## Phase 2: Apply Code Changes
+
+### Check if already applied
+
+```bash
+grep -q 'CALENDAR_MCP_VERSION' container/Dockerfile && \
+echo "ALREADY APPLIED — skip to Phase 3"
+```
+
+### Add MCP server to Dockerfile
+
+Edit `container/Dockerfile`. Find the pinned-version ARG block and add:
+
+```dockerfile
+ARG CALENDAR_MCP_VERSION=2.6.1
+```
+
+If `/add-gmail-tool` has already been applied, the pnpm global-install block already exists with its `zod-to-json-schema@3.22.5` pin. Just append the calendar package — **the calendar-mcp uses `zod@4.x` and does NOT need that pin**, but it's harmless to share the block:
+
+```dockerfile
+RUN --mount=type=cache,target=/root/.cache/pnpm \
+    pnpm install -g \
+        "@gongrzhe/server-gmail-autoauth-mcp@${GMAIL_MCP_VERSION}" \
+        "@cocal/google-calendar-mcp@${CALENDAR_MCP_VERSION}" \
+        "zod-to-json-schema@3.22.5"
+```
+
+If `/add-gmail-tool` hasn't been applied, install Calendar standalone:
+
+```dockerfile
+RUN --mount=type=cache,target=/root/.cache/pnpm \
+    pnpm install -g "@cocal/google-calendar-mcp@${CALENDAR_MCP_VERSION}"
+```
+
+**No `TOOL_ALLOWLIST` edit needed.** `container/agent-runner/src/providers/claude.ts` derives the allow-pattern dynamically from each group's `mcpServers` map (`Object.keys(this.mcpServers).map(mcpAllowPattern)`), so registering `calendar` in Phase 3 automatically allows `mcp__calendar__*`. Earlier versions of this skill instructed a static `TOOL_ALLOWLIST` edit — that's now redundant.
+
+### Rebuild the container image
+
+```bash
+./container/build.sh
+```
+
+## Phase 3: Wire Per-Agent-Group
+
+For each agent group, persist two changes to the **central DB** (`data/v2.db`): the `mcpServers.calendar` entry and an `additionalMounts` entry for `.calendar-mcp`. Both flow through `materializeContainerJson` on every spawn, so editing `groups/<folder>/container.json` by hand does **not** stick — that file is regenerated from the DB.
+
+### Register the MCP server
+
+For each chosen `<group-id>` (use `ncl groups list` to enumerate):
+
+```bash
+ncl groups config add-mcp-server \
+  --id <group-id> \
+  --name calendar \
+  --command google-calendar-mcp \
+  --args '[]' \
+  --env '{"GOOGLE_OAUTH_CREDENTIALS":"/workspace/extra/.calendar-mcp/gcp-oauth.keys.json","GOOGLE_CALENDAR_MCP_TOKEN_PATH":"/workspace/extra/.calendar-mcp/credentials.json"}'
+```
+
+Approval behaviour depends on where you run it: from inside an agent's container `ncl` write verbs are approval-gated (admin approves before it lands); from a host operator shell with full scope, it executes immediately. Either way, the response tells you which path it took.
+
+### Add the `.calendar-mcp` mount
+
+There is no `ncl groups config add-mount` verb yet (tracked in [#2395](https://github.com/nanocoai/nanoclaw/issues/2395)). Until that ships, edit the DB directly via the in-tree wrapper (`scripts/q.ts` — `setup/verify.ts:5` codifies that NanoClaw avoids depending on the `sqlite3` CLI binary, so don't shell out to it):
+
+```bash
+GROUP_ID='<group-id>'
+HOST_PATH="$HOME/.calendar-mcp"
+MOUNT=$(jq -cn --arg h "$HOST_PATH" '{hostPath:$h, containerPath:".calendar-mcp", readonly:false}')
+pnpm exec tsx scripts/q.ts data/v2.db "UPDATE container_configs \
+  SET additional_mounts = json_insert(additional_mounts, '\$[#]', json('$MOUNT')), \
+      updated_at = datetime('now') \
+  WHERE agent_group_id = '$GROUP_ID';"
+```
+
+Run from your NanoClaw project root (where `data/v2.db` lives). The `$[#]` placeholder is SQLite JSON1's append-to-end notation; it's `\$`-escaped so bash doesn't arithmetic-expand it before sqlite sees it. `updated_at` is ISO-string everywhere else in the schema, so use `datetime('now')` — not `strftime('%s','now')`, which would silently mix epoch ints into a column of YYYY-MM-DD HH:MM:SS strings.
+
+**Switch to `ncl groups config add-mount` once #2395 lands.** Update this skill at that time.
+
+`containerPath` is relative (mount-security rejects absolute paths — additional mounts land at `/workspace/extra/<relative>`).
+
+**Why this can't be `groups/<folder>/container.json`:** post-migration `014-container-configs`, `materializeContainerJson` in `src/container-config.ts` rewrites that file from the DB on every spawn. Anything hand-edited there is silently overwritten on next restart.
+
+**Same-group-as-gmail tip:** if this group already has the gmail MCP + `.gmail-mcp` mount, both coexist — `ncl groups config add-mcp-server` only updates the named entry, and `json_insert` appends to `additional_mounts` without disturbing existing entries.
+
+## Phase 4: Build and Restart
+
+```bash
+pnpm run build
+```
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
+```
+
+Kill any existing agent containers so they respawn with the new mcpServers config:
+
+```bash
+docker ps -q --filter 'name=nanoclaw-v2-' | xargs -r docker kill
+```
+
+## Phase 5: Verify
+
+### Test from a wired agent
+
+> Send: **"list my calendars"** or **"what's on my work calendar next Monday?"**.
+>
+> First call takes 2–3s while the MCP server starts and OneCLI does the token exchange.
+
+### Check logs if the tool isn't working
+
+```bash
+tail -100 logs/nanoclaw.log | grep -iE 'calendar|mcp'
+```
+
+Common signals:
+- `command not found: google-calendar-mcp` → image not rebuilt.
+- `ENOENT ...credentials.json` → mount missing. Check the mount allowlist.
+- `401 Unauthorized` from `*.googleapis.com` → OneCLI isn't injecting; verify agent's secret mode and that Google Calendar is connected.
+- Agent says "I don't have calendar tools" → the `calendar` MCP server isn't registered in this group's `mcpServers` (re-run the `ncl groups config add-mcp-server` step in Phase 3 for that group and restart it), or the agent-runner image is stale (`./container/build.sh`, `--no-cache` if suspicious).
+
+## Removal
+
+1. For each group that had Calendar wired, remove the MCP server from the DB:
+   ```bash
+   ncl groups config remove-mcp-server --id <group-id> --name calendar
+   ```
+2. Remove the `.calendar-mcp` mount from the DB (no `remove-mount` verb yet — same #2395 dependency):
+   ```bash
+   pnpm exec tsx scripts/q.ts data/v2.db "UPDATE container_configs \
+     SET additional_mounts = (SELECT json_group_array(value) FROM json_each(additional_mounts) \
+                              WHERE json_extract(value, '\$.containerPath') != '.calendar-mcp'), \
+         updated_at = datetime('now') \
+     WHERE agent_group_id = '<group-id>';"
+   ```
+3. Remove `CALENDAR_MCP_VERSION` ARG and the calendar package from the Dockerfile install block.
+4. `pnpm run build && ./container/build.sh && systemctl --user restart "$(. setup/lib/install-slug.sh && systemd_unit)"`.
+5. Optional: `rm -rf ~/.calendar-mcp/` and `onecli apps disconnect --provider google-calendar`.
+
+No `TOOL_ALLOWLIST` removal step — Phase 2 no longer edits it.
+
+## Credits & references
+
+- **MCP server:** [`@cocal/google-calendar-mcp`](https://github.com/cocal-com/google-calendar-mcp) — MIT-licensed, actively maintained, multi-account and multi-calendar.
+- **Why not gongrzhe:** earlier versions of this skill used `@gongrzhe/server-calendar-autoauth-mcp@1.0.2` which only supports the primary calendar with 5 event-level tools. The cocal server supersedes it.
+- **Skill pattern:** direct sibling of [`/add-gmail-tool`](../add-gmail-tool/SKILL.md); same OneCLI stub mechanism.
@@ -44,7 +44,7 @@ import './gchat.js';
 ### 4. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/gchat@4.26.0
+pnpm install @chat-adapter/gchat@4.27.0
 ```

 ### 5. Build
@@ -48,7 +48,7 @@ import './github.js';
 ### 4. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/github@4.26.0
+pnpm install @chat-adapter/github@4.27.0
 ```

 ### 5. Build
@@ -136,7 +136,15 @@ Use `per-thread` session mode so each PR/issue gets its own agent session.

 If you're in the middle of `/setup`, return to the setup flow now.

-Otherwise, restart the service (`systemctl --user restart nanoclaw` or `launchctl kickstart -k gui/$(id -u)/com.nanoclaw`) to pick up the new channel.
+Otherwise, restart the service to pick up the new channel.
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
+```

 ## Channel Info

@@ -0,0 +1,262 @@
+---
+name: add-gmail-tool
+description: Add Gmail as an MCP tool (read, search, send, label, draft) using OneCLI-managed OAuth. The agent gets Gmail tools in every enabled group; OneCLI injects real tokens at request time so no raw credentials are ever in the container or on disk in usable form.
+---
+
+# Add Gmail Tool (OneCLI-native)
+
+This skill wires the [`@gongrzhe/server-gmail-autoauth-mcp`](https://www.npmjs.com/package/@gongrzhe/server-gmail-autoauth-mcp) stdio MCP server into selected agent groups. The MCP server reads stub credentials containing the `onecli-managed` placeholder; the OneCLI gateway intercepts outbound calls to `gmail.googleapis.com` and injects the real OAuth bearer from its vault.
+
+Tools exposed (from `gmail-mcp@1.1.11`, surfaced to the agent as `mcp__gmail__<name>`): `search_emails`, `read_email`, `send_email`, `draft_email`, `delete_email`, `modify_email`, `batch_modify_emails`, `batch_delete_emails`, `download_attachment`, `list_email_labels`, `create_label`, `update_label`, `delete_label`, `get_or_create_label`, `list_filters`, `get_filter`, `create_filter`, `create_filter_from_template`, `delete_filter`.
+
+**Why this pattern:** v2's invariant is that containers never receive raw API keys — OneCLI is the sole credential path (see CHANGELOG v2.0.0). The stub-file pattern satisfies this: the container sees `"onecli-managed"` placeholders, the gateway swaps them in flight.
+
+## Phase 1: Pre-flight
+
+### Verify OneCLI has Gmail connected
+
+```bash
+onecli apps get --provider gmail
+```
+
+Expected: `"connection": { "status": "connected" }` with scopes including `gmail.readonly`, `gmail.modify`, `gmail.send`.
+
+If not connected, tell the user:
+
+> Open the OneCLI web UI at http://127.0.0.1:10254, go to Apps → Gmail, and click Connect. Sign in with the Google account you want the agent to act as.
+
+### Verify stub credentials exist
+
+```bash
+ls -la ~/.gmail-mcp/gcp-oauth.keys.json ~/.gmail-mcp/credentials.json 2>&1
+```
+
+If both exist and contain `"onecli-managed"`:
+
+```bash
+grep -l onecli-managed ~/.gmail-mcp/gcp-oauth.keys.json ~/.gmail-mcp/credentials.json
+```
+
+...skip to Phase 2.
+
+If either file exists but does **not** contain `onecli-managed`, **STOP** and tell the user — these are real OAuth credentials from a previous non-OneCLI install. Back them up, then delete before proceeding. The OneCLI migration normally handles this; if it didn't, something is wrong.
+
+If both files are absent, write them now:
+
+```bash
+mkdir -p ~/.gmail-mcp
+cat > ~/.gmail-mcp/gcp-oauth.keys.json <<'EOF'
+{
+  "installed": {
+    "client_id": "onecli-managed.apps.googleusercontent.com",
+    "client_secret": "onecli-managed",
+    "redirect_uris": ["http://localhost:3000/oauth2callback"]
+  }
+}
+EOF
+cat > ~/.gmail-mcp/credentials.json <<'EOF'
+{
+  "access_token": "onecli-managed",
+  "refresh_token": "onecli-managed",
+  "token_type": "Bearer",
+  "expiry_date": 99999999999999,
+  "scope": "https://www.googleapis.com/auth/gmail.readonly https://www.googleapis.com/auth/gmail.modify https://www.googleapis.com/auth/gmail.send"
+}
+EOF
+chmod 600 ~/.gmail-mcp/gcp-oauth.keys.json ~/.gmail-mcp/credentials.json
+```
+
+### Verify mount allowlist covers the path
+
+```bash
+cat ~/.config/nanoclaw/mount-allowlist.json
+```
+
+`~/.gmail-mcp` must sit under an `allowedRoots` entry (e.g. `/home/<user>`). If it doesn't, tell the user to run `/manage-mounts` first or add their home directory.
+
+### Check agent secret-mode
+
+For each target agent group, confirm OneCLI will inject Gmail secrets into its container. Find the OneCLI agent ID that matches the group's `agentGroupId`:
+
+```bash
+onecli agents list
+```
+
+If that agent's `secretMode` is `all`, you're done — Gmail secrets (identified by OneCLI's Gmail hostPattern) will auto-inject. If it's `selective`, explicitly assign the Gmail secrets using the safe merge pattern (`set-secrets` replaces the entire list — always read first):
+
+```bash
+GMAIL_IDS=$(onecli secrets list | jq -r '[.data[] | select(.name | test("(?i)gmail")) | .id] | join(",")')
+CURRENT=$(onecli agents secrets --id <agent-id> | jq -r '[.data[]] | join(",")')
+MERGED=$(printf '%s' "$CURRENT,$GMAIL_IDS" | tr ',' '\n' | sort -u | paste -sd ',' -)
+onecli agents set-secrets --id <agent-id> --secret-ids "$MERGED"
+onecli agents secrets --id <agent-id>
+```
+
+## Phase 2: Apply Code Changes
+
+### Check if already applied
+
+```bash
+grep -q 'GMAIL_MCP_VERSION' container/Dockerfile && \
+echo "ALREADY APPLIED — skip to Phase 3"
+```
+
+### Add MCP server to Dockerfile
+
+Edit `container/Dockerfile`. Find the pinned-version ARG block:
+
+```dockerfile
+ARG CLAUDE_CODE_VERSION=2.1.116
+ARG AGENT_BROWSER_VERSION=latest
+ARG VERCEL_VERSION=latest
+ARG BUN_VERSION=1.3.12
+```
+
+Add a new line:
+
+```dockerfile
+ARG GMAIL_MCP_VERSION=1.1.11
+```
+
+Then find the last pnpm global-install `RUN` block (the one that installs `@anthropic-ai/claude-code`) and add a new block after it, before `# ---- Entrypoint`:
+
+```dockerfile
+RUN --mount=type=cache,target=/root/.cache/pnpm \
+    pnpm install -g \
+        "@gongrzhe/server-gmail-autoauth-mcp@${GMAIL_MCP_VERSION}" \
+        "zod-to-json-schema@3.22.5"
+```
+
+Pinned version matters — `minimumReleaseAge` in `pnpm-workspace.yaml` gates trunk installs, and CLAUDE.md requires a fixed ARG version for all Node CLIs installed into the image.
+
+**Why the `zod-to-json-schema` pin:** `@gongrzhe/server-gmail-autoauth-mcp@1.1.11` has loose deps (`zod-to-json-schema: ^3.22.1`, `zod: ^3.22.4`). pnpm resolves `zod-to-json-schema` to the latest 3.25.x, which imports `zod/v3` — a subpath that only exists in `zod>=3.25`. But `zod` resolves to `3.24.x` (highest satisfying `^3.22.4` without breaking peer ranges). Result: `ERR_PACKAGE_PATH_NOT_EXPORTED` at import time. Pinning `zod-to-json-schema` to a pre-v3-subpath version avoids it. Re-check if you bump `GMAIL_MCP_VERSION`.
+
+**No `TOOL_ALLOWLIST` edit needed.** `container/agent-runner/src/providers/claude.ts` derives the allow-pattern dynamically from each group's `mcpServers` map (`Object.keys(this.mcpServers).map(mcpAllowPattern)`), so registering `gmail` in Phase 3 automatically allows `mcp__gmail__*`. Earlier versions of this skill instructed a static `TOOL_ALLOWLIST` edit — that's now redundant.
+
+### Rebuild the container image
+
+```bash
+./container/build.sh
+```
+
+Must complete cleanly. The new `pnpm install -g` layer is ~60s first time (cached on rebuild).
+
+## Phase 3: Wire Per-Agent-Group
+
+For each agent group that should have Gmail (ask the user — typically their personal DM and CLI agents, sometimes shared household agents), persist two changes to the **central DB** (`data/v2.db`): the `mcpServers.gmail` entry and an `additionalMounts` entry for `.gmail-mcp`. Both flow through `materializeContainerJson` on every spawn, so editing `groups/<folder>/container.json` by hand does **not** stick — that file is regenerated from the DB.
+
+### List groups, pick which ones get Gmail
+
+```bash
+ncl groups list
+```
+
+### Register the MCP server
+
+For each chosen `<group-id>`:
+
+```bash
+ncl groups config add-mcp-server \
+  --id <group-id> \
+  --name gmail \
+  --command gmail-mcp \
+  --args '[]' \
+  --env '{"GMAIL_OAUTH_PATH":"/workspace/extra/.gmail-mcp/gcp-oauth.keys.json","GMAIL_CREDENTIALS_PATH":"/workspace/extra/.gmail-mcp/credentials.json"}'
+```
+
+Approval behaviour depends on where you run it: from inside an agent's container `ncl` write verbs are approval-gated (admin approves before it lands); from a host operator shell with full scope, it executes immediately. Either way, the response tells you which path it took.
+
+### Add the `.gmail-mcp` mount
+
+There is no `ncl groups config add-mount` verb yet (tracked in [#2395](https://github.com/nanocoai/nanoclaw/issues/2395)). Until that ships, edit the DB directly via the in-tree wrapper (`scripts/q.ts` — `setup/verify.ts:5` codifies that NanoClaw avoids depending on the `sqlite3` CLI binary, so don't shell out to it):
+
+```bash
+GROUP_ID='<group-id>'
+HOST_PATH="$HOME/.gmail-mcp"
+MOUNT=$(jq -cn --arg h "$HOST_PATH" '{hostPath:$h, containerPath:".gmail-mcp", readonly:false}')
+pnpm exec tsx scripts/q.ts data/v2.db "UPDATE container_configs \
+  SET additional_mounts = json_insert(additional_mounts, '\$[#]', json('$MOUNT')), \
+      updated_at = datetime('now') \
+  WHERE agent_group_id = '$GROUP_ID';"
+```
+
+Run from your NanoClaw project root (where `data/v2.db` lives). The `$[#]` placeholder is SQLite JSON1's append-to-end notation; it's `\$`-escaped so bash doesn't arithmetic-expand it before sqlite sees it. `updated_at` is ISO-string everywhere else in the schema, so use `datetime('now')` — not `strftime('%s','now')`, which would silently mix epoch ints into a column of YYYY-MM-DD HH:MM:SS strings.
+
+**Switch to `ncl groups config add-mount` once #2395 lands.** Update this skill at that time.
+
+**Why the container path is relative:** `mount-security` rejects absolute `containerPath` values. Additional mounts are prefixed with `/workspace/extra/`, so `containerPath: ".gmail-mcp"` lands at `/workspace/extra/.gmail-mcp`. The MCP server's `GMAIL_OAUTH_PATH` / `GMAIL_CREDENTIALS_PATH` env vars point at that absolute location inside the container.
+
+**Why this can't be `groups/<folder>/container.json`:** post-migration `014-container-configs`, `materializeContainerJson` in `src/container-config.ts` rewrites that file from the DB on every spawn. Anything hand-edited there is silently overwritten on next restart.
+
+## Phase 4: Build and Restart
+
+```bash
+pnpm run build
+```
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
+```
+
+## Phase 5: Verify
+
+### Test from the wired agent
+
+Tell the user:
+
+> In your `<agent-name>` chat, send: **"list my gmail labels"** or **"search my inbox for invoices from last month"**.
+>
+> The agent should use `mcp__gmail__list_labels` / `mcp__gmail__search`. The first call may take a second or two while the MCP server starts and OneCLI does the token exchange.
+
+### Check logs if the tool isn't working
+
+```bash
+tail -100 logs/nanoclaw.log logs/nanoclaw.error.log | grep -iE 'gmail|mcp'
+# Per-container logs — session-scoped:
+ls data/v2-sessions/*/stderr.log | head
+```
+
+Common signals:
+- `command not found: gmail-mcp` → image wasn't rebuilt or PATH doesn't include `/pnpm` (should — `ENV PATH="$PNPM_HOME:$PATH"` in Dockerfile).
+- `ENOENT: no such file or directory, open '/workspace/extra/.gmail-mcp/credentials.json'` → mount is missing. Check `~/.config/nanoclaw/mount-allowlist.json` includes a parent of `~/.gmail-mcp`.
+- `401 Unauthorized` from `gmail.googleapis.com` → OneCLI isn't injecting. Check the agent's secret mode (`onecli agents secrets --id <agent-id>`) and that the Gmail app is connected (`onecli apps get --provider gmail`).
+- Agent says "I don't have Gmail tools" → the `gmail` MCP server isn't registered in this group's `mcpServers` (re-run the `ncl groups config add-mcp-server` step in Phase 3 for that group and restart it), or the agent-runner image is stale (rebuild with `./container/build.sh`, with `--no-cache` if suspicious).
+
+## Removal
+
+1. For each group that had Gmail wired, remove the MCP server from the DB:
+   ```bash
+   ncl groups config remove-mcp-server --id <group-id> --name gmail
+   ```
+2. Remove the `.gmail-mcp` mount from the DB (no `remove-mount` verb yet — same #2395 dependency):
+   ```bash
+   pnpm exec tsx scripts/q.ts data/v2.db "UPDATE container_configs \
+     SET additional_mounts = (SELECT json_group_array(value) FROM json_each(additional_mounts) \
+                              WHERE json_extract(value, '\$.containerPath') != '.gmail-mcp'), \
+         updated_at = datetime('now') \
+     WHERE agent_group_id = '<group-id>';"
+   ```
+3. Remove the `GMAIL_MCP_VERSION` ARG and the `pnpm install -g @gongrzhe/server-gmail-autoauth-mcp` block from `container/Dockerfile`.
+4. `pnpm run build && ./container/build.sh && systemctl --user restart "$(. setup/lib/install-slug.sh && systemd_unit)"`.
+5. (Optional) `rm -rf ~/.gmail-mcp/` if no other host-side tool needs the stubs.
+6. (Optional) Disconnect Gmail in OneCLI: `onecli apps disconnect --provider gmail`.
+
+No `TOOL_ALLOWLIST` removal step — Phase 2 no longer edits it.
+
+## Notes
+
+- **Stub format is OneCLI-prescribed.** The `access_token: "onecli-managed"` pattern with `expiry_date: 99999999999999` tells the Google auth client the token is valid; OneCLI intercepts the outgoing Gmail API call and rewrites `Authorization: Bearer onecli-managed` to the real token. `expiry_date: 0` (refresh-interception) is an alternative the OneCLI docs describe — both work but OneCLI's own `migrate` command writes the far-future variant, which is what this skill assumes.
+- **Scopes are set at OAuth connect time.** If the agent needs scopes beyond what's currently connected (e.g. the user later wants `calendar.readonly` for combined email/calendar workflows), disconnect and reconnect Gmail in the OneCLI web UI with the expanded scope set.
+- **This is tool-only.** Inbound email as a channel (emails trigger the agent) is a separate piece of work — it needs a `src/channels/gmail.ts` adapter that polls the inbox and routes to a messaging group. The pre-v2 qwibitai skill had this; it has not been ported to v2's channel architecture as of v2.0.0.
+
+## Credits & references
+
+- **MCP server:** [`@gongrzhe/server-gmail-autoauth-mcp`](https://github.com/GongRzhe/Gmail-MCP-Server) by GongRzhe — MIT-licensed.
+- **OneCLI credential stubs:** pattern documented at `https://onecli.sh/docs/guides/credential-stubs/gmail.md`.
+- **Skill pattern:** modeled on [`add-atomic-chat-tool`](../add-atomic-chat-tool/SKILL.md) and [`add-vercel`](../add-vercel/SKILL.md).
+- **Addresses:** [issue #1500](https://github.com/nanocoai/nanoclaw/issues/1500) (proxy Gmail/Calendar OAuth tokens through credential proxy) for the Gmail side.
+- **Related PRs:** [#1810](https://github.com/nanocoai/nanoclaw/pull/1810) (pre-install Gmail/Notion MCP) overlaps on the "install the MCP server in the image" idea but bundles many unrelated changes; this skill is the focused OneCLI-native version.
@@ -75,7 +75,7 @@ Stop and wait for the user to confirm before continuing.

 ### Remote Mode (Photon API)

-1. Set up a [Photon](https://photon.im) account
+1. Set up a [Photon](https://photon.codes) account
 2. Get your server URL and API key

 ### Configure environment
@@ -71,40 +71,16 @@ AskUserQuestion: "Want periodic wiki health checks?"
 2. **Monthly**
 3. **Skip** — lint manually

-If yes, create a NanoClaw scheduled task that runs in the wiki group. This is NOT a Claude Code cron job — it's a NanoClaw group task that runs in the agent container. Insert it into the SQLite database:
+If yes, ask the agent to schedule the lint task using the `schedule_task` MCP tool in conversation.
+
+## Step 6: Restart
+
+Run from your NanoClaw project root:

 ```bash
-pnpm exec tsx -e "
-const Database = require('better-sqlite3');
-const { CronExpressionParser } = require('cron-parser');
-const db = new Database('store/messages.db');
-const interval = CronExpressionParser.parse('<cron-expr>', { tz: process.env.TZ || 'UTC' });
-const nextRun = interval.next().toISOString();
-db.prepare('INSERT INTO scheduled_tasks (id, group_folder, chat_jid, prompt, schedule_type, schedule_value, context_mode, next_run, status, created_at) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)').run(
-  'wiki-lint',
-  '<group_folder>',
-  '<chat_jid>',
-  'Run a wiki lint pass per the wiki container skill. Check for contradictions, orphan pages, stale content, missing cross-references, and gaps. Report findings and offer to fix issues.',
-  'cron',
-  '<cron-expr>',
-  'group',
-  nextRun,
-  'active',
-  new Date().toISOString()
-);
-db.close();
-"
-```
-
-Use the group's `folder` and `chat_jid` from the registered groups table. Cron expressions: `0 10 * * 0` (weekly Sunday 10am) or `0 10 1 * *` (monthly 1st at 10am).
-
-## Step 6: Build and restart
-
-```bash
-pnpm run build
-./container/build.sh
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
-# Linux: systemctl --user restart nanoclaw
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
 ```

 Tell the user to test by sending a source to the wiki group.
@@ -87,7 +87,7 @@ Linear OAuth apps can't be @-mentioned, so the bridge's `onNewMention` handler n
 ### 5. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/linear@4.26.0
+pnpm install @chat-adapter/linear@4.27.0
 ```

 ### 6. Build
@@ -156,7 +156,15 @@ The `platform_id` must be `linear:<TEAM_KEY>` matching the `LINEAR_TEAM_KEY` env

 If you're in the middle of `/setup`, return to the setup flow now.

-Otherwise, restart the service (`systemctl --user restart nanoclaw` or `launchctl kickstart -k gui/$(id -u)/com.nanoclaw`) to pick up the new channel.
+Otherwise, restart the service to pick up the new channel.
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
+```

 ## Channel Info

@@ -0,0 +1,211 @@
+---
+name: add-mnemon
+description: Add persistent graph-based memory via mnemon. Agents recall past context before responding and remember insights after each turn.
+---
+
+# Add Mnemon — Persistent Memory
+
+Installs [mnemon](https://github.com/mnemon-dev/mnemon) in the agent container image. On each container start, `mnemon setup` registers Claude Code hooks that surface relevant memory before the agent responds and store new insights after each turn. Memory is written to the per-agent-group `.claude/` mount and survives container restarts.
+
+## Provider Compatibility
+
+**mnemon hooks only work with `--target claude-code`.** If the agent group uses `AGENT_PROVIDER=opencode`, hooks registered by `mnemon setup` will never fire — OpenCode spawns its own process and doesn't invoke the `claude` CLI at all.
+
+Check your provider:
+
+```bash
+grep AGENT_PROVIDER .env groups/*/container.json 2>/dev/null
+```
+
+- `AGENT_PROVIDER=claude` (default) — fully compatible, proceed with both Phase 2 steps.
+- `AGENT_PROVIDER=opencode` — use **Phase 2 (OpenCode path)** instead of the standard entrypoint step.
+
+## Phase 1: Pre-flight
+
+### Check if already applied
+
+```bash
+grep -q 'MNEMON_VERSION' container/Dockerfile && echo "Already applied" || echo "Not applied"
+```
+
+If already applied, skip to Phase 3 (Verify).
+
+### Check latest mnemon version
+
+```bash
+curl -fsSL https://api.github.com/repos/mnemon-dev/mnemon/releases/latest | grep '"tag_name"'
+```
+
+Note the version (e.g. `v0.1.1`) — use it as `MNEMON_VERSION` in the next step.
+
+## Phase 2: Apply Changes (Claude Code path)
+
+### 1. Dockerfile — install mnemon binary
+
+Add after the AWS CLI block, before the Bun runtime section:
+
+```dockerfile
+# ---- mnemon — persistent agent memory ----------------------------------------
+ARG MNEMON_VERSION=0.1.1
+RUN ARCH=$(dpkg --print-architecture) && \
+    curl -fsSL "https://github.com/mnemon-dev/mnemon/releases/download/v${MNEMON_VERSION}/mnemon_${MNEMON_VERSION}_linux_${ARCH}.tar.gz" \
+    | tar -xz -C /usr/local/bin mnemon && \
+    chmod +x /usr/local/bin/mnemon
+
+ENV MNEMON_DATA_DIR=/home/node/.claude/mnemon
+```
+
+`MNEMON_DATA_DIR` points into the per-agent-group `.claude/` mount so memory persists across container restarts. No extra volume mounts needed.
+
+### 2. Entrypoint — run mnemon setup on each container start
+
+`mnemon setup` is idempotent. Edit `container/entrypoint.sh` to run it right after `set -e`, before the `cat` that captures stdin:
+
+```bash
+#!/bin/bash
+# NanoClaw agent container entrypoint.
+#
+# ...existing header comment...
+
+set -e
+
+mnemon setup --target claude-code --yes --global >/dev/stderr 2>&1
+
+cat > /tmp/input.json
+
+exec bun run /app/src/index.ts < /tmp/input.json
+```
+
+`>/dev/stderr 2>&1` routes all mnemon output to stderr (docker logs) so it doesn't interfere with the JSON stdin handshake between host and agent-runner.
+
+### 3. Rebuild and smoke-test the image
+
+```bash
+./container/build.sh
+docker run --rm --entrypoint mnemon nanoclaw-agent:latest --version
+```
+
+## Phase 3: Restart and Verify
+
+### Restart the service
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+systemctl --user restart $(systemd_unit)              # Linux
+# launchctl kickstart -k gui/$(id -u)/$(launchd_label)   # macOS
+```
+
+### Confirm mnemon hooks are registered
+
+After the next container starts, check that setup ran:
+
+```bash
+docker logs $(docker ps --filter name=nanoclaw-v2 --format '{{.Names}}' | head -1) 2>&1 | grep -i mnemon
+```
+
+Then inspect the hooks inside the running container:
+
+```bash
+docker exec $(docker ps --filter name=nanoclaw-v2 --format '{{.Names}}' | head -1) \
+  cat /home/node/.claude/settings.json | grep -A5 mnemon
+```
+
+### Test memory recall
+
+Have a conversation with the agent, then start a new session and reference something from the earlier one. Mnemon should surface the relevant context automatically without you restating it.
+
+## Phase 2 (OpenCode path) — context injection
+
+mnemon hooks don't fire under OpenCode. Instead, the agent-runner injects mnemon context directly into every prompt via `wrapPromptWithContext()` in `container/agent-runner/src/providers/opencode.ts`. This is already implemented in NanoClaw — no code changes needed if you're on current `ester`/`main`.
+
+**How it works:** On each prompt, `readMnemonContext()` checks for `MNEMON_DATA_DIR` (set by the Dockerfile `ENV`). If the env var is present, it reads `$MNEMON_DATA_DIR/prompt/guide.md` (mnemon's custom prompt guide, written by `mnemon setup`) or falls back to an inline guide. The content is prepended as a `<system>` block, instructing the agent to run `mnemon recall` at the start of relevant tasks and `mnemon remember` after key decisions.
+
+**What this means for the agent:** The agent (running inside OpenCode) can call `mnemon recall`, `mnemon remember`, `mnemon link`, and `mnemon status` via its bash tool. mnemon writes its graph to `$MNEMON_DATA_DIR`, which is in the per-agent-group `.claude/` mount — so memory persists across container restarts.
+
+**Applying:** Only the Dockerfile step from Phase 2 is needed for OpenCode agents. Skip `container/entrypoint.sh` entirely.
+
+```dockerfile
+ARG MNEMON_VERSION=0.1.1
+RUN ARCH=$(dpkg --print-architecture) && \
+    curl -fsSL "https://github.com/mnemon-dev/mnemon/releases/download/v${MNEMON_VERSION}/mnemon_${MNEMON_VERSION}_linux_${ARCH}.tar.gz" \
+    | tar -xz -C /usr/local/bin mnemon && \
+    chmod +x /usr/local/bin/mnemon
+ENV MNEMON_DATA_DIR=/home/node/.claude/mnemon
+```
+
+Then rebuild: `./container/build.sh`
+
+### Verify (OpenCode)
+
+Start a session and ask the agent to run `mnemon status`. It should report empty graphs (no error) on first run.
+
+```bash
+# Also confirm the binary is present in the image:
+docker run --rm --entrypoint mnemon nanoclaw-agent:latest --version
+```
+
+## Memory Storage
+
+Mnemon writes to `/home/node/.claude/mnemon/` inside the container, which maps to the per-agent-group `.claude/` directory on the host. To find the exact host path:
+
+```bash
+docker inspect $(docker ps --filter name=nanoclaw-v2 --format '{{.Names}}' | head -1) \
+  --format '{{range .Mounts}}{{if eq .Destination "/home/node/.claude"}}{{.Source}}{{end}}{{end}}'
+```
+
+To reset all memory for an agent, stop the container and delete the `mnemon/` subdirectory from that host path.
+
+## Migration Guide Update
+
+If you are using `/migrate-nanoclaw`, add these entries to `.nanoclaw-migrations/05-dockerfile.md`:
+
+**Dockerfile — after AWS CLI, before Bun runtime:**
+```dockerfile
+ARG MNEMON_VERSION=0.1.1
+RUN ARCH=$(dpkg --print-architecture) && \
+    curl -fsSL "https://github.com/mnemon-dev/mnemon/releases/download/v${MNEMON_VERSION}/mnemon_${MNEMON_VERSION}_linux_${ARCH}.tar.gz" \
+    | tar -xz -C /usr/local/bin mnemon && \
+    chmod +x /usr/local/bin/mnemon
+ENV MNEMON_DATA_DIR=/home/node/.claude/mnemon
+```
+
+**`container/entrypoint.sh` — add after `set -e`:**
+```bash
+mnemon setup --target claude-code --yes --global >/dev/stderr 2>&1
+```
+
+## Troubleshooting
+
+### `mnemon: command not found` in container
+
+The image wasn't rebuilt after adding the Dockerfile layer. Run `./container/build.sh` and restart.
+
+### Memory not persisting across restarts
+
+Verify `MNEMON_DATA_DIR` resolves to a mounted path (not an in-container ephemeral directory):
+
+```bash
+docker exec <container> sh -c 'ls -la $MNEMON_DATA_DIR'
+```
+
+If the directory is empty after conversations, the mount is missing or the path is wrong. Check the host mount with the `docker inspect` command above.
+
+### Agent not using past memory
+
+`mnemon setup` writes hooks into `/home/node/.claude/settings.json`. Verify:
+
+```bash
+docker exec <container> cat /home/node/.claude/settings.json
+```
+
+If the hooks are absent, `mnemon setup` may have failed silently. Check container startup logs for errors from mnemon.
+
+### Setup fails at container start
+
+Run setup manually inside a running container to see the full error:
+
+```bash
+docker exec -it <container> mnemon setup --target claude-code --yes --global
+```
@@ -76,7 +76,7 @@ Then rebuild the container image: `./container/build.sh`

 Ask the user (plain text, not AskUserQuestion):

-1. **Which agent group?** List available groups: `sqlite3 data/v2.db "SELECT folder, name FROM agent_groups;"`
+1. **Which agent group?** List available groups: `pnpm exec tsx scripts/q.ts data/v2.db "SELECT folder, name FROM agent_groups;"`
 2. **Which Ollama model?** List available: `curl -s http://localhost:11434/api/tags | grep '"name"'`
 3. **Block Anthropic API?** Recommended yes — prevents accidental spend if config drifts.

@@ -111,7 +111,7 @@ Read the agent group's shared Claude settings:

 ```bash
 # Find the agent group ID
-AG_ID=$(sqlite3 data/v2.db "SELECT id FROM agent_groups WHERE folder='<FOLDER>';")
+AG_ID=$(pnpm exec tsx scripts/q.ts data/v2.db "SELECT id FROM agent_groups WHERE folder='<FOLDER>';")
 SETTINGS=data/v2-sessions/$AG_ID/.claude-shared/settings.json
 ```

@@ -130,12 +130,15 @@ file, not from env vars. This file is bind-mounted into the container as `~/.cla

 ## 5. Build and restart

+Run from your NanoClaw project root:
+
 ```bash
 export PATH="/opt/homebrew/bin:$PATH"
 pnpm run build
-launchctl unload ~/Library/LaunchAgents/com.nanoclaw.plist
-launchctl load ~/Library/LaunchAgents/com.nanoclaw.plist
-# Linux: systemctl --user restart nanoclaw
+source setup/lib/install-slug.sh
+launchctl unload ~/Library/LaunchAgents/$(launchd_label).plist
+launchctl load   ~/Library/LaunchAgents/$(launchd_label).plist
+# Linux: systemctl --user restart $(systemd_unit)
 ```

 ## 6. Verify
@@ -54,7 +54,7 @@ git remote -v
 If `upstream` is missing, add it:

 ```bash
-git remote add upstream https://github.com/qwibitai/nanoclaw.git
+git remote add upstream https://github.com/nanocoai/nanoclaw.git
 ```

 ### Merge the skill branch
@@ -122,9 +122,12 @@ OLLAMA_HOST=http://your-ollama-host:11434

 ### Restart the service

+Run from your NanoClaw project root:
+
 ```bash
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
-# Linux: systemctl --user restart nanoclaw
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
 ```

 ## Phase 4: Verify
@@ -132,12 +132,15 @@ Credentials: register provider API keys in OneCLI with the matching `--host-patt

 After adding a secret, **grant the agent access** — agents in `selective` mode only receive secrets they've been explicitly assigned:

-```bash
-# Find the agent id and secret id, then:
-onecli agents set-secrets --id <agent-id> --secret-ids <existing-ids>,<new-secret-id>
-```
+Use the safe merge pattern — `set-secrets` replaces the entire list, so always read first:

-Always include existing secret IDs in the list — `set-secrets` replaces, not appends.
+```bash
+AGENT_ID=$(onecli agents list | jq -r '.data[] | select(.identifier=="<agentGroupId>") | .id')
+CURRENT=$(onecli agents secrets --id "$AGENT_ID" | jq -r '[.data[]] | join(",")')
+MERGED=$(printf '%s' "$CURRENT,<new-secret-id>" | tr ',' '\n' | sort -u | paste -sd ',' -)
+onecli agents set-secrets --id "$AGENT_ID" --secret-ids "$MERGED"
+onecli agents secrets --id "$AGENT_ID"
+```

 #### Example: DeepSeek

@@ -208,7 +211,7 @@ onecli secrets create --name "OpenCode Zen" --type generic \

 ### Per group / per session

-Schema: **`agent_groups.agent_provider`** and **`sessions.agent_provider`**. Set to `opencode` for groups or sessions that should use OpenCode. The container receives `AGENT_PROVIDER` from the resolved value (session overrides group).
+Set `"provider": "opencode"` in the group's **`container.json`** (`groups/<folder>/container.json`) — the in-container runner reads `provider` from there, not from the DB. The DB columns **`agent_groups.agent_provider`** and **`sessions.agent_provider`** (session overrides group) only drive host-side provider contribution — per-session XDG mount, `OPENCODE_*` env passthrough — and do not propagate into `container.json` at spawn time. Set both, or just edit `container.json`; if they disagree, the runner uses `container.json` and the host-side resolver falls back through session → group → `container.json` → `'claude'`.

 Extra MCP servers still come from **`NANOCLAW_MCP_SERVERS`** / `container_config.mcpServers` on the host; the runner merges them into the same `mcpServers` object passed to **both** Claude and OpenCode providers.

@@ -229,19 +229,22 @@ echo '{}' | docker run -i --entrypoint /bin/echo nanoclaw-agent:latest "Containe

 ### 7. Restart Service

-Rebuild the main app and restart:
+Rebuild the main app and restart.
+
+Run from your NanoClaw project root:

 ```bash
 pnpm run build
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
-# Linux: systemctl --user restart nanoclaw
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+# Linux: systemctl --user restart $(systemd_unit)
 ```

 Wait 3 seconds for service to start, then verify:
 ```bash
 sleep 3
-launchctl list | grep nanoclaw  # macOS
-# Linux: systemctl --user status nanoclaw
+launchctl list | grep "$(. setup/lib/install-slug.sh && launchd_label)"  # macOS
+# Linux: systemctl --user status "$(. setup/lib/install-slug.sh && systemd_unit)"
 ```

 ### 8. Test Integration
@@ -275,7 +278,7 @@ Look for: `Parallel AI MCP servers configured`
 - Check agent-runner logs for "Parallel AI MCP servers configured" message

 **Task polling not working:**
- Verify scheduled task was created: `sqlite3 store/messages.db "SELECT * FROM scheduled_tasks"`
+- Verify scheduled task was created: `pnpm exec tsx scripts/q.ts store/messages.db "SELECT * FROM scheduled_tasks"`
 - Check task runs: `tail -f logs/nanoclaw.log | grep "scheduled task"`
 - Ensure task prompt includes proper Parallel MCP tool names

@@ -287,4 +290,4 @@ To remove Parallel AI integration:
 2. Revert changes to container-runner.ts and agent-runner/src/index.ts
 3. Remove Web Research Tools section from groups/main/CLAUDE.md
 4. Rebuild: `./container/build.sh && pnpm run build`
-5. Restart: `launchctl kickstart -k gui/$(id -u)/com.nanoclaw` (macOS) or `systemctl --user restart nanoclaw` (Linux)
+5. Restart: `source setup/lib/install-slug.sh && launchctl kickstart -k gui/$(id -u)/$(launchd_label)` (macOS) or `source setup/lib/install-slug.sh && systemctl --user restart $(systemd_unit)` (Linux)
@@ -0,0 +1,140 @@
+---
+name: add-rtk
+description: Install rtk token-compression proxy into agent containers. Routes Bash tool calls through rtk for 60–90% token savings on dev commands (git, cargo, pytest, docker, kubectl, etc.).
+---
+
+# Add rtk
+
+Install [rtk](https://github.com/rtk-ai/rtk) — a CLI proxy delivering 60–90% token savings on common dev commands (git, cargo, pytest, docker, kubectl, etc.) — and wire it transparently into agent containers via the Claude Code `PreToolUse` hook.
+
+## What this sets up
+
+- `rtk` binary at `~/.local/bin/rtk` on the host
+- `~/.local/bin/rtk` mounted read-only at `/usr/local/bin/rtk` inside the target agent group's containers
+- `PreToolUse` hook in the agent group's `settings.json` so every Bash call is automatically filtered through rtk — no CLAUDE.md instructions needed
+
+## Step 1 — Install rtk on the host
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh
+```
+
+If the script put the binary elsewhere, move it:
+
+```bash
+find ~/.local ~/.cargo/bin ~/bin -name rtk 2>/dev/null
+mv "$(which rtk 2>/dev/null)" ~/.local/bin/rtk
+```
+
+Verify:
+
+```bash
+~/.local/bin/rtk --version
+chmod +x ~/.local/bin/rtk   # if needed
+```
+
+## Step 2 — Identify the target agent group
+
+```bash
+ncl groups list
+```
+
+Note the group ID (e.g. `ag-1776342942165-ptgddd`). Repeat Steps 3–5 for each group.
+
+## Step 3 — Mount rtk into the container config
+
+`additional_mounts` is a JSON column not exposed via `ncl config update`. Update it directly via the DB helper, merging with any existing mounts.
+
+Read current mounts first:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "SELECT additional_mounts FROM container_configs WHERE agent_group_id = '<group-id>'"
+```
+
+Then write the merged array (include all existing entries plus the rtk entry):
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "UPDATE container_configs SET additional_mounts = '<merged-json>' WHERE agent_group_id = '<group-id>'"
+```
+
+The rtk entry to append: `{"hostPath":"/home/<user>/.local/bin/rtk","containerPath":"/usr/local/bin/rtk","readonly":true}`
+
+Verify:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "SELECT additional_mounts FROM container_configs WHERE agent_group_id = '<group-id>'"
+```
+
+## Step 4 — Add the PreToolUse hook to settings.json
+
+Each agent group has a `settings.json` at:
+
+```
+data/v2-sessions/<group-id>/.claude-shared/settings.json
+```
+
+This file is mounted at `/home/node/.claude/settings.json` inside the container and is read by Claude Code for hooks, env, and model config.
+
+Add the `PreToolUse` entry using `jq` to merge safely:
+
+```bash
+SETTINGS="data/v2-sessions/<group-id>/.claude-shared/settings.json"
+
+jq '.hooks.PreToolUse = [{"matcher":"Bash","hooks":[{"type":"command","command":"rtk hook claude"}]}]' \
+  "$SETTINGS" > /tmp/rtk-settings.json && mv /tmp/rtk-settings.json "$SETTINGS"
+```
+
+If `PreToolUse` already exists, append instead of overwriting:
+
+```bash
+jq '.hooks.PreToolUse += [{"matcher":"Bash","hooks":[{"type":"command","command":"rtk hook claude"}]}]' \
+  "$SETTINGS" > /tmp/rtk-settings.json && mv /tmp/rtk-settings.json "$SETTINGS"
+```
+
+## Step 5 — Restart the container
+
+```bash
+ncl groups restart --id <group-id>
+```
+
+No `--message` needed — the hook is transparent and requires no agent awareness.
+
+## Verify
+
+Ask the agent to run `git status` or any other supported command. rtk intercepts it silently. Check savings with:
+
+```bash
+~/.local/bin/rtk gain
+```
+
+## Troubleshooting
+
+### `rtk: command not found` inside the container
+
+Mount wasn't applied or container wasn't restarted:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "SELECT additional_mounts FROM container_configs WHERE agent_group_id = '<group-id>'"
+# Look for entry with /usr/local/bin/rtk
+ncl groups restart --id <group-id>
+```
+
+### Hook not firing
+
+Verify the hook is in `settings.json`:
+
+```bash
+jq '.hooks.PreToolUse' data/v2-sessions/<group-id>/.claude-shared/settings.json
+```
+
+If missing, re-run Step 4.
+
+### Binary won't execute — permission denied
+
+```bash
+chmod +x ~/.local/bin/rtk
+```
@@ -0,0 +1,13 @@
+# Remove Signal
+
+1. Comment out `import './signal.js'` in `src/channels/index.ts`
+2. Remove `SIGNAL_ACCOUNT` (and any other `SIGNAL_*` vars) from `.env`
+3. Rebuild and restart
+
+If you also want to unlink the Signal account from `signal-cli`:
+
+```bash
+signal-cli -a +1YOURNUMBER removeDevice --deviceId <id>
+```
+
+(Find the device id with `signal-cli -a +1YOURNUMBER listDevices`.)
@@ -0,0 +1,331 @@
+---
+name: add-signal
+description: Add Signal channel integration via signal-cli TCP daemon. Native adapter — no Chat SDK bridge.
+---
+
+# Add Signal Channel
+
+Adds Signal messaging support via a native adapter that speaks JSON-RPC to a [signal-cli](https://github.com/AsamK/signal-cli) TCP daemon. No Chat SDK bridge — only Node.js builtins (`node:net`, `node:child_process`, `node:fs`).
+
+Unlike Telegram or Discord, Signal has no bot API. NanoClaw registers as a full Signal account on a dedicated phone number (recommended) or links as a secondary device on your existing number.
+
+## Prerequisites
+
+### Java
+
+signal-cli requires Java 17+:
+
+```bash
+java -version
+```
+
+If missing:
+- **macOS:** `brew install --cask temurin@17`
+- **Debian/Ubuntu:** `sudo apt-get install -y default-jre`
+- **RHEL/Fedora:** `sudo dnf install -y java-17-openjdk`
+
+Java 17–25 all work.
+
+### signal-cli
+
+- **macOS:** `brew install signal-cli`
+- **Linux:** download the native binary from [GitHub releases](https://github.com/AsamK/signal-cli/releases):
+
+```bash
+SIGNAL_CLI_VERSION=$(curl -fsSL https://api.github.com/repos/AsamK/signal-cli/releases/latest | python3 -c "import sys,json; print(json.load(sys.stdin)['tag_name'][1:])")
+curl -fsSL "https://github.com/AsamK/signal-cli/releases/download/v${SIGNAL_CLI_VERSION}/signal-cli-${SIGNAL_CLI_VERSION}-Linux-native.tar.gz" \
+  | tar -xz -C ~/.local
+ln -sf ~/.local/signal-cli ~/.local/bin/signal-cli
+signal-cli --version
+```
+
+> The Linux native tarball extracts a single binary directly to `~/.local/signal-cli` (not into a subdirectory). The symlink above puts it on PATH.
+
+## Registration
+
+Two paths. The new-number path is recommended and battle-tested.
+
+### Path A: Register a new number (recommended)
+
+Use a dedicated SIM or VoIP number. NanoClaw owns it entirely.
+
+> **VoIP numbers:** Signal requires SMS verification before voice. Some VoIP providers are blocked even for voice calls. If registration fails with an auth error, try a different provider or a physical SIM.
+
+**Step 1: Solve the CAPTCHA**
+
+Signal requires a CAPTCHA on first registration:
+
+1. Open `https://signalcaptchas.org/registration/generate.html` in a browser
+2. Solve the captcha
+3. Right-click the **"Open Signal"** button → **Copy Link**
+4. The link starts with `signalcaptcha://` — the token is everything after that prefix
+
+**Step 2: Request SMS verification**
+
+```bash
+signal-cli -a +1YOURNUMBER register --captcha "PASTE_TOKEN_HERE"
+```
+
+**Step 3: Voice call fallback (if your number can't receive SMS)**
+
+Wait ~60 seconds after the SMS request, then:
+
+```bash
+signal-cli -a +1YOURNUMBER register --voice --captcha "SAME_TOKEN"
+```
+
+Signal calls your number and reads a 6-digit code. The same captcha token is reusable — no need to solve a new one.
+
+> You must request SMS first. Requesting voice immediately fails with `Invalid verification method: Before requesting voice verification…`
+
+**Step 4: Verify**
+
+```bash
+signal-cli -a +1YOURNUMBER verify CODE
+```
+
+No output = success.
+
+**Step 5: Set profile name (optional)**
+
+> ⚠ Stop NanoClaw before running signal-cli commands — the daemon holds an exclusive lock on its data directory while running.
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+
+# macOS
+launchctl unload ~/Library/LaunchAgents/$(launchd_label).plist
+signal-cli -a +1YOURNUMBER updateProfile --name "YourBotName"
+# optionally: --avatar /path/to/avatar.jpg
+launchctl load ~/Library/LaunchAgents/$(launchd_label).plist
+
+# Linux
+systemctl --user stop $(systemd_unit)
+signal-cli -a +1YOURNUMBER updateProfile --name "YourBotName"
+systemctl --user start $(systemd_unit)
+```
+
+### Path B: Link as secondary device
+
+Joins an existing Signal account as a secondary device. Simpler, but NanoClaw shares your personal number.
+
+```bash
+signal-cli -a +1YOURNUMBER link --name "NanoClaw"
+```
+
+This prints a `tsdevice:` URI. Scan it as a QR code on your phone: **Settings → Linked Devices → Link New Device**. QR codes expire in ~30 seconds — re-run if it expires.
+
+## Install
+
+### Pre-flight (idempotent)
+
+Skip to **Credentials** if all of these are already in place:
+
+- `src/channels/signal.ts` and `src/channels/signal.test.ts` both exist
+- `src/channels/index.ts` contains `import './signal.js';`
+
+Otherwise continue. Every step below is safe to re-run.
+
+### 1. Fetch the channels branch
+
+```bash
+git fetch origin channels
+```
+
+### 2. Copy the adapter and tests
+
+```bash
+git show origin/channels:src/channels/signal.ts      > src/channels/signal.ts
+git show origin/channels:src/channels/signal.test.ts > src/channels/signal.test.ts
+```
+
+### 3. Append the self-registration import
+
+Append to `src/channels/index.ts` (skip if the line is already present):
+
+```typescript
+import './signal.js';
+```
+
+### 4. Build
+
+```bash
+pnpm run build
+```
+
+No npm packages to install — the adapter uses only Node.js builtins.
+
+## Credentials
+
+Add to `.env`:
+
+```bash
+SIGNAL_ACCOUNT=+1YOURNUMBER
+```
+
+### Optional settings
+
+```bash
+# TCP daemon host and port (default: 127.0.0.1:7583)
+SIGNAL_TCP_HOST=127.0.0.1
+SIGNAL_TCP_PORT=7583
+
+# Path to the signal-cli binary (default: resolved on PATH)
+SIGNAL_CLI_PATH=/usr/local/bin/signal-cli
+
+# Whether NanoClaw manages the daemon lifecycle (default: true).
+# Set to false if you run signal-cli daemon externally.
+SIGNAL_MANAGE_DAEMON=true
+
+# signal-cli data directory (default: ~/.local/share/signal-cli)
+SIGNAL_DATA_DIR=~/.local/share/signal-cli
+```
+
+**Security note:** keep the TCP host on `127.0.0.1`. The daemon has no auth — binding it to a public interface would expose your full Signal account to the network.
+
+Sync to container: `mkdir -p data/env && cp .env data/env/env`
+
+### Restart
+
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+
+# macOS
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)
+
+# Linux
+systemctl --user restart $(systemd_unit)
+```
+
+## Wiring
+
+### DMs
+
+After the service starts, send any message to the Signal number from your personal Signal app. The router auto-creates a `messaging_groups` row. Then:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db \
+  "SELECT id, platform_id FROM messaging_groups WHERE channel_type='signal' ORDER BY created_at DESC LIMIT 5"
+```
+
+Pass the `id` to `/init-first-agent` or `/manage-channels` to wire it to an agent group.
+
+### Groups
+
+Add the Signal number to a group from your phone, send any message, then wire the resulting row the same way. For isolated per-group sessions:
+
+```bash
+NOW=$(date -u +"%Y-%m-%dT%H:%M:%S.000Z")
+pnpm exec tsx scripts/q.ts data/v2.db "
+INSERT OR IGNORE INTO messaging_group_agents
+  (id, messaging_group_id, agent_group_id, session_mode, priority, created_at)
+VALUES
+  ('mga-'||hex(randomblob(8)), 'mg-GROUPID', 'ag-AGENTID', 'isolated', 0, '$NOW');
+"
+```
+
+### Grant user access
+
+New Signal users (including the owner's Signal identity) are silently dropped with `not_member` until granted access. After the user's first message appears in `messaging_groups`:
+
+```bash
+NOW=$(date -u +"%Y-%m-%dT%H:%M:%S.000Z")
+pnpm exec tsx scripts/q.ts data/v2.db "
+INSERT OR REPLACE INTO user_roles (user_id, role, agent_group_id, granted_by, granted_at)
+  VALUES ('signal:UUID', 'owner', NULL, 'system', '$NOW');
+INSERT OR IGNORE INTO agent_group_members (user_id, agent_group_id, added_by, added_at)
+  VALUES ('signal:UUID', 'ag-AGENTID', 'system', '$NOW');
+"
+```
+
+Find the UUID from `messaging_groups.platform_id` or the `users` table.
+
+## Next Steps
+
+If you're in the middle of `/setup`, return to the setup flow now.
+
+Otherwise, run `/init-first-agent` to create an agent and wire it to your Signal DM, or `/manage-channels` to wire this channel to an existing agent group.
+
+## Channel Info
+
+- **type**: `signal`
+- **terminology**: Signal has "chats" (1:1 DMs) and "groups"
+- **supports-threads**: no
+- **platform-id-format**:
+  - DM: `signal:{UUID}` — sender's Signal UUID (ACI), **not** their phone number
+  - Group: `signal:{base64GroupId}` — base64-encoded GroupV2 ID
+- **how-to-find-id**: Send a message to the bot, then query `messaging_groups` as shown above
+- **typical-use**: Personal assistant via Signal DMs or small group chats
+- **default-isolation**: One agent per Signal account. Multiple chats with the same operator can share an agent group; groups with other people should typically use `isolated` session mode
+
+### Features
+
+- Markdown formatting — `**bold**`, `*italic*` / `_italic_`, `` `code` ``, ` ```code fence``` `, `~~strike~~`, `||spoiler||` (converted to Signal's offset-based text styles)
+- Quoted replies — `replyTo*` fields populated from Signal quotes
+- Typing indicators — DMs only (Signal doesn't support group typing)
+- Echo suppression — outbound messages matched on `(platformId, text)` within a 10 s TTL to avoid syncMessage loops
+- Note to Self — messages you send to your own account from another device route to the agent as inbound with `isFromMe: true`
+- Voice attachments — detected but not transcribed by default; the agent receives `[Voice Message]` placeholder text. Run `/add-voice-transcription` for local transcription via parakeet-mlx
+
+Not supported yet: outbound file attachments (logged and dropped), edit/delete messages, reactions.
+
+## Troubleshooting
+
+### Daemon not reachable
+
+```bash
+grep "Signal" logs/nanoclaw.log | tail
+```
+
+If you see `Signal daemon failed to start. Is signal-cli installed and your account linked?`:
+- Confirm `signal-cli` is on PATH (or set `SIGNAL_CLI_PATH`)
+- Confirm the account is linked: `signal-cli -a +1YOURNUMBER listIdentities` should succeed without prompting
+
+If you see `Signal daemon not reachable at 127.0.0.1:7583` and `SIGNAL_MANAGE_DAEMON=false`, start the daemon yourself: `signal-cli -a +1YOURNUMBER daemon --tcp 127.0.0.1:7583`.
+
+### Bot not responding
+
+1. Channel initialized: `grep "Signal channel connected" logs/nanoclaw.log | tail -1`
+2. Channel wired: `pnpm exec tsx scripts/q.ts data/v2.db "SELECT mg.platform_id, mg.name FROM messaging_groups mg JOIN messaging_group_agents mga ON mg.id = mga.messaging_group_id WHERE mg.channel_type='signal'"`
+3. Service running: `launchctl print gui/$(id -u)/"$(. setup/lib/install-slug.sh && launchd_label)"` (macOS) / `systemctl --user status "$(. setup/lib/install-slug.sh && systemd_unit)"` (Linux)
+4. **Check for duplicate service instances** — if `logs/nanoclaw.error.log` shows `No adapter for channel type channelType="signal"` despite the adapter starting, two NanoClaw processes are racing. See the `/debug` skill section "No adapter for channel type / Messages silently lost" for the full fix.
+
+### Messages delivered but never arrive (null platformMsgId)
+
+Signal responses show `platformMsgId=undefined` in the main log. This means the delivery poll ran but found no adapter — likely a duplicate service instance issue (see above). Affected messages cannot be retried; the user must resend.
+
+### Lost connection mid-session
+
+If you see `Signal channel lost TCP connection to signal-cli daemon` in the logs, the daemon dropped the connection. Restart the service to re-establish.
+
+### Messages dropped with `not_member`
+
+The Signal user hasn't been granted membership. See "Grant user access" above. This affects every new Signal user, including the owner's Signal identity — which is a separate user record from their identity on other channels even if it's the same person.
+
+### Captcha required
+
+Signal requires a captcha for new registrations. Go to `https://signalcaptchas.org/registration/generate.html`, solve it, right-click "Open Signal", copy the link, extract the token after `signalcaptcha://`.
+
+### `Invalid verification method: Before requesting voice verification…`
+
+You must request SMS first, wait ~60 seconds, then request voice. Both steps can use the same captcha token.
+
+### Config file in use / daemon lock
+
+signal-cli holds an exclusive lock on its data directory while the daemon is running. Stop NanoClaw before running any `signal-cli` commands directly, then restart afterward.
+
+### Group replies going to DM instead of group
+
+Modern Signal groups use GroupV2. The adapter must extract the group ID from `envelope?.dataMessage?.groupV2?.id` — not `groupInfo?.groupId`, which is GroupV1/legacy. If group messages are routing as DMs, check `src/channels/signal.ts` and confirm the groupId extraction falls through to `groupV2.id`.
+
+### Java not found
+
+Install Java 17+ — see the Prerequisites section above.
+
+### QR code expired (Path B)
+
+QR codes expire in ~30 seconds. Re-run the link command to generate a new one.
@@ -0,0 +1,5 @@
+# Verify Signal
+
+Send a message to your own Signal number (Note to Self) from another device, or have someone send your linked number a DM. The bot should respond within a few seconds.
+
+If nothing happens, tail `logs/nanoclaw.log` for `Signal channel connected` and `Signal message received`.
@@ -44,7 +44,7 @@ import './slack.js';
 ### 4. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/slack@4.26.0
+pnpm install @chat-adapter/slack@4.27.0
 ```

 ### 5. Build
@@ -60,7 +60,7 @@ pnpm run build
 1. Go to [api.slack.com/apps](https://api.slack.com/apps) and click **Create New App** > **From scratch**
 2. Name it (e.g., "NanoClaw") and select your workspace
 3. Go to **OAuth & Permissions** and add Bot Token Scopes:
-   - `chat:write`, `channels:history`, `groups:history`, `im:history`, `channels:read`, `groups:read`, `users:read`, `reactions:write`
+   - `chat:write`, `im:write`, `channels:history`, `groups:history`, `im:history`, `channels:read`, `groups:read`, `users:read`, `reactions:write`, `files:read`, `files:write`
 4. Click **Install to Workspace** and copy the **Bot User OAuth Token** (`xoxb-...`)
 5. Go to **Basic Information** and copy the **Signing Secret**

@@ -76,7 +76,13 @@ pnpm run build
 10. Under **Subscribe to bot events**, add:
    - `message.channels`, `message.groups`, `message.im`, `app_mention`
 11. Click **Save Changes**
-12. Slack will show a banner asking you to **reinstall the app** — click it to apply the new event subscriptions
+
+### Interactivity
+
+12. Go to **Interactivity & Shortcuts** and toggle **Interactivity** on
+13. Set the **Request URL** to the same `https://your-domain/webhook/slack`
+14. Click **Save Changes**
+15. Slack will show a banner asking you to **reinstall the app** — click it to apply the new settings

 ### Configure environment

@@ -44,7 +44,7 @@ import './teams.js';
 ### 4. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/teams@4.26.0
+pnpm install @chat-adapter/teams@4.27.0
 ```

 ### 5. Build
@@ -55,6 +55,47 @@ pnpm run build

 ## Credentials

+Two paths — manual (Azure Portal) or auto (Teams CLI).
+
+### Auto: Teams CLI
+
+Requires Node.js 18+, a Microsoft 365 account with sideloading permissions, and a public HTTPS endpoint (ngrok, Cloudflare Tunnel, or similar).
+
+1. Install the CLI:
+
+   ```bash
+   npm install -g @microsoft/teams.cli@preview
+   ```
+
+2. Sign in and verify:
+
+   ```bash
+   teams login
+   teams status
+   ```
+
+3. Create the Entra app, client secret, and bot registration:
+
+   ```bash
+   teams app create \
+     --name "NanoClaw" \
+     --endpoint "https://your-domain/api/webhooks/teams"
+   ```
+
+   The CLI prints the credentials as `CLIENT_ID`, `CLIENT_SECRET`, and `TENANT_ID`. Map them to NanoClaw's env keys:
+
+   - `CLIENT_ID` → `TEAMS_APP_ID`
+   - `CLIENT_SECRET` → `TEAMS_APP_PASSWORD`
+   - `TENANT_ID` → `TEAMS_APP_TENANT_ID`
+
+4. Pick **Install in Teams** from the post-create menu and confirm in the Teams dialog.
+
+Continue to [Configure environment](#configure-environment).
+
+---
+
+The steps below describe the **manual Azure Portal path**.
+
 ### Step 1: Create an Azure AD App Registration

 1. Go to [Azure Portal](https://portal.azure.com) > **App registrations** > **New registration**
@@ -58,7 +58,7 @@ In `setup/index.ts`, add this entry to the `STEPS` map (right after the `registe
 ### 5. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/telegram@4.26.0
+pnpm install @chat-adapter/telegram@4.27.0
 ```

 ### 6. Build
@@ -90,12 +90,12 @@ onecli secrets list | grep -i vercel
 OneCLI uses selective secret mode — secrets must be explicitly assigned to each agent. Get the Vercel secret ID from the output above, then assign it to every agent:

 ```bash
-# For each agent, add the Vercel secret to its assigned secrets list.
-# First get current assignments, then set them with the new secret appended.
-VERCEL_SECRET_ID=$(onecli secrets list 2>/dev/null | grep -B2 "Vercel" | grep '"id"' | head -1 | sed 's/.*"id": "//;s/".*//')
-for agent in $(onecli agents list 2>/dev/null | grep '"id"' | sed 's/.*"id": "//;s/".*//'); do
-  CURRENT=$(onecli agents secrets --id "$agent" 2>/dev/null | grep '"' | grep -v hint | grep -v data | sed 's/.*"//;s/".*//' | tr '\n' ',' | sed 's/,$//')
-  onecli agents set-secrets --id "$agent" --secret-ids "${CURRENT:+$CURRENT,}$VERCEL_SECRET_ID"
+# set-secrets replaces the entire list — read and merge for each agent.
+VERCEL_SECRET_ID=$(onecli secrets list | jq -r '.data[] | select(.name | test("(?i)vercel")) | .id' | head -1)
+for agent in $(onecli agents list | jq -r '.data[].id'); do
+  CURRENT=$(onecli agents secrets --id "$agent" | jq -r '[.data[]] | join(",")')
+  MERGED=$(printf '%s' "$CURRENT,$VERCEL_SECRET_ID" | tr ',' '\n' | sort -u | paste -sd ',' -)
+  onecli agents set-secrets --id "$agent" --secret-ids "$MERGED"
 done
 ```

@@ -41,9 +41,12 @@ DELETE FROM messaging_groups WHERE channel_type = 'wechat';

 ### 6. Rebuild and restart

+Run from your NanoClaw project root:
+
 ```bash
 pnpm run build
-systemctl --user restart nanoclaw   # Linux
+source setup/lib/install-slug.sh
+systemctl --user restart $(systemd_unit)              # Linux
 # or
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw   # macOS
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
 ```
@@ -82,12 +82,15 @@ Sync to container: `mkdir -p data/env && cp .env data/env/env`

 ### 2. Start the service and scan the QR

-Restart NanoClaw:
+Restart NanoClaw.
+
+Run from your NanoClaw project root:

 ```bash
-systemctl --user restart nanoclaw   # Linux
+source setup/lib/install-slug.sh
+systemctl --user restart $(systemd_unit)              # Linux
 # or
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw   # macOS
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
 ```

 The adapter will print a **QR URL** to the logs and save it to `data/wechat/qr.txt`:
@@ -44,7 +44,7 @@ import './whatsapp-cloud.js';
 ### 4. Install the adapter package (pinned)

 ```bash
-pnpm install @chat-adapter/whatsapp@4.26.0
+pnpm install @chat-adapter/whatsapp@4.27.0
 ```

 ### 5. Build
@@ -20,6 +20,7 @@ Skip to **Credentials** if all of these are already in place:
 - `setup/whatsapp-auth.ts` and `setup/groups.ts` both exist
 - `setup/index.ts`'s `STEPS` map contains both `'whatsapp-auth':` and `groups:`
 - `@whiskeysockets/baileys`, `qrcode`, `pino` are listed in `package.json` dependencies
+- `.claude/skills/add-whatsapp/scripts/wa-qr-browser.ts` exists (ships with this skill)

 Otherwise continue. Every step below is safe to re-run.

@@ -57,7 +58,7 @@ groups: () => import('./groups.js'),
 ### 5. Install the adapter packages (pinned)

 ```bash
-pnpm install @whiskeysockets/baileys@6.17.16 qrcode@1.5.4 @types/qrcode@1.5.6 pino@9.6.0
+pnpm install @whiskeysockets/baileys@7.0.0-rc.9 qrcode@1.5.4 @types/qrcode@1.5.6 pino@9.6.0
 ```

 ### 6. Build
@@ -95,7 +96,7 @@ If IS_HEADLESS=true AND not WSL → AskUserQuestion: How do you want to authenti
 - **QR code in terminal** - Displays QR code in the terminal (can be too small on some displays)

 Otherwise (macOS, desktop Linux, or WSL) → AskUserQuestion: How do you want to authenticate WhatsApp?
- **QR code in browser** (Recommended) - Opens a browser window with a large, scannable QR code
+- **QR code in browser** (Recommended) - Runs a small local HTTP server that renders the rotating QR as a PNG and auto-opens your default browser
 - **Pairing code** - Enter a numeric code on your phone (no camera needed, requires phone number)
 - **QR code in terminal** - Displays QR code in the terminal (can be too small on some displays)

@@ -114,11 +115,13 @@ rm -rf store/auth/
 For QR code in browser (recommended):

 ```bash
-pnpm exec tsx setup/index.ts --step whatsapp-auth -- --method qr-browser
+pnpm exec tsx .claude/skills/add-whatsapp/scripts/wa-qr-browser.ts
 ```

 (Bash timeout: 150000ms)

+The wrapper spawns `setup/index.ts --step whatsapp-auth -- --method qr`, parses each rotating QR from its `WHATSAPP_AUTH_QR` status blocks, and serves the current QR as a PNG on a local HTTP server (default port `8765`, falls back to a free port). Flags: `--clean` (wipes `store/auth/` before spawning) and `--port N`.
+
 Tell the user:

 > A browser window will open with a QR code.
@@ -130,11 +133,13 @@ Tell the user:
 For QR code in terminal:

 ```bash
-pnpm exec tsx setup/index.ts --step whatsapp-auth -- --method qr-terminal
+pnpm exec tsx setup/index.ts --step whatsapp-auth -- --method qr
 ```

 (Bash timeout: 150000ms)

+The setup driver emits each rotating QR as a `WHATSAPP_AUTH_QR` status block; when run directly (not through `setup:auto`) the raw QR string is printed and your terminal must render it as ASCII. If your terminal can't render it readably, use the browser method above.
+
 Tell the user:

 > 1. Open WhatsApp > **Settings** > **Linked Devices** > **Link a Device**
@@ -200,7 +205,7 @@ Otherwise, run `/manage-channels` to wire this channel to an agent group.

 - **type**: `whatsapp`
 - **terminology**: WhatsApp calls them "groups" and "chats." A "chat" is a 1:1 DM; a "group" has multiple members.
- **how-to-find-id**: DMs use `<phone>@s.whatsapp.net` (e.g. `14155551234@s.whatsapp.net`). Groups use `<id>@g.us`. To find your number: `node -e "const c=JSON.parse(require('fs').readFileSync('store/auth/creds.json','utf-8'));console.log(c.me?.id?.split(':')[0]+'@s.whatsapp.net')"`. Groups are auto-discovered — check `sqlite3 data/v2.db "SELECT platform_id, name FROM messaging_groups WHERE channel_type='whatsapp' AND is_group=1"`.
+- **how-to-find-id**: DMs use `<phone>@s.whatsapp.net` (e.g. `14155551234@s.whatsapp.net`). Groups use `<id>@g.us`. To find your number: `node -e "const c=JSON.parse(require('fs').readFileSync('store/auth/creds.json','utf-8'));console.log(c.me?.id?.split(':')[0]+'@s.whatsapp.net')"`. Groups are auto-discovered — check `pnpm exec tsx scripts/q.ts data/v2.db "SELECT platform_id, name FROM messaging_groups WHERE channel_type='whatsapp' AND is_group=1"`.
 - **supports-threads**: no
 - **typical-use**: Interactive chat — direct messages or small groups
 - **default-isolation**: Same agent group if you're the only participant across multiple chats. Separate agent group if different people are in different groups.
@@ -220,10 +225,10 @@ Not supported (WhatsApp linked device limitation): edit messages, delete message

 ### QR code expired

-QR codes expire after ~60 seconds. Re-run the auth command:
+QR codes expire after ~60 seconds. The browser wrapper rotates automatically as long as it's running; if it was stopped, re-run with `--clean`:

 ```bash
-rm -rf store/auth/ && pnpm exec tsx setup/index.ts --step whatsapp-auth -- --method qr-browser
+pnpm exec tsx .claude/skills/add-whatsapp/scripts/wa-qr-browser.ts --clean
 ```

 ### Pairing code not working
@@ -236,28 +241,31 @@ rm -rf store/auth/ && pnpm exec tsx setup/index.ts --step whatsapp-auth -- --met

 Ensure: digits only (no `+`), phone has internet, WhatsApp is updated.

-If pairing code keeps failing, switch to QR-browser auth instead:
+WhatsApp's pairing-code flow occasionally rejects valid codes with "Couldn't link device — An error happened. Please try again." This is a server-side rejection unrelated to the code itself; we've seen it happen twice in a row on fresh dedicated numbers. If you hit it more than once, switch to QR-browser auth — it has a noticeably higher success rate:

 ```bash
-rm -rf store/auth/ && pnpm exec tsx setup/index.ts --step whatsapp-auth -- --method qr-browser
+pnpm exec tsx .claude/skills/add-whatsapp/scripts/wa-qr-browser.ts --clean
 ```

 ### "waiting for this message" on reactions

-Signal sessions corrupted from rapid restarts. Clear sessions:
+Signal sessions corrupted from rapid restarts. Clear sessions.
+
+Run from your NanoClaw project root:

 ```bash
-systemctl --user stop nanoclaw
+source setup/lib/install-slug.sh
+systemctl --user stop $(systemd_unit)
 rm store/auth/session-*.json
-systemctl --user start nanoclaw
+systemctl --user start $(systemd_unit)
 ```

 ### Bot not responding

 1. Auth exists: `test -f store/auth/creds.json`
 2. Connected: `grep "Connected to WhatsApp" logs/nanoclaw.log | tail -1`
-3. Channel wired: `sqlite3 data/v2.db "SELECT mg.platform_id, mg.name FROM messaging_groups mg JOIN messaging_group_agents mga ON mg.id=mga.messaging_group_id WHERE mg.channel_type='whatsapp'"`
-4. Service running: `systemctl --user status nanoclaw`
+3. Channel wired: `pnpm exec tsx scripts/q.ts data/v2.db "SELECT mg.platform_id, mg.name FROM messaging_groups mg JOIN messaging_group_agents mga ON mg.id=mga.messaging_group_id WHERE mg.channel_type='whatsapp'"`
+4. Service running: `systemctl --user status "$(. setup/lib/install-slug.sh && systemd_unit)"`

 ### "conflict" disconnection

@@ -0,0 +1,246 @@
+/**
+ * scripts/wa-qr-browser.ts — serve WhatsApp pairing QR in the browser.
+ *
+ * Wraps `setup/index.ts --step whatsapp-auth -- --method qr` and renders the
+ * rotating QR string as a PNG in a small local HTTP page. Avoids the unreadable
+ * ASCII terminal QR. macOS / desktop-Linux only — no headless support needed.
+ *
+ * Usage:
+ *   pnpm exec tsx scripts/wa-qr-browser.ts [--clean] [--port 8765]
+ *
+ * --clean   rm -rf store/auth/ before spawning the auth step.
+ * --port N  bind to port N (default 8765, falls back to a free port).
+ */
+import { spawn, exec } from 'node:child_process';
+import http from 'node:http';
+import fs from 'node:fs';
+import path from 'node:path';
+import QRCode from 'qrcode';
+
+type Status = 'waiting' | 'ready' | 'success' | 'failed';
+type State = {
+  qr: string | null;
+  status: Status;
+  error?: string;
+  version: number;
+};
+
+const state: State = { qr: null, status: 'waiting', version: 0 };
+
+const args = process.argv.slice(2);
+const clean = args.includes('--clean');
+const portIdx = args.indexOf('--port');
+const requestedPort = portIdx >= 0 ? Number(args[portIdx + 1]) : 8765;
+
+if (clean) {
+  fs.rmSync(path.join(process.cwd(), 'store', 'auth'), {
+    recursive: true,
+    force: true,
+  });
+  console.log('[wa-qr-browser] cleaned store/auth/');
+}
+
+function htmlPage(): string {
+  return `<!doctype html>
+<html lang="en">
+<head>
+  <meta charset="utf-8" />
+  <meta name="viewport" content="width=device-width,initial-scale=1" />
+  <title>WhatsApp pairing</title>
+  <style>
+    body { margin: 0; min-height: 100vh; display: grid; place-items: center;
+           font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+           background: #0b141a; color: #e9edef; }
+    .card { background: #202c33; padding: 32px 40px; border-radius: 16px;
+            box-shadow: 0 12px 36px rgba(0,0,0,0.4); text-align: center;
+            min-width: 420px; }
+    h1 { font-size: 18px; font-weight: 500; margin: 0 0 20px; color: #aebac1; }
+    .qr-wrap { background: white; padding: 16px; border-radius: 12px;
+               display: inline-block; }
+    #qr { width: 360px; height: 360px; display: block; image-rendering: pixelated; }
+    #status { margin-top: 20px; font-size: 14px; color: #8696a0; min-height: 20px; }
+    #status.ok { color: #00d26a; font-size: 18px; font-weight: 500; }
+    #status.err { color: #ff6b6b; }
+    ol { text-align: left; color: #aebac1; font-size: 13px; line-height: 1.8;
+         margin: 20px 0 0; padding-left: 20px; }
+  </style>
+</head>
+<body>
+  <div class="card">
+    <h1>Scan with WhatsApp</h1>
+    <div class="qr-wrap"><img id="qr" alt="QR code" /></div>
+    <div id="status">Waiting for QR…</div>
+    <ol>
+      <li>Open WhatsApp on your phone</li>
+      <li>Settings &rarr; Linked Devices &rarr; Link a Device</li>
+      <li>Point the camera at this QR code</li>
+    </ol>
+  </div>
+  <script>
+    let lastVersion = -1;
+    const qr = document.getElementById('qr');
+    const status = document.getElementById('status');
+    async function tick() {
+      try {
+        const r = await fetch('/qr.json', { cache: 'no-store' });
+        const s = await r.json();
+        if (s.status === 'success') {
+          qr.style.display = 'none';
+          status.className = 'ok';
+          status.textContent = '✓ Authenticated!';
+          return;
+        }
+        if (s.status === 'failed') {
+          qr.style.display = 'none';
+          status.className = 'err';
+          status.textContent = '✗ ' + (s.error || 'failed');
+          return;
+        }
+        if (s.qr && s.version !== lastVersion) {
+          lastVersion = s.version;
+          qr.src = '/qr.png?v=' + s.version;
+          status.textContent = 'QR ready — scan within ~20s';
+        }
+      } catch (e) { /* server closing, ignore */ }
+      setTimeout(tick, 1500);
+    }
+    tick();
+  </script>
+</body>
+</html>`;
+}
+
+const server = http.createServer(async (req, res) => {
+  const url = req.url ?? '/';
+  if (url === '/' || url.startsWith('/?')) {
+    res.setHeader('content-type', 'text/html; charset=utf-8');
+    res.end(htmlPage());
+    return;
+  }
+  if (url === '/qr.json') {
+    res.setHeader('content-type', 'application/json');
+    res.setHeader('cache-control', 'no-store');
+    res.end(JSON.stringify(state));
+    return;
+  }
+  if (url.startsWith('/qr.png')) {
+    if (!state.qr) {
+      res.statusCode = 404;
+      res.end();
+      return;
+    }
+    try {
+      const buf = await QRCode.toBuffer(state.qr, { width: 360, margin: 1 });
+      res.setHeader('content-type', 'image/png');
+      res.setHeader('cache-control', 'no-store');
+      res.end(buf);
+    } catch (e) {
+      res.statusCode = 500;
+      res.end(String(e));
+    }
+    return;
+  }
+  res.statusCode = 404;
+  res.end();
+});
+
+function listen(port: number): Promise<number> {
+  return new Promise((resolve, reject) => {
+    server.once('error', (err: NodeJS.ErrnoException) => {
+      if (err.code === 'EADDRINUSE' && port === requestedPort) {
+        server.listen(0, () => {
+          const addr = server.address();
+          if (addr && typeof addr === 'object') resolve(addr.port);
+          else reject(new Error('unexpected address'));
+        });
+      } else {
+        reject(err);
+      }
+    });
+    server.listen(port, () => {
+      const addr = server.address();
+      if (addr && typeof addr === 'object') resolve(addr.port);
+      else reject(new Error('unexpected address'));
+    });
+  });
+}
+
+const port = await listen(requestedPort);
+const url = `http://localhost:${port}`;
+console.log(`[wa-qr-browser] QR server on ${url}`);
+
+const opener = process.platform === 'darwin' ? 'open' : 'xdg-open';
+exec(`${opener} ${url}`, (err) => {
+  if (err) console.log(`[wa-qr-browser] could not auto-open browser: ${err.message}`);
+  else console.log('[wa-qr-browser] opening browser…');
+});
+
+const child = spawn(
+  'pnpm',
+  ['exec', 'tsx', 'setup/index.ts', '--step', 'whatsapp-auth', '--', '--method', 'qr'],
+  { stdio: ['inherit', 'pipe', 'inherit'] },
+);
+
+let stdoutBuf = '';
+child.stdout.on('data', (chunk: Buffer) => {
+  const text = chunk.toString();
+  process.stdout.write(text);
+  stdoutBuf += text;
+
+  const blockRe = /=== NANOCLAW SETUP: (\w+) ===\n([\s\S]*?)\n=== END ===/g;
+  let m: RegExpExecArray | null;
+  let lastEnd = 0;
+  while ((m = blockRe.exec(stdoutBuf)) !== null) {
+    const [, name, body] = m;
+    const fields: Record<string, string> = {};
+    for (const line of body.split('\n')) {
+      const kv = line.match(/^(\w+):\s*(.*)$/);
+      if (kv) fields[kv[1]] = kv[2];
+    }
+    handleBlock(name, fields);
+    lastEnd = m.index + m[0].length;
+  }
+  if (lastEnd > 0) stdoutBuf = stdoutBuf.slice(lastEnd);
+});
+
+function handleBlock(name: string, fields: Record<string, string>): void {
+  if (name === 'WHATSAPP_AUTH_QR' && fields.QR) {
+    state.qr = fields.QR;
+    state.status = 'ready';
+    state.version++;
+    return;
+  }
+  if (name === 'WHATSAPP_AUTH') {
+    if (fields.STATUS === 'success') {
+      state.status = 'success';
+      console.log('[wa-qr-browser] authenticated');
+      setTimeout(() => server.close(() => process.exit(0)), 3000);
+    } else if (fields.STATUS === 'skipped') {
+      state.status = 'success';
+      state.error = `already authenticated (${fields.REASON ?? 'unknown'})`;
+      console.log(`[wa-qr-browser] ${state.error}`);
+      setTimeout(() => server.close(() => process.exit(0)), 3000);
+    } else if (fields.STATUS === 'failed') {
+      state.status = 'failed';
+      state.error = fields.ERROR ?? 'unknown error';
+      console.error(`[wa-qr-browser] failed: ${state.error}`);
+    }
+  }
+}
+
+child.on('exit', (code) => {
+  if (state.status === 'success') return;
+  if (state.status !== 'failed') {
+    state.status = 'failed';
+    state.error = `auth process exited (code=${code ?? 'null'})`;
+  }
+  setTimeout(() => {
+    server.close(() => process.exit(1));
+  }, 3000);
+});
+
+process.on('SIGINT', () => {
+  console.log('\n[wa-qr-browser] aborting…');
+  child.kill('SIGTERM');
+  server.close(() => process.exit(130));
+});
@@ -58,7 +58,7 @@ git remote -v
 If `upstream` is missing, add it:

 ```bash
-git remote add upstream https://github.com/qwibitai/nanoclaw.git
+git remote add upstream https://github.com/nanocoai/nanoclaw.git
 ```

 ### Merge the skill branch
@@ -171,9 +171,12 @@ Expected: Both operations succeed.

 ### Full integration test

+Run from your NanoClaw project root:
+
 ```bash
 pnpm run build
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)
 ```

 Send a message via WhatsApp and verify the agent responds.
@@ -88,15 +88,19 @@ Implementation:

 ## After Changes

-Always tell the user:
+Always tell the user.
+
+Run from your NanoClaw project root:
+
 ```bash
 # Rebuild and restart
 pnpm run build
+source setup/lib/install-slug.sh
 # macOS:
-launchctl unload ~/Library/LaunchAgents/com.nanoclaw.plist
-launchctl load ~/Library/LaunchAgents/com.nanoclaw.plist
+launchctl unload ~/Library/LaunchAgents/$(launchd_label).plist
+launchctl load ~/Library/LaunchAgents/$(launchd_label).plist
 # Linux:
-# systemctl --user restart nanoclaw
+# systemctl --user restart $(systemd_unit)
 ```

 ## Example Interaction
@@ -57,7 +57,50 @@ Debug level shows:

 ## Common Issues

-### 1. "Claude Code process exited with code 1"
+### 1. "No adapter for channel type" / Messages silently lost (null platformMsgId)
+
+**Symptom:** The bot stops replying. `logs/nanoclaw.error.log` shows repeated:
+```
+WARN No adapter for channel type channelType="telegram"
+WARN No adapter for channel type channelType="signal"
+```
+The main log shows "Message delivered" entries with `platformMsgId=undefined` — meaning the delivery poll ran, found no adapter, and permanently marked the message as delivered without sending it.
+
+**Root cause: two NanoClaw service instances running simultaneously.**
+
+When a second service instance (often `nanoclaw-v2-<id>.service` running alongside `nanoclaw.service`) is active with a stale binary, it has no channel adapters registered. Its delivery poll races against the working instance and wins — permanently marking outbound messages as delivered without ever sending them.
+
+**Diagnosis:**
+```bash
+# Check for duplicate running instances
+ps aux | grep 'nanoclaw/dist/index.js' | grep -v grep
+
+# Check which services are active
+systemctl --user list-units 'nanoclaw*' --all
+
+# Confirm channel adapters registered by the current process
+grep "Channel adapter started" logs/nanoclaw.log | tail -10
+```
+
+**Fix:**
+1. Identify which service has the correct binary and EnvironmentFile (the one showing `signal`, `telegram`, `cli` all started in the log).
+2. Stop and disable the stale duplicate service:
+   ```bash
+   systemctl --user stop nanoclaw.service   # or whichever is the old one
+   systemctl --user disable nanoclaw.service
+   ```
+3. If the remaining service unit is missing `EnvironmentFile`, add it:
+   ```bash
+   # Edit the service unit — add this line under [Service]:
+   # EnvironmentFile=/home/[user]/nanoclaw/.env
+   systemctl --user daemon-reload
+   systemctl --user restart nanoclaw-v2-<id>.service
+   ```
+4. Verify only one instance runs: `ps aux | grep nanoclaw/dist/index.js | grep -v grep`
+
+**Note:** Messages that were marked delivered with a null `platform_message_id` cannot be automatically retried — they are permanently lost. The user must resend their message.
+
+### 2. "Claude Code process exited with code 1"

 **Check the container log file** in `groups/{folder}/logs/container-*.log`

@@ -279,7 +322,7 @@ rm -rf data/sessions/
 rm -rf data/sessions/{groupFolder}/.claude/

 # Also clear the session ID from NanoClaw's tracking (stored in SQLite)
-sqlite3 store/messages.db "DELETE FROM sessions WHERE group_folder = '{groupFolder}'"
+pnpm exec tsx scripts/q.ts store/messages.db "DELETE FROM sessions WHERE group_folder = '{groupFolder}'"
 ```

 To verify session resumption is working, check the logs for the same session ID across messages:
@@ -9,7 +9,7 @@ Stand up the first NanoClaw agent for a channel and verify end-to-end delivery b

 ## Prerequisites

- **Service running.** Check: `launchctl list | grep nanoclaw` (macOS) or `systemctl --user status nanoclaw` (Linux). If stopped, tell the user to run `/setup` first.
+- **Service running.** Check: `launchctl list | grep "$(. setup/lib/install-slug.sh && launchd_label)"` (macOS) or `systemctl --user status "$(. setup/lib/install-slug.sh && systemd_unit)"` (Linux). If stopped, tell the user to run `/setup` first.
 - **Target channel installed.** At least one `/add-<channel>` skill has run, credentials are in `.env`, and the adapter is uncommented in `src/channels/index.ts`.
 - **Adapter connected.** Tail `logs/nanoclaw.log` — look for a recent `channel setup` / `adapter connected` line for the target channel.

@@ -54,7 +54,7 @@ Tell the user:
 Wait for the user's confirmation. Then look up the most recent DM messaging groups:

 ```bash
-sqlite3 data/v2.db "SELECT id, platform_id, name, created_at FROM messaging_groups WHERE channel_type='${CHANNEL}' AND is_group=0 ORDER BY created_at DESC LIMIT 5"
+pnpm exec tsx scripts/q.ts data/v2.db "SELECT id, platform_id, name, created_at FROM messaging_groups WHERE channel_type='${CHANNEL}' AND is_group=0 ORDER BY created_at DESC LIMIT 5"
 ```

 Show the top rows to the user and confirm which `platform_id` is theirs (usually the most recent). Record as `PLATFORM_ID`. If none appeared, check `logs/nanoclaw.log` for `unknown_sender` drops — the adapter might be rejecting inbound due to connection or permission issues.
@@ -103,7 +103,7 @@ Wait for the user's reply. If they confirm receipt, the skill is done.

 If they say it didn't arrive, then diagnose using the DB directly (no waiting loops required — the message either delivered or it didn't):

- `sqlite3 data/v2-sessions/<agent-group-id>/sessions/<session-id>/outbound.db "SELECT id, status, created_at FROM messages_out ORDER BY created_at DESC LIMIT 5"` — check for stuck `pending` rows. Replace `<agent-group-id>` and `<session-id>` with the values from the script's output.
+- `pnpm exec tsx scripts/q.ts data/v2-sessions/<agent-group-id>/sessions/<session-id>/outbound.db "SELECT id, status, created_at FROM messages_out ORDER BY created_at DESC LIMIT 5"` — check for stuck `pending` rows. Replace `<agent-group-id>` and `<session-id>` with the values from the script's output.
 - `grep -E 'Unauthorized channel destination|container.*exited|error' logs/nanoclaw.log | tail -20` — look for ACL rejections or container crashes.
 - `ls data/v2-sessions/<agent-group-id>/sessions/*/outbound.db` — confirm the session exists.

@@ -98,13 +98,13 @@ for i in $(seq 1 15); do
 done
 ```

-If it never becomes healthy, check if the gateway process is running:
+If it never becomes healthy, check the gateway containers. The gateway is a Docker Compose stack (project `onecli`, compose file at `~/.onecli/docker-compose.yml`), **not** a host process — `ps aux | grep onecli` will not find it, and there is no `onecli start` command (removed in OneCLI 1.4.x).

 ```bash
-ps aux | grep -i onecli | grep -v grep
+docker ps -a --filter "label=com.docker.compose.project=onecli" --format '{{.Names}}\t{{.Status}}'
 ```

-If it's not running, try starting it manually: `onecli start`. If that fails, show the error and stop — the user needs to debug their OneCLI installation.
+Both services have `restart: unless-stopped`, so they come back automatically once the Docker daemon is up. If Docker isn't running, start it (`open -a Docker` on macOS) and they'll restart on their own. To bring the stack up manually: `docker compose -f ~/.onecli/docker-compose.yml up -d`. If that fails, show the error and stop — the user needs to debug their OneCLI installation.

 ## Phase 3: Migrate existing credentials

@@ -236,9 +236,12 @@ pnpm run build

 If build fails, diagnose and fix. Common issue: `@onecli-sh/sdk` not installed — run `pnpm install` first.

-Restart the service:
- macOS (launchd): `launchctl kickstart -k gui/$(id -u)/com.nanoclaw`
- Linux (systemd): `systemctl --user restart nanoclaw`
+Restart the service.
+
+Run from your NanoClaw project root:
+
+- macOS (launchd): `launchctl kickstart -k gui/$(id -u)/"$(. setup/lib/install-slug.sh && launchd_label)"`
+- Linux (systemd): `systemctl --user restart "$(. setup/lib/install-slug.sh && systemd_unit)"`
 - WSL/manual: stop and re-run `bash start-nanoclaw.sh`

 ## Phase 5: Verify
@@ -259,9 +262,44 @@ Tell the user:
 - To manage secrets: `onecli secrets list`, or open ${ONECLI_URL}
 - To add rate limits or policies: `onecli rules create --help`

+## Granting secrets to agents (safe merge)
+
+`set-secrets` **replaces** the agent's entire secret list — it never appends. Always read the current list first and merge before calling it. This pattern is canonical across all skills that assign secrets:
+
+```bash
+AGENT_ID=$(onecli agents list | jq -r '.data[] | select(.identifier=="<agentGroupId>") | .id')
+CURRENT=$(onecli agents secrets --id "$AGENT_ID" | jq -r '[.data[]] | join(",")')
+MERGED=$(printf '%s' "$CURRENT,<new-secret-id>" | tr ',' '\n' | sort -u | paste -sd ',' -)
+onecli agents set-secrets --id "$AGENT_ID" --secret-ids "$MERGED"
+onecli agents secrets --id "$AGENT_ID"
+```
+
+- `<agentGroupId>` — the `agentGroupId` field in `groups/<folder>/container.json`
+- `<new-secret-id>` — the `id` from `onecli secrets list`
+- Multiple new secrets: append them comma-separated before the `printf` step
+
+### git over HTTPS
+
+OneCLI's proxy injects credentials proactively — `injections_applied=1` appears in `docker logs onecli` even when git sends no auth header. However, OneCLI sets `SSL_CERT_FILE` for Node/Python/Deno but not `GIT_SSL_CAINFO`. Without it, git rejects the OneCLI MITM certificate.
+
+**Auth format matters**: GitHub's git smart HTTP protocol (`github.com`) requires `Basic` auth, not `Bearer`. GitHub's REST API (`api.github.com`) accepts `Bearer`. These must be configured as separate secrets with different formats — see `/add-github` for the full setup.
+
+If an agent uses `git` or `gh`, add to `data/v2-sessions/<agent-group-id>/.claude-shared/settings.json`:
+
+```json
+"GIT_SSL_CAINFO": "/tmp/onecli-combined-ca.pem",
+"GIT_TERMINAL_PROMPT": "0",
+"GIT_CONFIG_COUNT": "1",
+"GIT_CONFIG_KEY_0": "credential.helper",
+"GIT_CONFIG_VALUE_0": "",
+"GH_TOKEN": "ghp_onecli_proxy_replaces_this"
+```
+
+**Debugging injection**: `docker logs onecli 2>&1 | grep "github.com"` shows every request with `injections_applied=N` and the HTTP status. If `injections_applied=1` but status is still 401, the injected credential value is wrong or uses the wrong auth format for that endpoint.
+
 ## Troubleshooting

-**"OneCLI gateway not reachable" in logs:** The gateway isn't running. Check with `curl -sf ${ONECLI_URL}/health`. Start it with `onecli start` if needed.
+**"OneCLI gateway not reachable" in logs:** The gateway isn't running. Check with `curl -sf ${ONECLI_URL}/health`. The most common cause is that Docker itself is down (the gateway is a Compose stack) — start Docker (`open -a Docker` on macOS) and the containers restart automatically. To bring them up manually: `docker compose -f ~/.onecli/docker-compose.yml up -d`.

 **Container gets no credentials:** Verify `ONECLI_URL` is set in `.env` and the gateway has an Anthropic secret (`onecli secrets list`).

@@ -11,7 +11,22 @@ Privilege is a **user-level** concept, not a channel-level one (see `src/db/user

 ## Assess Current State

-Read the central DB (`data/v2.db`) — query `agent_groups`, `messaging_groups`, `messaging_group_agents`, `users`, and `user_roles` tables. Also check `.env` for channel tokens and `src/channels/index.ts` for uncommented imports.
+Read the central DB (`data/v2.db`) using these canonical queries (column names match the schema, not the CLI flags — the `register` command's `--assistant-name` is stored in `agent_groups.name`).
+
+Run each via the in-tree wrapper — the host setup deliberately ships no `sqlite3` CLI:
+
+```bash
+pnpm exec tsx scripts/q.ts data/v2.db "<query>"
+```
+
+```sql
+SELECT id, name AS assistant_name, folder, agent_provider FROM agent_groups;
+SELECT id, channel_type, platform_id, name, unknown_sender_policy FROM messaging_groups;
+SELECT messaging_group_id, agent_group_id, session_mode, priority FROM messaging_group_agents;
+SELECT user_id, role, agent_group_id FROM user_roles ORDER BY role='owner' DESC;
+```
+
+Also check `.env` for channel tokens and `src/channels/index.ts` for uncommented imports.

 Categorize channels as: **wired** (has DB entities + messaging_group_agents row), **configured but unwired** (has credentials + barrel import, no DB entities), or **not configured**.

@@ -41,7 +41,12 @@ npx tsx setup/index.ts --step mounts --force -- --empty

 ## After Changes

-Restart the service so containers pick up the new config:
+Restart the service so containers pick up the new config (the unit/label names are per-install — see `setup/lib/install-slug.sh`).

- macOS: `launchctl kickstart -k gui/$(id -u)/com.nanoclaw`
- Linux: `systemctl --user restart nanoclaw`
+Run from your NanoClaw project root:
+
+```bash
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+systemctl --user restart $(systemd_unit)              # Linux
+```
@@ -0,0 +1,232 @@
+---
+name: migrate-from-v1
+description: Finish migrating a NanoClaw v1 install into v2. Run after `bash migrate-v2.sh` completes. Seeds the owner, cleans up CLAUDE.local.md files, reconciles container configs, and helps port custom v1 code. Triggers on "migrate from v1", "finish migration", "v1 migration".
+---
+
+# Finish v1 → v2 migration
+
+`bash migrate-v2.sh` already ran the deterministic migration. It handled:
+
+- .env keys merged
+- v2 DB seeded (agent_groups, messaging_groups, wiring)
+- Group folders copied (v1 CLAUDE.md → v2 CLAUDE.local.md)
+- Session data copied with conversation continuity (incl. Claude Code memory + JSONL transcripts)
+- Scheduled tasks ported
+- Channel code installed and auth state copied (incl. WhatsApp Baileys keystore)
+- WhatsApp LIDs resolved from `store/auth` and aliased into `messaging_groups`
+- Container skills copied
+- Container image built
+
+Your job is the parts that need human judgment: triage any failed steps, seed the owner, clean up CLAUDE.local.md files, reconcile configs, and port any fork customizations.
+
+Read `logs/setup-migration/handoff.json` first — it has `overall_status`, per-step results in `steps`, and a `followups` list.
+
+## Preflight: was the script run?
+
+Before anything else, check that `logs/setup-migration/handoff.json` exists. If it doesn't, the user is invoking this skill before `migrate-v2.sh` ran. Stop and tell them, verbatim:
+
+> This skill finishes a migration that `migrate-v2.sh` started. Run that first, in your terminal — not from inside Claude:
+>
+> ```bash
+> bash migrate-v2.sh
+> ```
+>
+> It needs interactive prompts (channel selection, service switchover) and runs Node/pnpm bootstrap, Docker, OneCLI setup, and a container build that don't fit inside a Claude session. When it finishes, it'll hand control back to Claude automatically — at which point this skill picks up.
+
+Do not attempt to run the script yourself, simulate its effects, or pick up the migration mid-stream. The deterministic side has dependencies on a real interactive shell.
+
+Once `handoff.json` exists, proceed to Phase 0.
+
+## Phase 0: Get v2 routing real messages
+
+Before any deeper migration work, prove v2 actually answers messages on the user's real channels. v1 is paused, not touched — flipping back is a service restart.
+
+### 0a — Fix blockers only
+
+Walk `handoff.steps`. Fix only the failures that would stop the bot from routing one message; defer the rest to its later phase.
+
+### 0b — Smoke test, then continue
+
+Tell the user the switch is non-destructive (v1 is paused, not modified; reverting is one command). Help them stop v1's service unit and start v2's, tail the host log for a clean boot, and have them send a real test message. Use `AskUserQuestion` to confirm the bot responded.
+
+If yes, continue to Phase 1. If no, diagnose from `logs/nanoclaw.log` and re-test — don't proceed to deeper work on a broken router.
+
+### Deferred failures
+
+Re-visit anything you skipped in 0a before declaring the migration done. Most surface naturally in later phases (`1c-groups` ↔ Phase 2, `1e-tasks` ↔ task verification).
+
+## Phase 1: Owner and access
+
+v2 auto-creates a `users` row for every sender it sees (via `extractAndUpsertUser` in `src/modules/permissions/index.ts`). By the time this skill runs, the owner's row likely already exists — it just needs the `owner` role granted.
+
+**User ID format**: always `<channel_type>:<platform_handle>`. Each channel populates this differently:
+- **Telegram**: `telegram:<numeric_user_id>` (e.g. `telegram:6037840640`)
+- **Discord**: `discord:<snowflake_user_id>` (e.g. `discord:123456789012345678`)
+- **WhatsApp**: `whatsapp:<phone>@s.whatsapp.net` (e.g. `whatsapp:14155551234@s.whatsapp.net`)
+- **Slack**: `slack:<user_id>` (e.g. `slack:U04ABCDEF`)
+- **Others**: `<channel_type>:<platform_id>`
+
+**Steps:**
+
+1. Query `users` table: `SELECT id, kind, display_name FROM users`.
+2. If exactly one user exists, confirm: `AskUserQuestion`: "Is `<display_name>` (`<id>`) you?" — Yes / No, let me type it.
+3. If multiple users exist, present them as options in `AskUserQuestion`.
+4. If no users exist yet (service hasn't received a message), ask the user to send a test message first, then re-query.
+5. Once confirmed, check `user_roles` — if the owner role already exists, skip. Otherwise insert:
+   ```sql
+   INSERT INTO user_roles (user_id, role, agent_group_id, granted_by, granted_at)
+   VALUES ('<user_id>', 'owner', NULL, NULL, datetime('now'))
+   ```
+
+Use the DB helpers in `src/db/user-roles.ts` — they keep indexes correct. Init the DB first:
+
+```ts
+import { initDb } from '../src/db/connection.js';
+import { runMigrations } from '../src/db/migrations/index.js';
+import { DATA_DIR } from '../src/config.js';
+import path from 'path';
+const db = initDb(path.join(DATA_DIR, 'v2.db'));
+runMigrations(db);
+```
+
+### Access policy
+
+After seeding the owner, discuss the access policy. v2's `messaging_groups.unknown_sender_policy` controls who can interact with the bot. `migrate-v2.sh` set it to `public` so the bot would respond during the switchover test, but the user may want to tighten it.
+
+Present the options via `AskUserQuestion`:
+
+1. **Public** (current) — anyone can message the bot. Good for personal DM bots.
+2. **Known users only** — only users in `agent_group_members` can trigger the bot. Others are silently dropped.
+3. **Approval required** — unknown senders trigger an approval request to the owner. Good for group chats where you want to vet new members.
+
+If the user picks option 2 or 3, seed the known users from v1's message history. The v1 database is at `<handoff.v1_path>/store/messages.db`. It has a `messages` table with `sender` and `sender_name` columns. For each group:
+
+```sql
+-- v1: unique senders per chat (excluding bot messages)
+SELECT DISTINCT sender, sender_name
+FROM messages
+WHERE chat_jid = '<v1_jid>' AND is_from_me = 0 AND sender IS NOT NULL
+```
+
+The `sender` value is a platform handle (e.g. `6037840640` for Telegram). Build the v2 user ID by inferring the channel type from the chat JID prefix (use `parseJid` from `setup/migrate-v2/shared.ts`) and combining: `<channel_type>:<sender>`.
+
+For each sender:
+1. Upsert into `users(id, kind, display_name)` if not already present.
+2. Insert into `agent_group_members(user_id, agent_group_id)` for each agent group wired to that messaging group.
+
+Show the user the list of senders being imported and let them deselect any they don't want.
+
+Then update the messaging groups:
+```sql
+UPDATE messaging_groups SET unknown_sender_policy = '<chosen_policy>'
+WHERE id IN (SELECT id FROM messaging_groups WHERE channel_type IN (<migrated_channels>))
+```
+
+## Phase 2: Clean up CLAUDE.local.md
+
+The migration copied v1's entire CLAUDE.md into CLAUDE.local.md for each group. This file now contains v1 boilerplate that v2 handles through its own composed fragments (`container/CLAUDE.md` + `.claude-fragments/module-*.md`). The user's customizations are buried inside.
+
+For each group that has a `CLAUDE.local.md`:
+
+1. Read the file.
+2. Read the v1 template it was based on. Determine which template by checking the v1 install:
+   - If the group had `is_main=1` in v1's `registered_groups`, the template was `groups/main/CLAUDE.md`
+   - Otherwise, the template was `groups/global/CLAUDE.md`
+   - The v1 path is in `handoff.json` → `v1_path`
+3. Diff the file against the template. Identify sections that are:
+   - **Stock boilerplate** (identical to template) — remove. v2's fragments cover this.
+   - **User customizations** (added sections, modified sections) — keep.
+4. The following v1 sections are now handled by v2 fragments and should be removed even if slightly modified:
+   - "What You Can Do" → v2 runtime system prompt
+   - "Communication" / "Internal thoughts" / "Sub-agents" → `container/CLAUDE.md` + `module-core.md`
+   - "Your Workspace" / workspace path references → `container/CLAUDE.md`
+   - "Memory" (the stock version) → `container/CLAUDE.md`
+   - "Message Formatting" → `container/CLAUDE.md`
+   - "Admin Context" → v2 uses `user_roles`, not is_main
+   - "Authentication" → v2 uses OneCLI
+   - "Container Mounts" → v2 mounts are different
+   - "Managing Groups" / "Finding Available Groups" / "Registered Groups Config" → v2 entity model, no IPC
+   - "Global Memory" → v2 has `.claude-shared.md` symlink
+   - "Scheduling for Other Groups" → `module-scheduling.md`
+   - "Task Scripts" → `module-scheduling.md`
+   - "Sender Allowlist" → v2 uses `unknown_sender_policy` + `user_roles`
+5. Fix path references in kept sections:
+   - `/workspace/group/` → `/workspace/agent/`
+   - `/workspace/project/` → these paths don't exist in v2; discuss with the user
+   - `/workspace/ipc/` → gone; remove references
+   - `/workspace/extra/` → v2 uses `container.json` `additionalMounts`; keep but note the path may change
+6. Keep the `# Name` heading and first paragraph (identity) — this is the user's agent personality.
+7. Show the user the proposed new CLAUDE.local.md before writing it. Use `AskUserQuestion`: "Here's what I'd keep — look right?" with options to approve, edit, or keep the original.
+
+If a CLAUDE.local.md has no user customizations (pure template copy), write a minimal file with just the identity heading.
+
+## Phase 3: Container config
+
+`migrate-v2.sh` writes `container.json` directly from v1's `container_config` (the `additionalMounts` shape is identical). If the v1 config was unparseable, it falls back to a `.v1-container-config.json` sidecar.
+
+For each group, check:
+
+1. If `container.json` exists, read it and verify the `additionalMounts` host paths are still valid on this machine. Flag any that don't exist.
+2. If `.v1-container-config.json` exists (parse failure fallback), read it, discuss with the user, and write a proper `container.json`. Then delete the sidecar.
+3. Check for `env` or `packages` fields — `env` may overlap with OneCLI vault, `packages` (apt/npm) are portable.
+
+## Phase 4: Fork customizations
+
+Check whether the user's v1 install was a customized fork.
+
+```bash
+cd <v1_path>
+git remote -v
+git log --oneline <upstream>/main..HEAD 2>/dev/null
+```
+
+If no commits ahead of upstream: stock v1, skip this phase.
+
+If there are commits:
+
+1. Show the commit list to the user.
+2. `AskUserQuestion`: "How do you want to handle your v1 customizations?"
+   - **Copy portable items** (recommended) — copy `container/skills/*`, `.claude/skills/*`, `docs/*`. Scan each with `scanForV1Patterns` from `setup/migrate-v2/shared.ts`.
+   - **Full walkthrough** — go commit by commit, decide together.
+   - **Reference only** — stash to `docs/v1-fork-reference/` for later.
+3. Source code (`src/*`, `container/agent-runner/src/*`) is NOT portable — v2's architecture is fundamentally different. Stash to `docs/v1-fork-reference/` with a README explaining what each file did. Don't translate.
+
+## Principles
+
+- **v1 checkout is read-only.** Never modify files under `handoff.v1_path`.
+- **Show before writing.** Show diffs/proposed content before modifying CLAUDE.local.md or container.json.
+- **Mask credentials** when displaying (first 4 + `...` + last 4 characters).
+- **`handoff.json` is the recovery point.** If context gets compacted, re-read it and `git status` to recover state.
+
+## Setup steps you can run
+
+The setup flow at `setup/index.ts` has individual steps you can invoke if something is missing or failed:
+
+```bash
+pnpm exec tsx setup/index.ts --step <name>
+```
+
+| Step | When to use |
+|------|-------------|
+| `onecli` | OneCLI not installed or not healthy |
+| `auth` | No Anthropic credential in vault |
+| `container` | Container image needs rebuild |
+| `service` | Service not installed or not running |
+| `mounts` | Mount allowlist missing |
+| `verify` | End-to-end health check (run after everything else) |
+| `environment` | System check (Node, dirs) |
+
+## When done
+
+1. Run the verify step to confirm everything works:
+   ```bash
+   pnpm exec tsx setup/index.ts --step verify
+   ```
+2. Delete `logs/setup-migration/handoff.json` — offer to save as `docs/migration-<date>.md` first.
+3. Restart the service if running so changes take effect:
+   ```bash
+   # Linux
+   systemctl --user restart nanoclaw-v2-*
+   # macOS
+   launchctl kickstart -k gui/$(id -u)/com.nanoclaw-v2-*
+   ```
@@ -28,13 +28,22 @@ Two phases: **Extract** (build the migration guide) and **Upgrade** (use it). If

 ---

+# Phase 0: Refresh this skill first
+
+The migration process itself evolves, so run its newest version before doing anything else:
+- Ensure the `upstream` remote exists (default `https://github.com/nanocoai/nanoclaw.git`) and fetch: `git fetch upstream --prune`. Detect the upstream branch (`main` or `master`).
+- Refresh this skill from upstream: `git checkout upstream/<branch> -- .claude/skills/migrate-nanoclaw/`
+- Re-read `.claude/skills/migrate-nanoclaw/SKILL.md`. If it changed, **follow the updated version from the top** instead of this one.
+
+This is the only working-tree change expected before the preflight check below; changes limited to `.claude/skills/migrate-nanoclaw/` are this self-refresh — ignore them in the 1.0 clean-tree check and proceed.
+
 # Phase 1: Extract

 ## 1.0 Preflight

 Run `git status --porcelain`. If non-empty, offer to stash or commit for them (AskUserQuestion: "Stash changes" / "Commit changes" / "I'll handle it"). If they want to commit, stage and commit with a descriptive message. If they want to stash, run `git stash push -m "pre-migration stash"`.

-Check remotes with `git remote -v`. If `upstream` is missing, ask for the URL (default: `https://github.com/qwibitai/nanoclaw.git`), add it, then `git fetch upstream --prune`.
+Check remotes with `git remote -v`. If `upstream` is missing, ask for the URL (default: `https://github.com/nanocoai/nanoclaw.git`), add it, then `git fetch upstream --prune`.

 Detect upstream branch: check `git branch -r | grep upstream/` for `main` or `master`. Store as UPSTREAM_BRANCH.

@@ -463,6 +472,11 @@ Do NOT use `git checkout -B` to create an intermediate branch — this caused is

 Run `npm install && pnpm run build` in the main tree to confirm.

+Stamp the upgrade marker (required — without it the startup tripwire stops the host on next start). Only do this after the build above succeeds:
+```bash
+pnpm exec tsx scripts/upgrade-state.ts set "" migrate-nanoclaw
+```
+
 Restart the service:
 ```bash
 launchctl load ~/Library/LaunchAgents/com.nanoclaw.plist
@@ -11,14 +11,15 @@ Run `/update-nanoclaw` in Claude Code.

 ## How it works

-**Preflight**: checks for clean working tree (`git status --porcelain`). If `upstream` remote is missing, asks you for the URL (defaults to `https://github.com/qwibitai/nanoclaw.git`) and adds it. Detects the upstream branch name (`main` or `master`).
+**Preflight**: checks for clean working tree (`git status --porcelain`). If `upstream` remote is missing, asks you for the URL (defaults to `https://github.com/nanocoai/nanoclaw.git`) and adds it. Detects the upstream branch name (`main` or `master`).

 **Backup**: creates a timestamped backup branch and tag (`backup/pre-update-<hash>-<timestamp>`, `pre-update-<hash>-<timestamp>`) before touching anything. Safe to run multiple times.

 **Preview**: runs `git log` and `git diff` against the merge base to show upstream changes since your last sync. Groups changed files into categories:
 - **Skills** (`.claude/skills/`): unlikely to conflict unless you edited an upstream skill
- **Source** (`src/`): may conflict if you modified the same files
- **Build/config** (`package.json`, `tsconfig*.json`, `container/`): review needed
+- **Host source** (`src/`): may conflict if you modified the same files
+- **Container** (`container/`): triggers container rebuild
+- **Build/config** (`package.json`, `pnpm-lock.yaml`, `tsconfig*.json`): lockfile changes trigger dep install

 **Update paths** (you pick one):
 - `merge` (default): `git merge upstream/<branch>`. Resolves all conflicts in one pass.
@@ -30,7 +31,7 @@ Run `/update-nanoclaw` in Claude Code.

 **Conflict resolution**: opens only conflicted files, resolves the conflict markers, keeps your local customizations intact.

-**Validation**: runs `pnpm run build` and `pnpm test`.
+**Validation**: runs `pnpm run build` and `pnpm test`. If container files changed, also runs the container typecheck and `./container/build.sh`.

 **Breaking changes check**: after validation, reads CHANGELOG.md for any `[BREAKING]` entries introduced by the update. If found, shows each breaking change and offers to run the recommended skill to migrate.

@@ -59,16 +60,25 @@ Help a user with a customized NanoClaw install safely incorporate upstream chang
 - Default to MERGE (one-pass conflict resolution). Offer REBASE as an explicit option.
 - Keep token usage low: rely on `git status`, `git log`, `git diff`, and open only conflicted files.

+# Step 0a: Refresh this skill first
+The update process itself evolves, so run its newest version before doing anything else:
+- Ensure the `upstream` remote exists (default `https://github.com/nanocoai/nanoclaw.git`) and fetch: `git fetch upstream --prune`. Detect the upstream branch (`main` or `master`).
+- Refresh this skill from upstream: `git checkout upstream/<branch> -- .claude/skills/update-nanoclaw/`
+- Re-read `.claude/skills/update-nanoclaw/SKILL.md`. If it changed, **follow the updated version from the top** instead of this one.
+
+This is the only working-tree change expected before the preflight check; the full update commits it along with everything else.
+
 # Step 0: Preflight (stop early if unsafe)
 Run:
 - `git status --porcelain`
 If output is non-empty:
 - Tell the user to commit or stash first, then stop.
+- Exception: changes limited to `.claude/skills/update-nanoclaw/` are the Step 0a self-refresh — ignore those and proceed.

 Confirm remotes:
 - `git remote -v`
 If `upstream` is missing:
- Ask the user for the upstream repo URL (default: `https://github.com/qwibitai/nanoclaw.git`).
+- Ask the user for the upstream repo URL (default: `https://github.com/nanocoai/nanoclaw.git`).
 - Add it: `git remote add upstream <user-provided-url>`
 - Then: `git fetch upstream --prune`

@@ -108,9 +118,10 @@ Show file-level impact from upstream:

 Bucket the upstream changed files:
 - **Skills** (`.claude/skills/`): unlikely to conflict unless the user edited an upstream skill
- **Source** (`src/`): may conflict if user modified the same files
- **Build/config** (`package.json`, `pnpm-lock.yaml`, `tsconfig*.json`, `container/`, `launchd/`): review needed
- **Other**: docs, tests, misc
+- **Host source** (`src/`): may conflict if user modified the same files
+- **Container** (`container/`): triggers container rebuild (+ typecheck if `agent-runner/src/` changed)
+- **Build/config** (`package.json`, `pnpm-lock.yaml`, `tsconfig*.json`): lockfile changes trigger dep install
+- **Other**: docs, tests, setup scripts, misc

 **Large drift check:** If the upstream commit count and age suggest the user has a lot of catching up to do, mention that `/migrate-nanoclaw` might be a better fit — it extracts customizations and reapplies them on clean upstream instead of merging. Offer it as an option but don't push.

@@ -173,11 +184,31 @@ If it gets messy (more than 3 rounds of conflicts):
  - `git rebase --abort`
  - Recommend merge instead.

+# Step 4.5: Install dependencies (if lockfiles changed)
+Check if the merge changed any lockfiles or package manifests:
+- `git diff <backup-tag-from-step-1>..HEAD --name-only | grep -E '^(pnpm-lock\.yaml|package\.json)$'`
+  - If matched: `pnpm install`
+- `git diff <backup-tag-from-step-1>..HEAD --name-only | grep -E '^container/agent-runner/(bun\.lock|package\.json)$'`
+  - If matched AND `command -v bun` succeeds: `cd container/agent-runner && bun install`
+  - If bun is not installed on the host, skip — container deps will be installed during `./container/build.sh`
+
+Skip this step if neither lockfile changed.
+
 # Step 5: Validation
-Run:
+Check which areas changed to determine what to validate:
+- `CHANGED_FILES=$(git diff --name-only <backup-tag-from-step-1>..HEAD)`
+
+**Host build** (always):
 - `pnpm run build`
 - `pnpm test` (do not fail the flow if tests are not configured)

+**Container typecheck** (only if `container/agent-runner/src/` files are in CHANGED_FILES AND bun types are available):
+- Check: `pnpm exec tsc -p container/agent-runner/tsconfig.json --noEmit`
+- If this fails because bun types are missing (`Cannot find type definition file for 'bun'`), skip with a note — type errors will surface at container runtime instead
+
+**Container image rebuild** (only if any `container/` files are in CHANGED_FILES):
+- `./container/build.sh`
+
 If build fails:
 - Show the error.
 - Only fix issues clearly caused by the merge (missing imports, type mismatches from merged code).
@@ -209,8 +240,10 @@ If one or more `[BREAKING]` lines are found:
 - For each skill the user selects, invoke it using the Skill tool.
 - After all selected skills complete (or if user chose Skip), proceed to Step 7 (skill updates check).

-# Step 7: Check for skill updates
-After the summary, check if skills are distributed as branches in this repo:
+# Step 7: Check for skill and channel/provider updates
+
+## 7a: Skill branches
+Check if skills are distributed as branches in this repo:
 - `git branch -r --list 'upstream/skill/*'`

 If any `upstream/skill/*` branches exist:
@@ -218,7 +251,31 @@ If any `upstream/skill/*` branches exist:
  - Option 1: "Yes, check for updates" (description: "Runs /update-skills to check for and apply skill branch updates")
  - Option 2: "No, skip" (description: "You can run /update-skills later any time")
 - If user selects yes, invoke `/update-skills` using the Skill tool.
- After the skill completes (or if user selected no), proceed to Step 8.
+
+## 7b: Channel and provider updates
+Detect installed channels by reading `src/channels/index.ts` and collecting all `import './<name>.js';` lines (excluding `cli`). For providers, check `src/providers/index.ts` the same way.
+
+If any channels/providers are installed AND `upstream/channels` or `upstream/providers` branches exist:
+- List the installed channels/providers.
+- Use AskUserQuestion to ask: "Would you like to update your installed channels/providers? Re-running `/add-<name>` is safe — it only updates code files, credentials and wiring are untouched."
+  - One option per installed channel/provider (e.g., "Update Slack (/add-slack)")
+  - "Skip — I'll update them later"
+  - Set `multiSelect: true`
+- For each selected option, invoke the corresponding `/add-<channel>` or `/add-<provider>` skill.
+
+If no channels/providers are installed, skip silently.
+
+Proceed to Step 7.9.
+
+# Step 7.9: Stamp the upgrade marker (required)
+After validation has **succeeded**, record that this install reached the new version through the supported path. Without this, the startup tripwire stops the host on its next start.
+
+- `pnpm exec tsx scripts/upgrade-state.ts set "" update-nanoclaw`
+  - The empty version argument stamps the current `package.json` version.
+
+If validation did NOT succeed, do not stamp — leave the tripwire to catch the broken state.
+
+Proceed to Step 8.

 # Step 8: Summary + rollback instructions
 Show:
@@ -232,9 +289,10 @@ Show:
 Tell the user:
 - To rollback: `git reset --hard <backup-tag-from-step-1>`
 - Backup branch also exists: `backup/pre-update-<HASH>-<TIMESTAMP>`
- Restart the service to apply changes:
-  - If using launchd: `launchctl unload ~/Library/LaunchAgents/com.nanoclaw.plist && launchctl load ~/Library/LaunchAgents/com.nanoclaw.plist`
-  - If running manually: restart `pnpm run dev`
+- Restart the service to apply changes. The unit/label names are per-install — derive them with `setup/lib/install-slug.sh`. Run from your NanoClaw project root:
+  - **macOS (Darwin)**: `source setup/lib/install-slug.sh && launchctl kickstart -k gui/$(id -u)/$(launchd_label)`
+  - **Linux**: `source setup/lib/install-slug.sh && systemctl --user restart $(systemd_unit)` (or, if you want to confirm the unit name first: `systemctl --user list-units --type=service | grep "$(. setup/lib/install-slug.sh && systemd_unit)"`)
+  - **Manual** (no service found): restart `pnpm run dev`


 ## Diagnostics
@@ -42,7 +42,7 @@ Check remotes:
 - `git remote -v`

 If `upstream` is missing:
- Ask the user for the upstream repo URL (default: `https://github.com/qwibitai/nanoclaw.git`).
+- Ask the user for the upstream repo URL (default: `https://github.com/nanocoai/nanoclaw.git`).
 - `git remote add upstream <url>`

 Fetch:
@@ -40,7 +40,7 @@ git remote -v
 If `upstream` is missing, add it:

 ```bash
-git remote add upstream https://github.com/qwibitai/nanoclaw.git
+git remote add upstream https://github.com/nanocoai/nanoclaw.git
 ```

 ### Merge the skill branch
@@ -128,9 +128,12 @@ echo 'ANTHROPIC_API_KEY=<key>' >> .env
 pnpm run build
 ```

-Then restart the service:
- macOS: `launchctl kickstart -k gui/$(id -u)/com.nanoclaw`
- Linux: `systemctl --user restart nanoclaw`
+Then restart the service.
+
+Run from your NanoClaw project root:
+
+- macOS: `launchctl kickstart -k gui/$(id -u)/"$(. setup/lib/install-slug.sh && launchd_label)"`
+- Linux: `systemctl --user restart "$(. setup/lib/install-slug.sh && systemd_unit)"`
 - WSL/manual: stop and re-run `bash start-nanoclaw.sh`

 2. Check logs for successful proxy startup:
@@ -38,6 +38,8 @@ Before using this skill, ensure:

 ## Quick Start

+Run from your NanoClaw project root:
+
 ```bash
 # 1. Setup authentication (interactive)
 pnpm exec dotenv -e .env -- pnpm exec tsx .claude/skills/x-integration/scripts/setup.ts
@@ -49,9 +51,10 @@ pnpm exec dotenv -e .env -- pnpm exec tsx .claude/skills/x-integration/scripts/s

 # 3. Rebuild host and restart service
 pnpm run build
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
-# Linux: systemctl --user restart nanoclaw
-# Verify: launchctl list | grep nanoclaw (macOS) or systemctl --user status nanoclaw (Linux)
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+# Linux: systemctl --user restart $(systemd_unit)
+# Verify: launchctl list | grep "$(launchd_label)" (macOS) or systemctl --user status $(systemd_unit) (Linux)
 ```

 ## Configuration
@@ -270,16 +273,23 @@ cat data/x-auth.json  # Should show {"authenticated": true, ...}

 ### 4. Restart Service

+Run from your NanoClaw project root:
+
 ```bash
 pnpm run build
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
-# Linux: systemctl --user restart nanoclaw
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+# Linux: systemctl --user restart $(systemd_unit)
 ```

-**Verify success:**
+**Verify success.**
+
+Run from your NanoClaw project root:
+
 ```bash
-launchctl list | grep nanoclaw  # macOS — should show PID and exit code 0 or -
-# Linux: systemctl --user status nanoclaw
+source setup/lib/install-slug.sh
+launchctl list | grep "$(launchd_label)"  # macOS — should show PID and exit code 0 or -
+# Linux: systemctl --user status $(systemd_unit)
 ```

 ## Usage via WhatsApp
@@ -343,10 +353,13 @@ echo '{"content":"Test"}' | pnpm exec tsx .claude/skills/x-integration/scripts/p

 ### Authentication Expired

+Run from your NanoClaw project root:
+
 ```bash
 pnpm exec dotenv -e .env -- pnpm exec tsx .claude/skills/x-integration/scripts/setup.ts
-launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
-# Linux: systemctl --user restart nanoclaw
+source setup/lib/install-slug.sh
+launchctl kickstart -k gui/$(id -u)/$(launchd_label)  # macOS
+# Linux: systemctl --user restart $(systemd_unit)
 ```

 ### Browser Lock Files
@@ -7,7 +7,7 @@ on:

 jobs:
  bump-version:
-    if: github.repository == 'qwibitai/nanoclaw'
+    if: github.repository == 'nanocoai/nanoclaw'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/create-github-app-token@v1
@@ -18,12 +18,20 @@ jobs:

      - uses: actions/checkout@v4
        with:
+          fetch-depth: 0
          token: ${{ steps.app-token.outputs.token }}

      - uses: pnpm/action-setup@v4

      - name: Bump patch version
        run: |
+          # Skip the auto-bump when the pushed commits already changed the
+          # version themselves (e.g. a release PR that set a minor/major).
+          # Otherwise the bot would patch a deliberate 2.1.0 up to 2.1.1.
+          if git diff --name-only "${{ github.event.before }}" "${{ github.sha }}" | grep -qx 'package.json'; then
+            echo "package.json already changed in this push; skipping auto-bump."
+            exit 0
+          fi
          pnpm version patch --no-git-tag-version
          git add package.json
          git diff --cached --quiet && exit 0
@@ -1,7 +1,12 @@
 name: Label PR

+# SECURITY: this workflow runs with write access to the base repo on fork PRs,
+# because `pull_request_target` executes in the context of the base branch.
+# Keep it metadata-only — do NOT add actions/checkout or any step that
+# executes PR-supplied content (install scripts, build commands, etc.).
+# See https://securitylab.github.com/resources/github-actions-preventing-pwn-requests/
 on:
-  pull_request:
+  pull_request_target:
    types: [opened, edited]

 jobs:
@@ -8,7 +8,7 @@ on:

 jobs:
  update-tokens:
-    if: github.repository == 'qwibitai/nanoclaw'
+    if: github.repository == 'nanocoai/nanoclaw'
    runs-on: ubuntu-latest
    steps:
      - uses: actions/create-github-app-token@v1
@@ -1 +1,5 @@
+staged=$(git diff --cached --name-only --diff-filter=ACM -- 'src/**/*.ts')
 pnpm run format:fix
+if [ -n "$staged" ]; then
+  echo "$staged" | xargs git add
+fi
@@ -2,7 +2,45 @@

 All notable changes to NanoClaw will be documented in this file.

-For detailed release notes, see the [full changelog on the documentation site](https://docs.nanoclaw.dev/changelog).
+## [2.1.0] - 2026-06-07
+
+- [BREAKING] **Startup now requires an upgrade marker.** The host refuses to boot unless `data/upgrade-state.json` records that this install reached the current version through a sanctioned path (`/setup`, `/update-nanoclaw`, `/migrate-nanoclaw`). After this update completes — and before restarting the service — stamp the marker by running `pnpm exec tsx scripts/upgrade-state.ts set`. If the host has already tripped on restart with "update did not go through the supported path", that same command clears it. See [docs/upgrade-recovery.md](docs/upgrade-recovery.md).
+
+## [2.0.64] - 2026-05-18
+
+- **`ncl destinations add` and `remove` through the approval flow now reach the receiver immediately.** Approved destinations weren't being projected into the receiving agent's local session state, so a freshly-added destination silently failed at `send_message` with `unknown destination`, and a removed destination stayed resolvable until the next container restart. Both now take effect the moment the approval executes. Direct (non-approval) calls were unaffected.
+
+## [2.0.63] - 2026-05-15
+
+Rollup release covering v2.0.55 through v2.0.63 — everything merged since the v2.0.54 tag. Starting with this release, the goal is to publish a GitHub Release for every `package.json` version bump that lands on `main`; see [RELEASING.md](RELEASING.md).
+
+- [BREAKING] **Service names are now per-install.** On v2 installs the launchd label and systemd unit are slugged to your project root: `com.nanoclaw.<sha1(projectRoot)[:8]>` and `nanoclaw-<slug>.service`. The old `com.nanoclaw` / `nanoclaw.service` names no longer match a real service — update any copy-pasted restart or status commands. Find your install's names with `source setup/lib/install-slug.sh && launchd_label` (macOS) or `systemd_unit` (Linux). The `ncl` transport-error help text and 26 skill files now use the canonical helper-driven pattern; see [setup/lib/install-slug.sh](setup/lib/install-slug.sh).
+- **Compaction destination reminder placement fixed.** The reminder injected after SDK auto-compaction now appears at the end of the compaction summary so it isn't stripped during truncation. Replaces the placement shipped in v2.0.54.
+- **Stronger message-wrapping enforcement.** The poll loop nudges the agent when its output lacks `<message>` wrapping, and `CLAUDE.md` core instructions now require wrapping even for single-destination agents. The welcome flow no longer double-greets.
+- **OneCLI credentials after MCP install.** MCP servers added through `add_mcp_server` now inherit OneCLI gateway routing — fixes the case where the agent kept asking for API keys after installing a new server.
+- **CLI scope hardening.** `scopeField` now fails closed when scope is missing, and `sessions get` is guarded against cross-group oracle access from group-scoped agents.
+- **gmail/gcal skills aligned with v2.** `/add-gmail-tool` and `/add-gcal-tool` now reflect the v2 container-config model — DB-backed mounts, no dead `TOOL_ALLOWLIST` edits, no `container.json` writes that get clobbered on next spawn. Manual sqlite3/JSON1 invocations corrected.
+- **Repo-rename cleanup.** Remaining `qwibitai/nanoclaw` references swept to `nanocoai/nanoclaw` across code and docs; CI workflow guards updated so they no longer no-op after the rename.
+- Slack scope checklist now includes `files:read` and `files:write` for skills that read or post attachments.
+- The internal-tag description in destination instructions no longer mentions scratchpads (which confused agents into routing them incorrectly).
+- Container startup is now graceful when the `on_wake` column is missing on older sessions DBs.
+
+## [2.0.54] - 2026-05-10
+
+- **Per-group model and effort overrides.** Agent groups can now run a specific Claude model and effort level, set via `ncl groups config update --model <model> --effort <level>`. Defaults to the host-configured model when unset.
+- **Claude Code 2.1.128.** Container claude-code bumped from 2.1.116 to 2.1.128.
+- CLI help text improvements for `ncl groups config` and `ncl groups restart`.
+
+## [2.0.48] - 2026-05-09
+
+- **Container config moved to DB.** Per-agent-group container runtime config (provider, model, packages, MCP servers, mounts, skills) now lives in the `container_configs` table instead of `groups/<folder>/container.json`. Existing filesystem configs are backfilled automatically on startup. Managed via `ncl groups config get/update` and `config add-mcp-server/remove-mcp-server/add-package/remove-package`.
+- **Explicit restart with on-wake messages.** Config CLI operations no longer auto-kill containers. New `ncl groups restart` command with `--rebuild` and `--message` flags. On-wake messages (`on_wake` column on `messages_in`) are only picked up by a fresh container's first poll, preventing dying containers from stealing them during the SIGTERM grace period. Self-mod approval handlers (`install_packages`, `add_mcp_server`) use the same race-free mechanism.
+- **Per-group CLI scope.** New `cli_scope` setting on container config (`disabled` / `group` / `global`, default `group`). Controls what the agent can access via `ncl` from inside the container. `disabled` excludes CLI instructions from CLAUDE.md and blocks all requests. `group` (default) restricts to own-group resources with auto-filled args. `global` gives unrestricted access (set automatically for owner agent groups). Includes post-handler result filtering to prevent cross-group data leaks and blocks `cli_scope` escalation from group-scoped agents.
+
+## [2.0.45] - 2026-05-08
+
+- **Admin CLI (`ncl`).** New `ncl` command for querying and modifying the central DB — agent groups, messaging groups, wirings, users, roles, members, destinations, sessions, approvals, and dropped messages. Host-side transport via Unix socket; container-side transport via session DB. Write operations from inside containers go through the approval flow. `list` supports column filtering and `--limit`. Run `ncl help` for usage.
+- **v1 → v2 migration.** Run `bash migrate-v2.sh` from the v2 checkout. Finds your v1 install (sibling directory or `NANOCLAW_V1_PATH`), merges `.env`, seeds the v2 DB from `registered_groups`, copies group folders (`CLAUDE.md` → `CLAUDE.local.md`), copies session data with conversation continuity, ports scheduled tasks, interactively selects and installs channels (clack multiselect), copies container skills, builds the agent container, and offers a service switchover to test. Hands off to Claude (`/migrate-from-v1`) for owner seeding, access policy, CLAUDE.md cleanup, and fork customization porting. See [docs/migration-dev.md](docs/migration-dev.md) and [docs/v1-to-v2-changes.md](docs/v1-to-v2-changes.md).

 ## [2.0.0] - 2026-04-22

@@ -6,8 +6,8 @@ This is NanoClaw v2, a ground-up rewrite with breaking changes throughout. It ca

 **Do this instead:**
 1. Abort the merge: `git merge --abort` (or `git reset --hard ORIG_HEAD` if the merge already completed).
-2. Tell the user, verbatim: *"This is the v2 rewrite — it can't be merged into your existing install. Run `bash migrate-v2.sh` to migrate instead."*
-3. Wait for the user to confirm before doing anything else. Do not attempt the migration yourself.
+2. Tell the user, verbatim: *"This is the v2 rewrite — it can't be merged into your existing install. Exit Claude Code (or open a separate terminal) and run `bash migrate-v2.sh` from the shell."*
+3. Wait for the user to confirm before doing anything else. Do not run the migration script yourself — it requires an interactive terminal and cannot be run from within Claude Code.

 If you are a fresh install (you ran `git clone`, not `git pull`) and there are no conflicts, ignore this banner and continue below.

@@ -53,6 +53,8 @@ Exactly one writer per file — no cross-mount lock contention. Heartbeat is a f

 `data/v2.db` holds everything that isn't per-session: users, user_roles, agent_groups, messaging_groups, wiring, pending_approvals, user_dms, chat_sdk_* (for the Chat SDK bridge), schema_version. Migrations live at `src/db/migrations/`.

+For ad-hoc queries from skills or scripts, use the in-tree wrapper rather than the `sqlite3` CLI: `pnpm exec tsx scripts/q.ts <db> "<sql>"`. The host setup intentionally avoids depending on the `sqlite3` binary (`setup/verify.ts:5`); the wrapper goes through the `better-sqlite3` dep that setup already installs and verifies. Default-output format matches `sqlite3 -list` (pipe-separated, no header) so existing skill text reads identically.
+
 ## Key Files

 | File | Purpose |
@@ -70,13 +72,43 @@ Exactly one writer per file — no cross-mount lock contention. Heartbeat is a f
 | `src/onecli-approvals.ts` | OneCLI credentialed-action approval bridge |
 | `src/user-dm.ts` | Cold-DM resolution + `user_dms` cache |
 | `src/group-init.ts` | Per-agent-group filesystem scaffold (CLAUDE.md, skills, agent-runner-src overlay) |
-| `src/db/` | DB layer — agent_groups, messaging_groups, sessions, user_roles, user_dms, pending_*, migrations |
+| `src/db/container-configs.ts` | CRUD for `container_configs` table (per-group container runtime config) |
+| `src/backfill-container-configs.ts` | Migrates legacy `container.json` files into the DB on startup |
+| `src/container-restart.ts` | Kill + on-wake respawn for agent group containers |
+| `src/db/` | DB layer — agent_groups, messaging_groups, sessions, container_configs, user_roles, user_dms, pending_*, migrations |
 | `src/channels/` | Channel adapter infra (registry, Chat SDK bridge); specific channel adapters are skill-installed from the `channels` branch |
 | `src/providers/` | Host-side provider container-config (`claude` baked in; `opencode` etc. installed from the `providers` branch) |
 | `container/agent-runner/src/` | Agent-runner: poll loop, formatter, provider abstraction, MCP tools, destinations |
-| `container/skills/` | Container skills mounted into every agent session |
+| `container/skills/` | Container skills mounted into every agent session (`onecli-gateway`, `welcome`, `self-customize`, `agent-browser`, `slack-formatting`) |
 | `groups/<folder>/` | Per-agent-group filesystem (CLAUDE.md, skills, per-group `agent-runner-src/` overlay) |
 | `scripts/init-first-agent.ts` | Bootstrap the first DM-wired agent (used by `/init-first-agent` skill) |
+| `migrate-v2.sh` + `setup/migrate-v2/` | v1→v2 migration. Standalone script: `bash migrate-v2.sh`. Seeds DB, copies groups/sessions, installs channels, builds container, offers service switchover, then hands off to `/migrate-from-v1` skill for owner setup and CLAUDE.md cleanup. See [docs/migration-dev.md](docs/migration-dev.md). |
+
+## Admin CLI (`ncl`)
+
+`ncl` queries and modifies the central DB — agent groups, messaging groups, wirings, users, roles, and more. On the host it connects via Unix socket (`src/cli/socket-server.ts`); inside containers it uses the session DB transport (`container/agent-runner/src/cli/ncl.ts`).
+
+```
+ncl <resource> <verb> [<id>] [--flags]
+ncl <resource> help
+ncl help
+```
+
+| Resource | Verbs | What it is |
+|----------|-------|------------|
+| groups | list, get, create, update, delete, restart, config get/update, config add-mcp-server/remove-mcp-server, config add-package/remove-package | Agent groups (workspace, personality, container config) |
+| messaging-groups | list, get, create, update, delete | A single chat/channel on one platform |
+| wirings | list, get, create, update, delete | Links a messaging group to an agent group (session mode, triggers) |
+| users | list, get, create, update | Platform identities (`<channel>:<handle>`) |
+| roles | list, grant, revoke | Owner / admin privileges (global or scoped to an agent group) |
+| members | list, add, remove | Unprivileged access gate for an agent group |
+| destinations | list, add, remove | Where an agent group can send messages |
+| sessions | list, get | Active sessions (read-only) |
+| user-dms | list | Cold-DM cache (read-only) |
+| dropped-messages | list | Messages from unregistered senders (read-only) |
+| approvals | list, get | Pending approval requests (read-only) |
+
+Key files: `src/cli/dispatch.ts` (dispatcher + approval handler), `src/cli/crud.ts` (generic CRUD registration), `src/cli/resources/` (per-resource definitions).

 ## Channels and Providers (skill-installed)

@@ -91,39 +123,47 @@ Each `/add-<name>` skill is idempotent: `git fetch origin <branch>` → copy mod

 One tier of agent self-modification today:

-1. **`install_packages` / `add_mcp_server`** — changes to the per-agent-group container config only (apt/npm deps, wire an existing MCP server). Single admin approval per request; on approve, the handler in `src/modules/self-mod/apply.ts` rebuilds the image when needed (`install_packages` only) and restarts the container. `container/agent-runner/src/mcp-tools/self-mod.ts`.
+1. **`install_packages` / `add_mcp_server`** — changes to the per-agent-group container config in the DB (apt/npm deps, wire an existing MCP server). Single admin approval per request; on approve, the handler in `src/modules/self-mod/apply.ts` rebuilds the image when needed (`install_packages` only), writes an `on_wake` message, kills the container, and respawns via `onExit` callback. The on-wake message is only picked up by the fresh container's first poll — dying containers can never steal it. `container/agent-runner/src/mcp-tools/self-mod.ts`.

 A second tier (direct source-level self-edits via a draft/activate flow) is planned but not yet implemented.

+## Container Config
+
+Per-agent-group container runtime config (provider, model, packages, MCP servers, mounts, etc.) lives in the `container_configs` table in the central DB. Materialized to `groups/<folder>/container.json` at spawn time so the container runner can read it. Managed via `ncl groups config get/update` and the self-mod MCP tools.
+
+**`cli_scope`** — controls what the agent can do with `ncl` from inside the container:
+
+| Value | Behavior |
+|-------|----------|
+| `disabled` | Agent never learns about ncl (instructions excluded from CLAUDE.md). Host dispatch rejects any `cli_request`. |
+| `group` (default) | Agent can access `groups`, `sessions`, `destinations`, `members` only, scoped to its own agent group. `--id` and group args are auto-filled. Cross-group access rejected. `cli_scope` changes blocked. |
+| `global` | Unrestricted. Set automatically for owner agent groups via `init-first-agent`. |
+
+Key files: `src/db/container-configs.ts`, `src/container-config.ts`, `src/cli/dispatch.ts` (scope enforcement), `src/claude-md-compose.ts` (instructions exclusion).
+
+## Container Restart
+
+`ncl groups restart --id <group-id> [--rebuild] [--message <text>]`. Kills running containers; if `--message` is provided, writes an `on_wake` message and respawns via `onExit` callback. Without `--message`, containers come back on the next user message. From inside a container, `--id` is auto-filled and only the calling session is restarted.
+
+The `on_wake` column on `messages_in` ensures wake messages are only picked up by a fresh container's first poll iteration. This prevents the race where a dying container (still in its SIGTERM grace period) could steal the message. `killContainer` accepts an optional `onExit` callback that fires after the process exits, guaranteeing the old container is gone before the new one spawns.
+
+Key files: `src/container-restart.ts`, `src/container-runner.ts` (`killContainer`), `container/agent-runner/src/db/messages-in.ts` (`getPendingMessages`).
+
 ## Secrets / Credentials / OneCLI

-API keys, OAuth tokens, and auth credentials are managed by the OneCLI gateway. Secrets are injected into per-agent containers at request time — none are passed in env vars or through chat context. `src/onecli-approvals.ts`, `ensureAgent()` in `container-runner.ts`. Run `onecli --help`.
+API keys, OAuth tokens, and auth credentials are managed by the OneCLI gateway. Secrets are injected into per-agent containers at request time — none are passed in env vars or through chat context. The container agent sees this via the `onecli-gateway` container skill (`container/skills/onecli-gateway/SKILL.md`), which teaches it how the proxy works, how to handle auth errors, and to never ask for raw credentials. Host-side wiring: `src/onecli-approvals.ts`, `ensureAgent()` in `container-runner.ts`. Run `onecli --help`.

-### Gotcha: auto-created agents start in `selective` secret mode
+### Secret modes

-When the host first spawns a session for a new agent group, `container-runner.ts:385` calls `onecli.ensureAgent({ name, identifier })`. The OneCLI `POST /api/agents` endpoint creates the agent in **`selective`** secret mode — meaning **no secrets are assigned to it by default**, even if the secrets exist in the vault and have host patterns that would otherwise match.
-
-Symptom: container starts, the proxy + CA cert are wired correctly, but the agent gets `401 Unauthorized` (or similar) from APIs whose credentials *are* in the vault. The credential just isn't in this agent's allow-list.
-
-The SDK does not expose `setSecretMode` — the only fix is the CLI (or the web UI at `http://127.0.0.1:10254`).
+Auto-created agents default to `all` secret mode — every vault secret whose host pattern matches is injected automatically, so the common case needs no per-agent setup. If an agent is in `selective` mode it gets no secrets until you assign them, which shows up as a `401` from an API whose credential *is* in the vault. The SDK can't change this; use the CLI (or the web UI at `http://127.0.0.1:10254`):

 ```bash
-# Find the agent (identifier is the agent group id)
-onecli agents list
-
-# Flip to "all" so every vault secret with a matching host pattern gets injected
-onecli agents set-secret-mode --id <agent-id> --mode all
-
-# Or, stay selective and assign specific secrets
-onecli secrets list                                    # find secret ids
-onecli agents set-secrets --id <agent-id> --secret-ids <id1>,<id2>
-
-# Inspect what an agent currently has
-onecli agents secrets --id <agent-id>                  # secrets assigned to this agent
-onecli secrets list                                    # all vault secrets (with host patterns)
+onecli agents list                                          # check secretMode
+onecli agents set-secret-mode --id <agent-id> --mode all    # inject all matching secrets
+onecli agents set-secrets --id <agent-id> --secret-ids ...  # or stay selective, assign specific ones
 ```

-If you've just enabled `mode all`, no container restart is needed — the gateway looks up secrets per request, so the next API call from the running container will see the new credentials.
+No container restart needed — the gateway looks up secrets per request.

 ### Requiring approval for credential use

@@ -141,7 +181,7 @@ Four types of skills. See [CONTRIBUTING.md](CONTRIBUTING.md) for the full taxono
 - **Channel/provider install skills** — copy the relevant module(s) in from the `channels` or `providers` branch, wire imports, install pinned deps (e.g. `/add-discord`, `/add-slack`, `/add-whatsapp`, `/add-opencode`).
 - **Utility skills** — ship code files alongside `SKILL.md` (e.g. `/claw`).
 - **Operational skills** — instruction-only workflows (`/setup`, `/debug`, `/customize`, `/init-first-agent`, `/manage-channels`, `/init-onecli`, `/update-nanoclaw`).
- **Container skills** — loaded inside agent containers at runtime (`container/skills/`: `welcome`, `self-customize`, `agent-browser`, `slack-formatting`).
+- **Container skills** — loaded inside agent containers at runtime (`container/skills/`: `onecli-gateway`, `welcome`, `self-customize`, `agent-browser`, `slack-formatting`).

 | Skill | When to Use |
 |-------|-------------|
@@ -157,6 +197,17 @@ Four types of skills. See [CONTRIBUTING.md](CONTRIBUTING.md) for the full taxono

 Before creating a PR, adding a skill, or preparing any contribution, you MUST read [CONTRIBUTING.md](CONTRIBUTING.md). It covers accepted change types, the four skill types and their guidelines, `SKILL.md` format rules, and the pre-submission checklist.

+## PR Hygiene
+
+Before creating a PR, run these checks:
+
+```bash
+git diff upstream/main --stat HEAD
+git log upstream/main..HEAD --oneline
+```
+
+Show the output and wait for approval. Installation-specific files (group files, .claude/settings.json, local configs) should not be included.
+
 ## Development

 Run commands directly — don't tell the user to run them.
@@ -186,7 +237,17 @@ launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # restart
 systemctl --user start|stop|restart nanoclaw
 ```

-Host logs: `logs/nanoclaw.log` (normal) and `logs/nanoclaw.error.log` (errors only — some delivery/approval failures only show up here).
+## Troubleshooting
+
+Check these first when something goes wrong:
+
+| What | Where |
+|------|-------|
+| Host logs | `logs/nanoclaw.error.log` first (delivery failures, crash-loop backoff, warnings), then `logs/nanoclaw.log` for the full routing chain |
+| Setup logs | `logs/setup.log` (overall), `logs/setup-steps/*.log` (per-step: bootstrap, environment, container, onecli, mounts, service, etc.) |
+| Session DBs | `data/v2-sessions/<agent-group>/<session>/` — `inbound.db` (`messages_in`: did the message reach the container?), `outbound.db` (`messages_out`: did the agent produce a response?) |
+
+Note: container logs are lost after the container exits (`--rm` flag). If the agent silently failed inside the container, there's no persistent log to inspect.

 ## Supply Chain Security (pnpm)

@@ -211,6 +272,8 @@ This project uses pnpm with `minimumReleaseAge: 4320` (3 days) in `pnpm-workspac
 | [docs/setup-wiring.md](docs/setup-wiring.md) | What's wired, what's open in the setup flow |
 | [docs/architecture-diagram.md](docs/architecture-diagram.md) | Diagram version of the architecture |
 | [docs/build-and-runtime.md](docs/build-and-runtime.md) | Runtime split (Node host + Bun container), lockfiles, image build surface, CI, key invariants |
+| [docs/v1-to-v2-changes.md](docs/v1-to-v2-changes.md) | v1→v2 architecture diff — vocabulary for where v1 things moved |
+| [docs/migration-dev.md](docs/migration-dev.md) | Migration development guide — testing, debugging, dev loop |

 ## Container Build Cache

@@ -4,8 +4,8 @@

 1. **Check for existing work.** Search open PRs and issues before starting:
   ```bash
-   gh pr list --repo qwibitai/nanoclaw --search "<your feature>"
-   gh issue list --repo qwibitai/nanoclaw --search "<your feature>"
+   gh pr list --repo nanocoai/nanoclaw --search "<your feature>"
+   gh issue list --repo nanocoai/nanoclaw --search "<your feature>"
   ```
   If a related PR or issue exists, build on it rather than duplicating effort.

@@ -43,7 +43,7 @@ Add capabilities to NanoClaw by merging a git branch. The SKILL.md contains setu
 3. Claude walks through interactive setup (env vars, bot creation, etc.)

 **Contributing a feature skill:**
-1. Fork `qwibitai/nanoclaw` and branch from `main`
+1. Fork `nanocoai/nanoclaw` and branch from `main`
 2. Make the code changes (new files, modified source, updated `package.json`, etc.)
 3. Add a SKILL.md in `.claude/skills/<name>/` with setup instructions — step 1 should be merging the branch
 4. Open a PR. We'll create the `skill/<name>` branch from your work
@@ -123,7 +123,8 @@ Test your contribution on a fresh clone before submitting. For skills, run the s

 1. **Link related issues.** If your PR resolves an open issue, include `Closes #123` in the description so it's auto-closed on merge.
 2. **Test thoroughly.** Run the feature yourself. For skills, test on a fresh clone.
-3. **Check the right box** in the PR template. Labels are auto-applied based on your selection:
+3. **Check for installation-specific files.** Before creating a PR, verify no installation-specific files are in your diff (see PR Hygiene in CLAUDE.md).
+4. **Check the right box** in the PR template. Labels are auto-applied based on your selection:

 | Checkbox | Label |
 |----------|-------|
@@ -16,6 +16,7 @@ Thanks to everyone who has contributed to NanoClaw!
 - [flobo3](https://github.com/flobo3) — Flo
 - [edwinwzhe](https://github.com/edwinwzhe) — Edwin He
 - [scottgl9](https://github.com/scottgl9) — Scott Glover
+- [ingyukoh](https://github.com/ingyukoh) — Ingyu Koh
 - [cschmidt](https://github.com/cschmidt) — Carl Schmidt
 - [leonalfredbot-ship-it](https://github.com/leonalfredbot-ship-it) — Alfred-the-buttler
 - [moktamd](https://github.com/moktamd)
@@ -26,11 +26,36 @@ NanoClaw provides that same core functionality, but in a codebase small enough t
 ## Quick Start

 ```bash
-git clone https://github.com/qwibitai/nanoclaw.git && cd nanoclaw && bash nanoclaw.sh
+git clone https://github.com/nanocoai/nanoclaw.git nanoclaw-v2
+cd nanoclaw-v2
+bash nanoclaw.sh
 ```

 `nanoclaw.sh` walks you from a fresh machine to a named agent you can message. It installs Node, pnpm, and Docker if missing, registers your Anthropic credential with OneCLI, builds the agent container, and pairs your first channel (Telegram, Discord, WhatsApp, or a local CLI). If a step fails, Claude Code is invoked automatically to diagnose and resume from where it broke.

+<details>
+<summary><strong>Migrating from NanoClaw v1?</strong></summary>
+
+Run from a fresh v2 checkout next to your v1 install:
+
+```bash
+git clone https://github.com/nanocoai/nanoclaw.git nanoclaw-v2
+cd nanoclaw-v2
+bash migrate-v2.sh
+```
+
+`migrate-v2.sh` finds your v1 install (sibling directory, or `NANOCLAW_V1_PATH=/path/to/nanoclaw`), migrates state into the v2 checkout, then `exec`s into Claude Code to finish the parts that need judgment (owner seeding, CLAUDE.local.md cleanup, fork-customisation replay).
+
+Run the script directly, not from inside a Claude session — the deterministic side needs interactive prompts and real shell I/O for Node/pnpm bootstrap, Docker, OneCLI, and the container build.
+
+**What it does:** merges `.env`, seeds the v2 DB from `registered_groups`, copies group folders + session data + scheduled tasks, installs the channel adapters you select, copies channel auth state (including Baileys keystore + LID mappings for WhatsApp), builds the agent container.
+
+**What it doesn't:** flip the system service. Pick *"switch to v2"* at the prompt, or do it manually after testing — your v1 install is left untouched.
+
+See [docs/v1-to-v2-changes.md](docs/v1-to-v2-changes.md) for what's different and [docs/migration-dev.md](docs/migration-dev.md) for development notes.
+
+</details>
+
 ## Philosophy

 **Small enough to understand.** One process, a few source files and no microservices. If you want to understand the full NanoClaw codebase, just ask Claude Code to walk you through it.
@@ -190,3 +215,5 @@ See [CHANGELOG.md](CHANGELOG.md) for breaking changes, or the [full release hist
 ## License

 MIT
+
+<img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=47894bd5-353b-42fe-bb97-74144e6df0bf" />
@@ -8,92 +8,56 @@

 <p align="center">
  <a href="https://nanoclaw.dev">nanoclaw.dev</a>&nbsp; • &nbsp;
+  <a href="https://docs.nanoclaw.dev">ドキュメント</a>&nbsp; • &nbsp;
  <a href="README.md">English</a>&nbsp; • &nbsp;
  <a href="README_zh.md">中文</a>&nbsp; • &nbsp;
  <a href="https://discord.gg/VDdww8qS42"><img src="https://img.shields.io/discord/1470188214710046894?label=Discord&logo=discord&v=2" alt="Discord" valign="middle"></a>&nbsp; • &nbsp;
-  <a href="repo-tokens"><img src="repo-tokens/badge.svg" alt="34.9k tokens, 17% of context window" valign="middle"></a>
+  <a href="repo-tokens"><img src="repo-tokens/badge.svg" alt="repo tokens" valign="middle"></a>
 </p>

-> **注意:** この日本語訳は v1 時点のもので、最新の v2 アーキテクチャは反映されていません。最新の内容は [README.md](README.md) をご覧ください。
-
---
-
-<h2 align="center">🐳 Dockerサンドボックスで動作</h2>
-<p align="center">各エージェントはマイクロVM内の独立したコンテナで実行されます。<br>ハイパーバイザーレベルの分離。ミリ秒で起動。複雑なセットアップ不要。</p>
-
-**macOS (Apple Silicon)**
-```bash
-curl -fsSL https://nanoclaw.dev/install-docker-sandboxes.sh | bash
-```
-
-**Windows (WSL)**
-```bash
-curl -fsSL https://nanoclaw.dev/install-docker-sandboxes-windows.sh | bash
-```
-
-> 現在、macOS（Apple Silicon）とWindows（x86）に対応しています。Linux対応は近日公開予定。
-
-<p align="center"><a href="https://nanoclaw.dev/blog/nanoclaw-docker-sandboxes">発表記事を読む →</a>&nbsp; · &nbsp;<a href="docs/docker-sandboxes.md">手動セットアップガイド →</a></p>
-
 ---

 ## NanoClawを作った理由

-[OpenClaw](https://github.com/openclaw/openclaw)は素晴らしいプロジェクトですが、理解しきれない複雑なソフトウェアに自分の生活へのフルアクセスを与えたまま安心して眠れるとは思えませんでした。OpenClawは約50万行のコード、53の設定ファイル、70以上の依存関係を持っています。セキュリティはアプリケーションレベル（許可リスト、ペアリングコード）であり、真のOS レベルの分離ではありません。すべてが共有メモリを持つ1つのNodeプロセスで動作します。
+[OpenClaw](https://github.com/openclaw/openclaw)は素晴らしいプロジェクトですが、自分が理解しきれない複雑なソフトウェアに生活へのフルアクセスを与えたまま安心して眠れるとは思えませんでした。OpenClawは約50万行のコード、53の設定ファイル、70以上の依存関係を持っています。セキュリティはアプリケーションレベル（許可リスト、ペアリングコード）であり、真のOSレベルの分離ではありません。すべてが共有メモリを持つ1つのNodeプロセスで動作します。

-NanoClawは同じコア機能を提供しますが、理解できる規模のコードベースで実現しています：1つのプロセスと少数のファイル。Claudeエージェントは単なるパーミッションチェックの背後ではなく、ファイルシステム分離された独自のLinuxコンテナで実行されます。
+NanoClawは同じコア機能を提供しますが、理解できる規模のコードベースで実現しています。1つのプロセスと少数のファイル。Claudeエージェントは単なるパーミッションチェックの背後ではなく、ファイルシステム分離された独自のLinuxコンテナで実行されます。

 ## クイックスタート

 ```bash
-gh repo fork qwibitai/nanoclaw --clone
-cd nanoclaw
-claude
+git clone https://github.com/nanocoai/nanoclaw.git nanoclaw-v2
+cd nanoclaw-v2
+bash nanoclaw.sh
 ```

-<details>
-<summary>GitHub CLIなしの場合</summary>
-
-1. GitHub上で[qwibitai/nanoclaw](https://github.com/qwibitai/nanoclaw)をフォーク（Forkボタンをクリック）
-2. `git clone https://github.com/<あなたのユーザー名>/nanoclaw.git`
-3. `cd nanoclaw`
-4. `claude`
-
-</details>
-
-その後、`/setup`を実行します。Claude Codeがすべてを処理します：依存関係、認証、コンテナセットアップ、サービス設定。
-
-> **注意:** `/`で始まるコマンド（`/setup`、`/add-whatsapp`など）は[Claude Codeスキル](https://code.claude.com/docs/en/skills)です。通常のターミナルではなく、`claude` CLIプロンプト内で入力してください。Claude Codeをインストールしていない場合は、[claude.com/product/claude-code](https://claude.com/product/claude-code)から入手してください。
+`nanoclaw.sh`は、まっさらなマシンから、メッセージを送れる名前付きエージェントが動く状態までを一気通貫で案内します。NodeやpnpmやDockerが無ければインストールし、AnthropicクレデンシャルをOneCLIに登録し、エージェントコンテナをビルドし、最初のチャネル（Telegram、Discord、WhatsApp、またはローカルCLI）とペアリングします。途中でステップが失敗すれば自動的にClaude Codeが呼び出され、原因を診断して中断箇所から再開します。

 ## 設計思想

-**理解できる規模。** 1つのプロセス、少数のソースファイル、マイクロサービスなし。NanoClawのコードベース全体を理解したい場合は、Claude Codeに説明を求めるだけです。
+**理解できる規模。** 1つのプロセス、少数のソースファイル、マイクロサービスなし。NanoClawのコードベース全体を把握したいなら、Claude Codeに説明を求めれば十分です。

-**分離によるセキュリティ。** エージェントはLinuxコンテナ（macOSではApple Container、またはDocker）で実行され、明示的にマウントされたものだけが見えます。コマンドはホストではなくコンテナ内で実行されるため、Bashアクセスは安全です。
+**分離によるセキュリティ。** エージェントはLinuxコンテナで実行され、明示的にマウントされたものだけが見えます。コマンドはホストではなくコンテナ内で実行されるため、Bashアクセスも安全です。

-**個人ユーザー向け。** NanoClawはモノリシックなフレームワークではなく、各ユーザーのニーズに正確にフィットするソフトウェアです。肥大化するのではなく、オーダーメイドになるよう設計されています。自分のフォークを作成し、Claude Codeにニーズに合わせて変更させます。
+**個人ユーザー向け。** NanoClawはモノリシックなフレームワークではなく、各ユーザーのニーズに正確にフィットするソフトウェアです。肥大化するのではなく、オーダーメイドであるよう設計されています。自分のフォークを作り、Claude Codeにニーズに合わせて変更させます。

-**カスタマイズ＝コード変更。** 設定ファイルの肥大化なし。動作を変えたい？コードを変更するだけ。コードベースは変更しても安全な規模です。
+**カスタマイズ＝コード変更。** 設定の肥大化はありません。動作を変えたいならコードを変える。コードベースは変更しても安全な規模です。

-**AIネイティブ。**
- インストールウィザードなし — Claude Codeがセットアップを案内。
- モニタリングダッシュボードなし — Claudeに状況を聞くだけ。
- デバッグツールなし — 問題を説明すればClaudeが修正。
+**AIネイティブ、設計としてハイブリッド。** インストールとオンボーディングは最適化されたスクリプトのパスで、速く決定的です。判断が必要なところ（インストール失敗、対話的な決定、カスタマイズ）では、制御はシームレスにClaude Codeへ渡されます。セットアップ以降も、監視ダッシュボードやデバッグUIは用意しません。問題をチャットで説明すれば、Claude Codeが処理します。

-**機能追加ではなくスキル。** コードベースに機能（例：Telegram対応）を追加する代わりに、コントリビューターは`/add-telegram`のような[Claude Codeスキル](https://code.claude.com/docs/en/skills)を提出し、あなたのフォークを変換します。あなたが必要なものだけを正確に実行するクリーンなコードが手に入ります。
+**機能ではなくスキル。** トランクにはレジストリとインフラのみを同梱し、個別のチャネルアダプターや代替プロバイダーは含めません。チャネル（Discord、Slack、Telegram、WhatsAppなど）は長期運用される`channels`ブランチに、代替プロバイダー（OpenCode、Ollama）は`providers`ブランチに置かれます。`/add-telegram`や`/add-opencode`などを実行すると、スキルが必要なモジュールだけを正確にフォークへコピーします。要求していない機能は一切入りません。

-**最高のハーネス、最高のモデル。** NanoClawはClaude Agent SDK上で動作します。つまり、Claude Codeを直接実行しているということです。Claude Codeは高い能力を持ち、そのコーディングと問題解決能力によってNanoClawを変更・拡張し、各ユーザーに合わせてカスタマイズできます。
+**最高のハーネス、最高のモデル。** NanoClawはAnthropic公式のClaude Agent SDK経由でネイティブにClaude Codeを使用します。最新のClaudeモデルとClaude Codeの全ツールセット（自分のNanoClawフォークを変更・拡張する能力を含む）が手に入ります。他プロバイダーはドロップイン・オプションです。OpenAIのCodex（ChatGPTサブスクリプションまたはAPIキー）向けには`/add-codex`、OpenCode経由のOpenRouter、Google、DeepSeekなどには`/add-opencode`、ローカルのオープンウェイトモデルには`/add-ollama-provider`。プロバイダーはエージェントグループごとに設定可能です。

 ## サポート機能

- **マルチチャネルメッセージング** - WhatsApp、Telegram、Discord、Slack、Gmailからアシスタントと会話。`/add-whatsapp`や`/add-telegram`などのスキルでチャネルを追加。1つでも複数でも同時に実行可能。
- **グループごとの分離コンテキスト** - 各グループは独自の`CLAUDE.md`メモリ、分離されたファイルシステムを持ち、そのファイルシステムのみがマウントされた専用コンテナサンドボックスで実行。
- **メインチャネル** - 管理制御用のプライベートチャネル（セルフチャット）。各グループは完全に分離。
- **スケジュールタスク** - Claudeを実行し、メッセージを返せる定期ジョブ。
- **Webアクセス** - Webからのコンテンツ検索・取得。
- **コンテナ分離** - エージェントは[Dockerサンドボックス](https://nanoclaw.dev/blog/nanoclaw-docker-sandboxes)（マイクロVM分離）、Apple Container（macOS）、またはDocker（macOS/Linux）でサンドボックス化。
- **エージェントスウォーム** - 複雑なタスクで協力する専門エージェントチームを起動。
- **オプション連携** - Gmail（`/add-gmail`）などをスキルで追加。
+- **マルチチャネルメッセージング** — WhatsApp、Telegram、Discord、Slack、Microsoft Teams、iMessage、Matrix、Google Chat、Webex、Linear、GitHub、WeChat、Resend経由のメール。`/add-<channel>`スキルでオンデマンドにインストール。1つでも複数でも同時に実行可能。
+- **柔軟な分離モデル** — チャネルごとに専用エージェントを割り当てて完全プライバシーを確保することも、複数チャネルで1つのエージェントを共有して会話は分離しつつメモリを統一することも、複数チャネルを1つの共有セッションにまとめて会話を横断させることもできます。`/manage-channels`でチャネル単位に選択。[docs/isolation-model.md](docs/isolation-model.md)参照。
+- **エージェントごとのワークスペース** — 各エージェントグループは独自の`CLAUDE.md`、独自のメモリ、独自のコンテナ、そしてあなたが許可したマウントのみを持ちます。明示的に配線しない限り、境界を越えるものはありません。
+- **スケジュールタスク** — Claudeを実行し、結果を返信できる定期ジョブ。
+- **Webアクセス** — Webからの検索とコンテンツ取得。
+- **コンテナ分離** — エージェントはDockerでサンドボックス化されます（macOS/Linux/WSL2）。[Docker Sandboxes](docs/docker-sandboxes.md)によるマイクロVM分離や、macOSネイティブのオプトインとしてApple Containerも選択可能です。
+- **クレデンシャルのセキュリティ** — エージェントは生のAPIキーを保持しません。アウトバウンドリクエストは[OneCLI Agent Vault](https://github.com/onecli/onecli)を経由し、リクエスト時に認証情報を注入して、エージェントごとのポリシーとレート制限を適用します。

 ## 使い方

@@ -105,7 +69,7 @@ claude
@Andy 毎週月曜の朝8時に、Hacker NewsとTechCrunchからAI関連のニュースをまとめてブリーフィングを送って
 ```

-メインチャネル（セルフチャット）から、グループやタスクを管理できます：
+所有または管理しているチャネルからは、グループやタスクを管理できます：
 ```
@Andy 全グループのスケジュールタスクを一覧表示して
@Andy 月曜のブリーフィングタスクを一時停止して
@@ -114,14 +78,14 @@ claude

 ## カスタマイズ

-NanoClawは設定ファイルを使いません。変更するには、Claude Codeに伝えるだけです：
+NanoClawは設定ファイルを使いません。変更したいときは、Claude Codeにやりたいことを伝えるだけです：

 - 「トリガーワードを@Bobに変更して」
 - 「今後はレスポンスをもっと短く直接的にして」
 - 「おはようと言ったらカスタム挨拶を追加して」
 - 「会話の要約を毎週保存して」

-または`/customize`を実行してガイド付きの変更を行えます。
+または`/customize`を実行すればガイド付きで変更できます。

 コードベースは十分に小さいため、Claudeが安全に変更できます。

@@ -129,105 +93,101 @@ NanoClawは設定ファイルを使いません。変更するには、Claude Co

 **機能を追加するのではなく、スキルを追加してください。**

-Telegram対応を追加したい場合、コアコードベースにTelegramを追加するPRを作成しないでください。代わりに、NanoClawをフォークし、ブランチでコード変更を行い、PRを開いてください。あなたのPRから`skill/telegram`ブランチを作成し、他のユーザーが自分のフォークにマージできるようにします。
+新しいチャネルやエージェントプロバイダーを追加したい場合、トランクには追加しないでください。新しいチャネルアダプターは`channels`ブランチに、新しいエージェントプロバイダーは`providers`ブランチに追加します。ユーザーはそれぞれのフォークで`/add-<name>`スキルを実行し、スキルが必要なモジュールを標準パスへコピーし、登録を配線し、依存関係をピン留めします。

-ユーザーは自分のフォークで`/add-telegram`を実行するだけで、あらゆるユースケースに対応しようとする肥大化したシステムではなく、必要なものだけを正確に実行するクリーンなコードが手に入ります。
+こうすることでトランクは純粋なレジストリ／インフラのまま保たれ、どのフォークもスリムなままです。ユーザーは求めたチャネルとプロバイダーだけを受け取り、それ以外は入りません。

 ### RFS（スキル募集）

-私たちが求めているスキル：
+私たちが見たいスキル：

 **コミュニケーションチャネル**
- `/add-signal` - Signalをチャネルとして追加
-
-**セッション管理**
- `/clear` - 会話をコンパクト化する`/clear`コマンドの追加（同一セッション内で重要な情報を保持しながらコンテキストを要約）。Claude Agent SDKを通じてプログラム的にコンパクト化をトリガーする方法の解明が必要。
+- `/add-signal` — Signalをチャネルとして追加

 ## 必要条件

- macOSまたはLinux
- Node.js 20以上
- [Claude Code](https://claude.ai/download)
- [Apple Container](https://github.com/apple/container)（macOS）または[Docker](https://docker.com/products/docker-desktop)（macOS/Linux）
+- macOSまたはLinux（WindowsはWSL2経由）
+- Node.js 20以上とpnpm 10以上（インストーラーが未インストールなら両方をインストールします）
+- [Docker Desktop](https://docker.com/products/docker-desktop)（macOS/Windows）または Docker Engine（Linux）
+- [Claude Code](https://claude.ai/download)（`/customize`、`/debug`、セットアップ時のエラー復旧、全ての`/add-<channel>`スキルで使用）

 ## アーキテクチャ

 ```
-チャネル --> SQLite --> ポーリングループ --> コンテナ（Claude Agent SDK） --> レスポンス
+メッセージングアプリ → ホストプロセス（ルーター） → inbound.db → コンテナ（Bun、Claude Agent SDK） → outbound.db → ホストプロセス（配信） → メッセージングアプリ
 ```

-単一のNode.jsプロセス。チャネルはスキルで追加され、起動時に自己登録します — オーケストレーターは認証情報が存在するチャネルを接続します。エージェントはファイルシステム分離された独立したLinuxコンテナで実行されます。マウントされたディレクトリのみアクセス可能。グループごとのメッセージキューと同時実行制御。ファイルシステム経由のIPC。
+単一のNodeホストがセッションごとのエージェントコンテナをオーケストレーションします。メッセージが到着すると、ホストはエンティティモデル（ユーザー → メッセージンググループ → エージェントグループ → セッション）に沿ってルーティングし、セッションの`inbound.db`に書き込み、コンテナを起こします。コンテナ内部のagent-runnerは`inbound.db`をポーリングしてClaudeを実行し、レスポンスを`outbound.db`に書き込みます。ホストは`outbound.db`をポーリングし、チャネルアダプターを通じて配信します。

-詳細なアーキテクチャについては、[docs/SPEC.md](docs/SPEC.md)を参照してください。
+セッションごとに2つのSQLiteファイル、各ファイルにライターは1つだけ — クロスマウントの競合なし、IPCなし、stdinパイプなし。チャネルと代替プロバイダーは起動時に自己登録します。トランクはレジストリとChat SDKブリッジを同梱し、アダプター本体はフォークごとにスキルでインストールされます。
+
+詳しいアーキテクチャ説明は[docs/architecture.md](docs/architecture.md)を、3階層の分離モデルについては[docs/isolation-model.md](docs/isolation-model.md)を参照してください。

 主要ファイル：
- `src/index.ts` - オーケストレーター：状態、メッセージループ、エージェント呼び出し
- `src/channels/registry.ts` - チャネルレジストリ（起動時の自己登録）
- `src/ipc.ts` - IPCウォッチャーとタスク処理
- `src/router.ts` - メッセージフォーマットとアウトバウンドルーティング
- `src/group-queue.ts` - グローバル同時実行制限付きのグループごとのキュー
- `src/container-runner.ts` - ストリーミングエージェントコンテナの起動
- `src/task-scheduler.ts` - スケジュールタスクの実行
- `src/db.ts` - SQLite操作（メッセージ、グループ、セッション、状態）
- `groups/*/CLAUDE.md` - グループごとのメモリ
+- `src/index.ts` — エントリーポイント：DB初期化、チャネルアダプター、配信ポーリング、sweep
+- `src/router.ts` — インバウンドルーティング：メッセージンググループ → エージェントグループ → セッション → `inbound.db`
+- `src/delivery.ts` — `outbound.db`をポーリングし、アダプター経由で配信、システムアクションを処理
+- `src/host-sweep.ts` — 60秒ごとのsweep：ストール検出、期限到来メッセージの起動、繰り返し
+- `src/session-manager.ts` — セッションの解決、`inbound.db`と`outbound.db`のオープン
+- `src/container-runner.ts` — エージェントグループごとのコンテナ起動、OneCLIによるクレデンシャル注入
+- `src/db/` — セントラルDB（ユーザー、ロール、エージェントグループ、メッセージンググループ、配線、マイグレーション）
+- `src/channels/` — チャネルアダプターのインフラ（アダプターは`/add-<channel>`スキルでインストール）
+- `src/providers/` — ホスト側プロバイダー設定（`claude`はバンドル、その他はスキル経由）
+- `container/agent-runner/` — Bun製agent-runner：ポーリングループ、MCPツール、プロバイダー抽象化
+- `groups/<folder>/` — エージェントグループごとのファイルシステム（`CLAUDE.md`、スキル、コンテナ設定）

 ## FAQ

 **なぜDockerなのか？**

-Dockerはクロスプラットフォーム対応（macOS、Linux、さらにWSL2経由のWindows）と成熟したエコシステムを提供します。macOSでは、`/convert-to-apple-container`でオプションとしてApple Containerに切り替え、より軽量なネイティブランタイムを使用できます。
+Dockerはクロスプラットフォーム対応（macOS、Linux、WSL2経由のWindows）と成熟したエコシステムを提供します。macOSでは、`/convert-to-apple-container`でオプションとしてApple Containerに切り替え、より軽量なネイティブランタイムを使えます。さらに強い分離が必要なら、[Docker Sandboxes](docs/docker-sandboxes.md)が各コンテナをマイクロVM内で動作させます。

-**Linuxで実行できますか？**
+**LinuxやWindowsで実行できますか？**

-はい。DockerがデフォルトのランタイムでmacOSとLinuxの両方で動作します。`/setup`を実行するだけです。
+はい。Dockerがデフォルトのランタイムで、macOS、Linux、Windows（WSL2経由）で動作します。`bash nanoclaw.sh`を実行するだけです。

 **セキュリティは大丈夫ですか？**

-エージェントはアプリケーションレベルのパーミッションチェックの背後ではなく、コンテナで実行されます。明示的にマウントされたディレクトリのみアクセスできます。実行するものをレビューすべきですが、コードベースは十分に小さいため実際にレビュー可能です。完全なセキュリティモデルについては[docs/SECURITY.md](docs/SECURITY.md)を参照してください。
+エージェントはアプリケーションレベルのパーミッションチェックではなく、コンテナ内で実行されます。明示的にマウントされたディレクトリのみアクセス可能です。クレデンシャルはコンテナに渡されず、アウトバウンドAPIリクエストは[OneCLI Agent Vault](https://github.com/onecli/onecli)を経由し、プロキシレベルで認証を注入し、レートリミットやアクセスポリシーをサポートします。実行するものはレビューすべきですが、コードベースは実際にレビュー可能な規模です。完全なセキュリティモデルについては[セキュリティドキュメント](https://docs.nanoclaw.dev/concepts/security)を参照してください。

 **なぜ設定ファイルがないのか？**

-設定の肥大化を避けたいからです。すべてのユーザーがNanoClawをカスタマイズし、汎用的なシステムを設定するのではなく、コードが必要なことを正確に実行するようにすべきです。設定ファイルが欲しい場合は、Claudeに追加するよう伝えることができます。
+設定の肥大化を避けたいからです。すべてのユーザーがNanoClawをカスタマイズし、汎用的なシステムを設定するのではなくコードが自分の望み通りに動くようにすべきです。設定ファイルが欲しければClaudeに追加するよう伝えれば実現できます。

 **サードパーティやオープンソースモデルを使えますか？**

-はい。NanoClawはClaude API互換のモデルエンドポイントに対応しています。`.env`ファイルで以下の環境変数を設定してください：
+はい。推奨される方法は`/add-opencode`（OpenCode設定経由でOpenRouter、OpenAI、Google、DeepSeekなど）か`/add-ollama-provider`（Ollama経由でローカルのオープンウェイトモデル）です。どちらもエージェントグループごとに設定可能なので、同じインストール内で異なるエージェントが異なるバックエンドで動作できます。
+
+一時的な実験用には、Claude API互換のエンドポイントも`.env`で利用できます：

 ```bash
 ANTHROPIC_BASE_URL=https://your-api-endpoint.com
 ANTHROPIC_AUTH_TOKEN=your-token-here
 ```

-以下が使用可能です：
- [Ollama](https://ollama.ai)とAPIプロキシ経由のローカルモデル
- [Together AI](https://together.ai)、[Fireworks](https://fireworks.ai)等でホストされたオープンソースモデル
- Anthropic互換APIのカスタムモデルデプロイメント
-
-注意：最高の互換性のため、モデルはAnthropic APIフォーマットに対応している必要があります。
-
 **問題のデバッグ方法は？**

 Claude Codeに聞いてください。「スケジューラーが動いていないのはなぜ？」「最近のログには何がある？」「このメッセージに返信がなかったのはなぜ？」これがNanoClawの基盤となるAIネイティブなアプローチです。

 **セットアップがうまくいかない場合は？**

-問題がある場合、セットアップ中にClaudeが動的に修正を試みます。それでもうまくいかない場合は、`claude`を実行してから`/debug`を実行してください。Claudeが他のユーザーにも影響する可能性のある問題を見つけた場合は、セットアップのSKILL.mdを修正するPRを開いてください。
+ステップが失敗した場合、`nanoclaw.sh`は診断と再開のためにClaude Codeへ制御を渡します。それでも解決しなければ、`claude`を実行して`/debug`を呼び出してください。他のユーザーにも影響しそうな問題をClaudeが特定した場合は、該当のセットアップステップまたはスキルにPRを送ってください。

 **どのような変更がコードベースに受け入れられますか？**

-セキュリティ修正、バグ修正、明確な改善のみが基本設定に受け入れられます。それだけです。
+ベース設定に受け入れられるのは、セキュリティ修正、バグ修正、明確な改善のみです。それだけです。

-それ以外のすべて（新機能、OS互換性、ハードウェアサポート、機能拡張）はスキルとしてコントリビューションすべきです。
+それ以外（新機能、OS互換性、ハードウェアサポート、拡張など）は、`channels`または`providers`ブランチのスキルとしてコントリビュートしてください。

-これにより、基本システムを最小限に保ち、すべてのユーザーが不要な機能を継承することなく、自分のインストールをカスタマイズできます。
+これにより、ベースシステムを最小限に保ち、全ユーザーが不要な機能を継承することなく自分のインストールをカスタマイズできます。

 ## コミュニティ

-質問やアイデアは？[Discordに参加](https://discord.gg/VDdww8qS42)してください。
+質問やアイデアがありますか？[Discordに参加](https://discord.gg/VDdww8qS42)してください。

 ## 変更履歴

-破壊的変更と移行ノートについては[CHANGELOG.md](CHANGELOG.md)を参照してください。
+破壊的変更については[CHANGELOG.md](CHANGELOG.md)を、完全なリリース履歴はドキュメントサイトの[full release history](https://docs.nanoclaw.dev/changelog)を参照してください。

 ## ライセンス

@@ -3,93 +3,87 @@
 </p>

 <p align="center">
-  NanoClaw —— 您的专属 Claude 助手，在容器中安全运行。它轻巧易懂，并能根据您的个人需求灵活定制。
+  一个将智能体安全运行在独立容器中的 AI 助手。轻量、易于理解，并可根据您的需求完全定制。
 </p>

 <p align="center">
  <a href="https://nanoclaw.dev">nanoclaw.dev</a>&nbsp; • &nbsp;
+  <a href="https://docs.nanoclaw.dev">文档</a>&nbsp; • &nbsp;
  <a href="README.md">English</a>&nbsp; • &nbsp;
  <a href="README_ja.md">日本語</a>&nbsp; • &nbsp;
  <a href="https://discord.gg/VDdww8qS42"><img src="https://img.shields.io/discord/1470188214710046894?label=Discord&logo=discord&v=2" alt="Discord" valign="middle"></a>&nbsp; • &nbsp;
-  <a href="repo-tokens"><img src="repo-tokens/badge.svg" alt="34.9k tokens, 17% of context window" valign="middle"></a>
+  <a href="repo-tokens"><img src="repo-tokens/badge.svg" alt="repo tokens" valign="middle"></a>
 </p>

-> **注意：** 此中文翻译对应 v1 版本，已不反映最新的 v2 架构。请参考 [README.md](README.md) 获取最新内容。
+---

-通过 Claude Code，NanoClaw 可以动态重写自身代码，根据您的需求定制功能。
+## 我为什么创建 NanoClaw

-**新功能：** 首个支持 [Agent Swarms（智能体集群）](https://code.claude.com/docs/en/agent-teams) 的 AI 助手。可轻松组建智能体团队，在您的聊天中高效协作。
+[OpenClaw](https://github.com/openclaw/openclaw) 是一个令人印象深刻的项目，但我无法安心使用一个我不了解、却能访问我个人隐私的复杂软件。OpenClaw 有近 50 万行代码、53 个配置文件和 70+ 个依赖项。其安全性是应用级别的（白名单、配对码），而非真正的操作系统级隔离。所有东西都在一个共享内存的 Node 进程中运行。

-## 我为什么创建这个项目
-
-[OpenClaw](https://github.com/openclaw/openclaw) 是一个令人印象深刻的项目，但我无法安心使用一个我不了解却能访问我个人隐私的软件。OpenClaw 有近 50 万行代码、53 个配置文件和 70+ 个依赖项。其安全性是应用级别的（通过白名单、配对码实现），而非操作系统级别的隔离。所有东西都在一个共享内存的 Node 进程中运行。
-
-NanoClaw 用一个您能快速理解的代码库，为您提供了同样的核心功能。只有一个进程，少数几个文件。智能体（Agent）运行在具有文件系统隔离的真实 Linux 容器中，而不是依赖于权限检查。
+NanoClaw 用一个您能轻松理解的代码库提供了同样的核心功能：一个进程，少数几个文件。Claude 智能体运行在具有文件系统隔离的独立 Linux 容器中，而不是仅靠权限检查。

 ## 快速开始

 ```bash
-git clone https://github.com/qwibitai/nanoclaw.git
-cd nanoclaw
-claude
+git clone https://github.com/nanocoai/nanoclaw.git nanoclaw-v2
+cd nanoclaw-v2
+bash nanoclaw.sh
 ```

-然后运行 `/setup`。Claude Code 会处理一切：依赖安装、身份验证、容器设置、服务配置。
-
-> **注意：** 以 `/` 开头的命令（如 `/setup`、`/add-whatsapp`）是 [Claude Code 技能](https://code.claude.com/docs/en/skills)。请在 `claude` CLI 提示符中输入，而非在普通终端中。
+`nanoclaw.sh` 会把您从一台全新机器一直带到一个可以直接发消息的命名智能体。它会在缺失时安装 Node、pnpm 和 Docker，向 OneCLI 注册您的 Anthropic 凭据，构建智能体容器，并配对您的第一个渠道（Telegram、Discord、WhatsApp 或本地 CLI）。如果某一步失败，会自动调用 Claude Code 进行诊断并从中断处继续。

 ## 设计哲学

-**小巧易懂：** 单一进程，少量源文件。无微服务、无消息队列、无复杂抽象层。让 Claude Code 引导您轻松上手。
+**小到可以理解。** 单一进程，少量源文件，无微服务。如果您想了解完整的 NanoClaw 代码库，直接让 Claude Code 给您讲一遍就行。

-**通过隔离保障安全:** 智能体运行在 Linux 容器（在 macOS 上是 Apple Container，或 Docker）中。它们只能看到被明确挂载的内容。即便通过 Bash 访问也十分安全，因为所有命令都在容器内执行，不会直接操作您的宿主机。
+**通过隔离实现安全。** 智能体运行在 Linux 容器中，只能看到明确挂载的内容。Bash 访问是安全的，因为命令在容器内执行，而不是在您的宿主机上。

-**为单一用户打造:** 这不是一个框架，是一个完全符合您个人需求的、可工作的软件。您可以 Fork 本项目，然后让 Claude Code 根据您的精确需求进行修改和适配。
+**为个人用户打造。** NanoClaw 不是一个单体框架，而是能精确匹配每个用户需求的软件。它被设计成量身定制的，而不是臃肿膨胀。您创建自己的 fork，让 Claude Code 按您的需求修改它。

-**定制即代码修改:** 没有繁杂的配置文件。想要不同的行为？直接修改代码。代码库足够小，这样做是安全的。
+**定制 = 修改代码。** 没有配置膨胀。想要不同的行为？改代码。代码库小到改动是安全的。

-**AI 原生:** 无安装向导（由 Claude Code 指导安装）。无需监控仪表盘，直接询问 Claude 即可了解系统状况。无调试工具（描述问题，Claude 会修复它）。
+**AI 原生，混合式设计。** 安装与上手流程走的是经过优化的脚本路径，快速且确定。当某一步需要判断（安装失败、引导决策、定制化）时，控制权会无缝地交给 Claude Code。安装之后也不提供监控仪表盘或调试 UI：您在聊天中描述问题，Claude Code 来处理。

-**技能（Skills）优于功能（Features）:** 贡献者不应该向代码库添加新功能（例如支持 Telegram）。相反，他们应该贡献像 `/add-telegram` 这样的 [Claude Code 技能](https://code.claude.com/docs/en/skills)，这些技能可以改造您的 fork。最终，您得到的是只做您需要事情的整洁代码。
+**技能优于功能。** 主干只发布注册表和基础设施，不包含具体的渠道适配器或替代智能体提供者。各个渠道（Discord、Slack、Telegram、WhatsApp……）放在长期存在的 `channels` 分支上；替代提供者（OpenCode、Ollama）放在 `providers` 分支上。您运行 `/add-telegram`、`/add-opencode` 等，技能会把您所需要的模块精确地复制到您的 fork 里。不会出现您没要求的功能。

-**最好的工具套件，最好的模型:** 本项目运行在 Claude Agent SDK 之上，这意味着您直接运行的就是 Claude Code。Claude Code 高度强大，其编码和问题解决能力使其能够修改和扩展 NanoClaw，为每个用户量身定制。
+**最强的 harness，最强的模型。** NanoClaw 通过 Anthropic 官方的 Claude Agent SDK 原生使用 Claude Code，所以您能用上最新的 Claude 模型以及 Claude Code 的完整工具集——包括修改和扩展自己的 NanoClaw fork 的能力。其他提供者是可插拔选项：`/add-codex` 对应 OpenAI 的 Codex（ChatGPT 订阅或 API key），`/add-opencode` 通过 OpenCode 接入 OpenRouter、Google、DeepSeek 等，`/add-ollama-provider` 用于本地开源权重模型。提供者可按智能体组单独配置。

 ## 功能支持

- **多渠道消息** - 通过 WhatsApp、Telegram、Discord、Slack 或 Gmail 与您的助手对话。使用 `/add-whatsapp` 或 `/add-telegram` 等技能添加渠道，可同时运行一个或多个。
- **隔离的群组上下文** - 每个群组都拥有独立的 `CLAUDE.md` 记忆和隔离的文件系统。它们在各自的容器沙箱中运行，且仅挂载所需的文件系统。
- **主频道** - 您的私有频道（self-chat），用于管理控制；其他所有群组都完全隔离
- **计划任务** - 运行 Claude 的周期性作业，并可以给您回发消息
- **网络访问** - 搜索和抓取网页内容
- **容器隔离** - 智能体在 Apple Container (macOS) 或 Docker (macOS/Linux) 的沙箱中运行
- **智能体集群（Agent Swarms）** - 启动多个专业智能体团队，协作完成复杂任务（首个支持此功能的个人 AI 助手）
- **可选集成** - 通过技能添加 Gmail (`/add-gmail`) 等更多功能
+- **多渠道消息** — WhatsApp、Telegram、Discord、Slack、Microsoft Teams、iMessage、Matrix、Google Chat、Webex、Linear、GitHub、WeChat，以及通过 Resend 的邮件。按需通过 `/add-<channel>` 技能安装。可同时运行一个或多个。
+- **灵活的隔离模式** — 可为每个渠道配一个独立智能体以获得完全隐私，也可让一个智能体在多个渠道上共享、统一记忆但会话独立，或者把多个渠道合并到一个共享会话里，让一场对话横跨多个入口。通过 `/manage-channels` 按渠道选择。详见 [docs/isolation-model.md](docs/isolation-model.md)。
+- **每个智能体的独立工作区** — 每个智能体组都有自己的 `CLAUDE.md`、自己的记忆、自己的容器，以及您允许的挂载点。除非您明确接线，否则不会有东西越过边界。
+- **计划任务** — 运行 Claude 的周期性作业，可以给您回发消息。
+- **网络访问** — 搜索和抓取网页内容。
+- **容器隔离** — 智能体在 Docker（macOS/Linux/WSL2）中沙箱化运行，可选 [Docker Sandboxes](docs/docker-sandboxes.md) 的微虚拟机隔离，或在 macOS 上选用 Apple Container 作为原生运行时。
+- **凭据安全** — 智能体不持有原始 API key。出站请求经由 [OneCLI 的 Agent Vault](https://github.com/onecli/onecli)，在请求时注入凭据，并按每个智能体执行策略和速率限制。

 ## 使用方法

-使用触发词（默认为 `@Andy`）与您的助手对话：
+用触发词（默认为 `@Andy`）与您的助手对话：

 ```
-@Andy 每周一到周五早上9点，给我发一份销售渠道的概览（需要访问我的 Obsidian vault 文件夹）
-@Andy 每周五回顾过去一周的 git 历史，如果与 README 有出入，就更新它
-@Andy 每周一早上8点，从 Hacker News 和 TechCrunch 收集关于 AI 发展的资讯，然后发给我一份简报
+@Andy 每个工作日早上 9 点给我发一份销售渠道概览（可以访问我的 Obsidian vault 文件夹）
+@Andy 每周五回顾过去一周的 git 历史，如果与 README 有出入就更新它
+@Andy 每周一早上 8 点，从 Hacker News 和 TechCrunch 收集 AI 相关资讯，给我发一份简报
 ```

-在主频道（您的self-chat）中，可以管理群组和任务：
+在您拥有或管理的渠道里，还可以管理群组和任务：
 ```
-@Andy 列出所有群组的计划任务
+@Andy 列出所有群组里的计划任务
@Andy 暂停周一简报任务
@Andy 加入"家庭聊天"群组
 ```

 ## 定制

-没有需要学习的配置文件。直接告诉 Claude Code 您想要什么：
+NanoClaw 不用配置文件。想改就直接告诉 Claude Code：

 - "把触发词改成 @Bob"
- "记住以后回答要更简短直接"
- "当我说早上好的时候，加一个自定义的问候"
- "每周存储一次对话摘要"
+- "以后回答请更简短、更直接"
+- "我说早上好的时候加一个自定义问候"
+- "每周保存一次会话摘要"

 或者运行 `/customize` 进行引导式修改。

@@ -97,107 +91,103 @@ claude

 ## 贡献

-**不要添加功能，而是添加技能。**
+**不要加功能，要加技能。**

-如果您想添加 Telegram 支持，不要创建一个 PR 同时添加 Telegram 和 WhatsApp。而是贡献一个技能文件 (`.claude/skills/add-telegram/SKILL.md`)，教 Claude Code 如何改造一个 NanoClaw 安装以使用 Telegram。
+如果您想添加新的渠道或智能体提供者，不要把它加到主干上。新的渠道适配器进入 `channels` 分支；新的智能体提供者进入 `providers` 分支。用户在自己的 fork 上运行 `/add-<name>` 技能，由技能把相关模块复制到标准路径、接好注册、固定依赖版本。

-然后用户在自己的 fork 上运行 `/add-telegram`，就能得到只做他们需要事情的整洁代码，而不是一个试图支持所有用例的臃肿系统。
+这样主干始终保持为纯粹的注册表和基础设施，每个 fork 也都保持精简——用户只获得他们要求的渠道和提供者，其它什么也不会混进来。

-### RFS (技能征集)
+### RFS（技能征集）

 我们希望看到的技能：

 **通信渠道**
- `/add-signal` - 添加 Signal 作为渠道
-
-**会话管理**
- `/clear` - 添加一个 `/clear` 命令，用于压缩会话（在同一会话中总结上下文，同时保留关键信息）。这需要研究如何通过 Claude Agent SDK 以编程方式触发压缩。
+- `/add-signal` — 添加 Signal 作为渠道

 ## 系统要求

- macOS 或 Linux
- Node.js 20+
- [Claude Code](https://claude.ai/download)
- [Apple Container](https://github.com/apple/container) (macOS) 或 [Docker](https://docker.com/products/docker-desktop) (macOS/Linux)
+- macOS 或 Linux（Windows 通过 WSL2）
+- Node.js 20+ 和 pnpm 10+（安装脚本会在缺失时自动安装）
+- [Docker Desktop](https://docker.com/products/docker-desktop)（macOS/Windows）或 Docker Engine（Linux）
+- [Claude Code](https://claude.ai/download)，用于 `/customize`、`/debug`、安装过程中的错误恢复以及所有 `/add-<channel>` 技能

 ## 架构

 ```
-渠道 --> SQLite --> 轮询循环 --> 容器 (Claude Agent SDK) --> 响应
+消息应用 → 主机进程（路由器） → inbound.db → 容器（Bun、Claude Agent SDK） → outbound.db → 主机进程（投递） → 消息应用
 ```

-单一 Node.js 进程。渠道通过技能添加，启动时自注册 — 编排器连接具有凭据的渠道。智能体在具有文件系统隔离的 Linux 容器中执行。每个群组的消息队列带有并发控制。通过文件系统进行 IPC。
+单一 Node 主机编排每个会话的智能体容器。当一条消息到来时，主机按实体模型（用户 → 消息组 → 智能体组 → 会话）进行路由，写入该会话的 `inbound.db`，并唤醒容器。容器内部的 agent-runner 轮询 `inbound.db`，调用 Claude，并把响应写入 `outbound.db`。主机轮询 `outbound.db`，通过渠道适配器投递回去。

-完整架构详情请见 [docs/SPEC.md](docs/SPEC.md)。
+每个会话两个 SQLite 文件，每个文件只有一个写入者——没有跨挂载的锁争用，没有 IPC，没有 stdin 管道。渠道和替代提供者在启动时自注册；主干提供注册表和 Chat SDK 桥接，而适配器本身在每个 fork 里通过技能安装。
+
+完整架构说明见 [docs/architecture.md](docs/architecture.md)；三级隔离模型见 [docs/isolation-model.md](docs/isolation-model.md)。

 关键文件：
- `src/index.ts` - 编排器：状态管理、消息循环、智能体调用
- `src/channels/registry.ts` - 渠道注册表（启动时自注册）
- `src/ipc.ts` - IPC 监听与任务处理
- `src/router.ts` - 消息格式化与出站路由
- `src/group-queue.ts` - 带全局并发限制的群组队列
- `src/container-runner.ts` - 生成流式智能体容器
- `src/task-scheduler.ts` - 运行计划任务
- `src/db.ts` - SQLite 操作（消息、群组、会话、状态）
- `groups/*/CLAUDE.md` - 各群组的记忆
+- `src/index.ts` — 入口：数据库初始化、渠道适配器、投递轮询、sweep
+- `src/router.ts` — 入站路由：消息组 → 智能体组 → 会话 → `inbound.db`
+- `src/delivery.ts` — 轮询 `outbound.db`，通过适配器投递，处理系统动作
+- `src/host-sweep.ts` — 60 秒 sweep：失效检测、到期消息唤醒、循环任务
+- `src/session-manager.ts` — 解析会话，打开 `inbound.db` / `outbound.db`
+- `src/container-runner.ts` — 为每个智能体组启动容器，OneCLI 凭据注入
+- `src/db/` — 中心数据库（用户、角色、智能体组、消息组、接线、迁移）
+- `src/channels/` — 渠道适配器基础设施（适配器通过 `/add-<channel>` 技能安装）
+- `src/providers/` — 主机侧提供者配置（`claude` 内置，其他通过技能安装）
+- `container/agent-runner/` — Bun 版 agent-runner：轮询循环、MCP 工具、提供者抽象
+- `groups/<folder>/` — 每个智能体组的文件系统（`CLAUDE.md`、技能、容器配置）

 ## FAQ

-**为什么是 Docker？**
+**为什么用 Docker？**

-Docker 提供跨平台支持（macOS 和 Linux）和成熟的生态系统。在 macOS 上，您可以选择通过运行 `/convert-to-apple-container` 切换到 Apple Container，以获得更轻量级的原生运行时体验。
+Docker 提供跨平台支持（macOS、Linux、Windows via WSL2）和成熟的生态。在 macOS 上，您可以选择通过 `/convert-to-apple-container` 切换到 Apple Container，以获得更轻量的原生运行时。如需更强隔离，[Docker Sandboxes](docs/docker-sandboxes.md) 会把每个容器放到一台微虚拟机里运行。

-**我可以在 Linux 上运行吗？**
+**我可以在 Linux 或 Windows 上运行吗？**

-可以。Docker 是默认的容器运行时，在 macOS 和 Linux 上都可以使用。只需运行 `/setup`。
+可以。Docker 是默认运行时，可在 macOS、Linux 以及 Windows（通过 WSL2）上工作。运行 `bash nanoclaw.sh` 就行。

 **这个项目安全吗？**

-智能体在容器中运行，而不是在应用级别的权限检查之后。它们只能访问被明确挂载的目录。您仍然应该审查您运行的代码，但这个代码库小到您真的可以做到。完整的安全模型请见 [docs/SECURITY.md](docs/SECURITY.md)。
+智能体运行在容器里，而不是躲在应用级权限检查之后。它们只能访问明确挂载的目录。凭据不会进入容器——出站 API 请求通过 [OneCLI 的 Agent Vault](https://github.com/onecli/onecli) 在代理层注入认证，并支持速率限制和访问策略。您仍然应该审查自己要运行的代码，但代码库小到您真的能做到。完整的安全模型见 [安全文档](https://docs.nanoclaw.dev/concepts/security)。

 **为什么没有配置文件？**

-我们不希望配置泛滥。每个用户都应该定制它，让代码完全符合他们的需求，而不是去配置一个通用的系统。如果您喜欢用配置文件，告诉 Claude 让它加上。
+我们不想让配置泛滥。每位用户都应该定制 NanoClaw，让代码精确地做他们想要的事，而不是去配置一个通用系统。如果您更喜欢有配置文件，可以让 Claude 给您加。

 **我可以使用第三方或开源模型吗？**

-可以。NanoClaw 支持任何 API 兼容的模型端点。在 `.env` 文件中设置以下环境变量：
+可以。推荐做法是 `/add-opencode`（通过 OpenCode 配置接入 OpenRouter、OpenAI、Google、DeepSeek 等）或 `/add-ollama-provider`（通过 Ollama 使用本地开源权重模型）。两者都可以按智能体组单独配置，所以同一套安装里不同的智能体可以运行在不同的后端上。
+
+对于一次性实验，任何 Claude API 兼容的端点也可以通过 `.env` 使用：

 ```bash
 ANTHROPIC_BASE_URL=https://your-api-endpoint.com
 ANTHROPIC_AUTH_TOKEN=your-token-here
 ```

-这使您能够使用：
- 通过 [Ollama](https://ollama.ai) 配合 API 代理运行的本地模型
- 托管在 [Together AI](https://together.ai)、[Fireworks](https://fireworks.ai) 等平台上的开源模型
- 兼容 Anthropic API 格式的自定义模型部署
-
-注意：为获得最佳兼容性，模型需支持 Anthropic API 格式。
-
 **我该如何调试问题？**

-问 Claude Code。"为什么计划任务没有运行？" "最近的日志里有什么？" "为什么这条消息没有得到回应？" 这就是 AI 原生的方法。
+问 Claude Code。"为什么计划任务没运行？""最近的日志里有什么？""为什么这条消息没有得到回复？"这就是 NanoClaw 底层的 AI 原生方式。

-**为什么我的安装不成功？**
+**为什么安装对我不成功？**

-如果遇到问题，安装过程中 Claude 会尝试动态修复。如果问题仍然存在，运行 `claude`，然后运行 `/debug`。如果 Claude 发现一个可能影响其他用户的问题，请开一个 PR 来修改 setup SKILL.md。
+如果某一步失败，`nanoclaw.sh` 会把控制权交给 Claude Code 进行诊断并从中断处继续。如果还是没解决，运行 `claude`，然后 `/debug`。如果 Claude 发现一个可能影响其他用户的问题，请对相关的安装步骤或技能提 PR。

-**什么样的代码更改会被接受？**
+**什么样的更改会被接受进代码库？**

-安全修复、bug 修复，以及对基础配置的明确改进。仅此而已。
+进入基础配置的只会是：安全修复、bug 修复、明显的改进。仅此而已。

-其他一切（新功能、操作系统兼容性、硬件支持、增强功能）都应该作为技能来贡献。
+其他一切（新能力、操作系统兼容、硬件支持、增强）都应作为技能贡献到 `channels` 或 `providers` 分支。

-这使得基础系统保持最小化，并让每个用户可以定制他们的安装，而无需继承他们不想要的功能。
+这样基础系统保持最小化，每位用户都可以定制自己的安装，而不必继承他们不想要的功能。

 ## 社区

-有任何疑问或建议？欢迎[加入 Discord 社区](https://discord.gg/VDdww8qS42)与我们交流。
+有问题或想法？欢迎[加入 Discord](https://discord.gg/VDdww8qS42)。

 ## 更新日志

-破坏性变更和迁移说明请见 [CHANGELOG.md](CHANGELOG.md)。
+破坏性变更见 [CHANGELOG.md](CHANGELOG.md)，完整发布历史见文档站的 [full release history](https://docs.nanoclaw.dev/changelog)。

 ## 许可证

@@ -0,0 +1,50 @@
+# Releasing NanoClaw
+
+Starting with v2.0.63, the goal is to publish a GitHub Release for every `package.json` version bump that lands on `main`. Releases are cut manually by a maintainer, so there can be lag between a bump merging and its release being published. The intent is *timeliness*, not strict 1:1 correlation with every bump.
+
+Each release ships:
+
+- A tagged commit on `main` (`vX.Y.Z`).
+- A `CHANGELOG.md` entry under `## [<version>] - <YYYY-MM-DD>`.
+- A GitHub Release whose body mirrors the CHANGELOG entry plus a contributors section.
+
+## When to cut a release
+
+A release is cut by a maintainer publishing it. The trigger is a `package.json` bump on `main`, but the publish step is manual — there is no fixed schedule, and bumps that land back-to-back may be rolled into a single release (as v2.0.55 through v2.0.63 were). Cutting more frequently is preferable to batching: smaller releases are easier to read, pin, and revert.
+
+## What goes in a release
+
+`CHANGELOG.md` is the canonical record of user-visible change. The release body on GitHub mirrors it. Aim for:
+
+- **Bold lead-ins** per major feature or fix, then a sentence-case prose explanation.
+- **`[BREAKING]` prefix** for any change that requires user action. Always include the workaround inline — never link to a separate doc for the fix.
+- **Doc links** for major features (relative paths into the repo, e.g. `[setup/lib/install-slug.sh](setup/lib/install-slug.sh)`).
+- **Inline commands** for actionable steps, in backticks.
+- **Minor items** as single plain bullets at the bottom of the entry, no bold lead-in.
+- **No PR numbers** in the user-facing prose. PR references can live in the GitHub Release's `## Contributors` section.
+
+## Publishing the release
+
+1. Bump `package.json` and add a `CHANGELOG.md` entry in the same commit (commit message: `chore: bump version to vX.Y.Z`).
+2. Once the bump commit lands on `main`, open a draft GitHub Release:
+   - **Tag:** `vX.Y.Z`, target `main`.
+   - **Title:** `vX.Y.Z` (bare version — descriptive content lives in the body, matching the CHANGELOG header pattern).
+   - **Body:** copy the CHANGELOG entry verbatim. Append a `## Contributors` section listing every PR author who landed work in the release window. Append a `**Full Changelog**: https://github.com/nanocoai/nanoclaw/compare/<prev-tag>...vX.Y.Z` line at the bottom.
+3. If anyone in the window opened their first NanoClaw PR, add a `## New Contributors` section above `## Contributors`, with each first-timer's first PR link and an invite to Discord.
+4. Publish (not just save draft).
+
+## Rollup releases
+
+If multiple `package.json` bumps land between two GitHub Releases (as happened between v2.0.54 and v2.0.63), the next release is a rollup: its CHANGELOG entry covers everything merged since the last released tag, and the body opens with a one-line "Rollup release covering vX.Y.Z through vX.Y.W." note. After the catchup, return to one release per bump.
+
+## Channels and stability
+
+NanoClaw currently ships a single channel: every published release is a stable release.
+
+- **Latest** — the most recent release on `main`, shown as "Latest release" on the GitHub Releases page. Consumers that want auto-bump follow GitHub's `/releases/latest` pointer.
+- **Stable** — currently identical to latest. NanoClaw has no separate stable branch and no pre-release/RC channel.
+- **Pinned** — any tagged release. Reproducible and the recommended choice for packagers and forks; published tags are not moved or retracted.
+
+If a pre-release channel is introduced later (e.g. `vX.Y.Z-rc.N`), those releases will be marked "Pre-release" on GitHub so they do not become the `latest` pointer, and this section will be updated to describe the promotion path.
+
+The tag is the source of truth — a GitHub Release's `target_commitish` always points to a tagged commit.
@@ -0,0 +1,30 @@
+
+     [38;2;43;183;206m⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠰⣄⠘⣦⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢹⡆⢸⡆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀[39m  [2m[38;2;43;183;206m°[39m[22m
+     [38;2;43;183;206m⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⡇⢸⡇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⠀⠀⠀⠀⠀⢀⣠⣴⠾⠟⠛⠛⠿⢶⣦⣾⠇⣾⠁⠀⠀⠀⢀⣤⣤⠀⢀⣄⠀[39m   
+     [38;2;43;183;206m⠀⠀⠀⠀⣴⡿⡋⠀⠀⠀⠀⠀⢤⣾⣿⢛⢿⣏⠀⠀⠀⢰⣟⣽⡏⠀⣸⡿⣧[39m   
+   [2m[38;2;43;183;206mo[39m[22m [38;2;43;183;206m⠀⠀⢀⣾⠋⠀⠀⠀⠀⠀⠀⠀⠀⠘⠈⣧⣀⣿⣧⠀⠀⣿⣼⣿⣇⣾⠋⢠⣿[39m   
+     [38;2;43;183;206m⠀⠀⣾⢃⠀⢲⣷⡋⣰⡀⢀⣀⣀⡀⠠⣿⣿⣠⣿⣇⠀⣿⢻⣉⠉⠙⠠⣼⠇[39m   
+     [38;2;43;183;206m⠀⣼⡏⠃⠀⢸⣿⣿⡿⠃⣾⣷⣻⣿⡏⢹⠿⠿⣿⣿⢀⣿⣐⠙⣷⣦⡾⠋⠀[39m  [2m[38;2;43;183;206mo[39m[22m
+     [38;2;43;183;206m⢠⣿⡃⠀⠀⠀⠀⠀⠈⠀⠀⠉⠙⠁⠀⠀⠀⠐⣿⣿⣟⠁⣿⣿⠟⠋⠀⠀⠀[39m   
+   [2m[38;2;43;183;206m°[39m[22m [38;2;43;183;206m⢸⣿⣧⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣀⣨⣿⣿⣿⣿⣿⠟⠁⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⢸⣿⣿⣷⣤⣤⠀⣀⢀⠀⢀⣀⣠⣴⣶⣿⣿⣿⣿⡿⠛⠁⠀⠀⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⣿⢋⠿⣿⣿⣿⣿⡿⣿⣿⣿⣿⣿⣿⠿⠿⠿⣿⣅⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀[39m  [38;2;43;183;206mO[39m
+     [38;2;43;183;206m⣿⣿⠙⢾⣽⣟⣿⣿⣼⣿⣿⣿⣩⣶⣶⣦⠀⠀⠩⢻⣆⠀⠀⠀⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⠘⣿⣶⣤⣿⣿⣿⣿⣵⢖⡀⠉⠹⡛⢷⣝⡿⠁⠀⠀⣿⡆⠀⠀⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⠀⢹⣯⣽⣟⣛⣻⣿⣿⣾⣽⢶⣽⣿⣿⣿⣏⠀⠠⣤⣿⡇⠀⠀⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⠀⠀⠻⣿⣶⣾⣿⢿⣻⣿⣿⣿⣿⣿⣿⣏⣛⣧⣦⣿⣿⣧⣄⠀⠀⠀⠀⠀⠀[39m   
+   [38;2;43;183;206mo[39m [38;2;43;183;206m⠀⠀⠀⠈⠻⣿⣶⣥⣼⣿⣿⣽⣿⣿⣿⣷⣶⣾⣿⣿⣯⣘⣿⣧⠀⠀⠀⠀⠀[39m   
+     [38;2;43;183;206m⠀⠀⠀⠀⠤⣤⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⣿⠿⠿⠿⠋⠀⠀⠀⠀⠀[39m   
+
+[1m _  _               [22m[38;2;43;183;206m[1m  ___ _             [22m[39m
+[1m| \| |__ _ _ _  ___ [22m[38;2;43;183;206m[1m / __| |__ ___ __ __[22m[39m
+[1m| .` / _` | ' \/ _ \[22m[38;2;43;183;206m[1m| (__| / _` \ V  V /[22m[39m
+[1m|_|\_\__,_|_||_\___/[22m[38;2;43;183;206m[1m \___|_\__,_|\_/\_/ [22m[39m
+
+                 [2mSmall.[22m
+         [2mRuns on your machine.[22m
+            [2mYours to modify.[22m
+
+[38;2;5;62;165m════════════════════════════════════════[39m
@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+#
+# ncl — NanoClaw CLI launcher.
+#
+# Resolves the project root from this script's location, cd's there so the
+# host-resolved DATA_DIR matches the running host, and execs the TS entry
+# via tsx. Symlink this file into a directory on your PATH (or alias `ncl`
+# to its full path) to invoke from anywhere:
+#
+#   ln -s "$(pwd)/bin/ncl" /usr/local/bin/ncl
+#   # or
+#   alias ncl="$(pwd)/bin/ncl"
+
+set -euo pipefail
+
+SCRIPT="${BASH_SOURCE[0]}"
+# Resolve symlinks so PROJECT_ROOT points at the real checkout.
+while [ -h "$SCRIPT" ]; do
+  DIR="$(cd -P "$(dirname "$SCRIPT")" && pwd)"
+  SCRIPT="$(readlink "$SCRIPT")"
+  [[ "$SCRIPT" != /* ]] && SCRIPT="$DIR/$SCRIPT"
+done
+SCRIPT_DIR="$(cd -P "$(dirname "$SCRIPT")" && pwd)"
+PROJECT_ROOT="$(dirname "$SCRIPT_DIR")"
+
+cd "$PROJECT_ROOT"
+exec pnpm exec tsx src/cli/client.ts "$@"
@@ -19,9 +19,9 @@ ARG INSTALL_CJK_FONTS=false
 # Pin CLI versions for reproducibility. Bump deliberately — unpinned installs
 # mean every rebuild silently picks up the latest and can break in lockstep
 # across all users.
-ARG CLAUDE_CODE_VERSION=2.1.116
+ARG CLAUDE_CODE_VERSION=2.1.154
 ARG AGENT_BROWSER_VERSION=latest
-ARG VERCEL_VERSION=latest
+ARG VERCEL_VERSION=52.2.1
 ARG BUN_VERSION=1.3.12

 # ---- System dependencies -----------------------------------------------------
@@ -91,7 +91,13 @@ RUN --mount=type=cache,target=/root/.bun/install/cache \
 #     the SDK fails at spawn time with "native binary not found".
 ENV PNPM_HOME="/pnpm"
 ENV PATH="$PNPM_HOME:$PATH"
-RUN corepack enable
+# Pin pnpm to match the host (package.json packageManager). pnpm 11 stopped
+# honoring `only-built-dependencies[]=` in .npmrc for global installs, which
+# silently skips claude-code's native-binary postinstall and agent-browser's
+# bin chmod — the agent then crashes at runtime with "native binary not
+# installed". Keep this in lockstep with package.json's `packageManager`.
+ARG PNPM_VERSION=10.33.0
+RUN corepack enable && corepack prepare pnpm@${PNPM_VERSION} --activate

 RUN --mount=type=cache,target=/root/.cache/pnpm \
    echo "only-built-dependencies[]=agent-browser" > /root/.npmrc && \
@@ -104,6 +110,11 @@ RUN --mount=type=cache,target=/root/.cache/pnpm \
 RUN --mount=type=cache,target=/root/.cache/pnpm \
    pnpm install -g "@anthropic-ai/claude-code@${CLAUDE_CODE_VERSION}"

+# ---- ncl CLI wrapper ----------------------------------------------------------
+# Actual script lives in the mounted source at /app/src/cli/ncl.ts.
+RUN printf '#!/bin/sh\nexec bun /app/src/cli/ncl.ts "$@"\n' > /usr/local/bin/ncl && \
+    chmod +x /usr/local/bin/ncl
+
 # ---- Entrypoint --------------------------------------------------------------
 COPY entrypoint.sh /app/entrypoint.sh
 RUN chmod +x /app/entrypoint.sh
@@ -5,8 +5,9 @@
    "": {
      "name": "nanoclaw-agent-runner",
      "dependencies": {
-        "@anthropic-ai/claude-agent-sdk": "^0.2.116",
-        "@modelcontextprotocol/sdk": "^1.12.1",
+        "@anthropic-ai/claude-agent-sdk": "^0.3.154",
+        "@anthropic-ai/sdk": "^0.100.0",
+        "@modelcontextprotocol/sdk": "^1.29.0",
        "cron-parser": "^5.0.0",
        "zod": "^4.0.0",
      },
@@ -18,25 +19,25 @@
    },
  },
  "packages": {
-    "@anthropic-ai/claude-agent-sdk": ["@anthropic-ai/claude-agent-sdk@0.2.116", "", { "dependencies": { "@anthropic-ai/sdk": "^0.81.0", "@modelcontextprotocol/sdk": "^1.29.0" }, "optionalDependencies": { "@anthropic-ai/claude-agent-sdk-darwin-arm64": "0.2.116", "@anthropic-ai/claude-agent-sdk-darwin-x64": "0.2.116", "@anthropic-ai/claude-agent-sdk-linux-arm64": "0.2.116", "@anthropic-ai/claude-agent-sdk-linux-arm64-musl": "0.2.116", "@anthropic-ai/claude-agent-sdk-linux-x64": "0.2.116", "@anthropic-ai/claude-agent-sdk-linux-x64-musl": "0.2.116", "@anthropic-ai/claude-agent-sdk-win32-arm64": "0.2.116", "@anthropic-ai/claude-agent-sdk-win32-x64": "0.2.116" }, "peerDependencies": { "zod": "^4.0.0" } }, "sha512-5NKpgaOZkzNCGCvLxJZUVGimf5IcYmpQ2x2XrR9ilK+2UkWrnnwcUfIWo8bBz9e7lSYcUf9XleGigq2eOOF7aw=="],
+    "@anthropic-ai/claude-agent-sdk": ["@anthropic-ai/claude-agent-sdk@0.3.154", "", { "optionalDependencies": { "@anthropic-ai/claude-agent-sdk-darwin-arm64": "0.3.154", "@anthropic-ai/claude-agent-sdk-darwin-x64": "0.3.154", "@anthropic-ai/claude-agent-sdk-linux-arm64": "0.3.154", "@anthropic-ai/claude-agent-sdk-linux-arm64-musl": "0.3.154", "@anthropic-ai/claude-agent-sdk-linux-x64": "0.3.154", "@anthropic-ai/claude-agent-sdk-linux-x64-musl": "0.3.154", "@anthropic-ai/claude-agent-sdk-win32-arm64": "0.3.154", "@anthropic-ai/claude-agent-sdk-win32-x64": "0.3.154" }, "peerDependencies": { "@anthropic-ai/sdk": ">=0.93.0", "@modelcontextprotocol/sdk": "^1.29.0", "zod": "^4.0.0" } }, "sha512-iEn25urI2QrMPFIhId3h7v/7EG5gsmF7ooe+6EvsAosePeLmpVVerp5nXtHnlmBkMinLecurcPA+OddKw76jYw=="],

-    "@anthropic-ai/claude-agent-sdk-darwin-arm64": ["@anthropic-ai/claude-agent-sdk-darwin-arm64@0.2.116", "", { "os": "darwin", "cpu": "arm64" }, "sha512-mG19ovtXCpETmd5KmTU1JO2iIHZBG09IP8DmgZjLA3wLmTzpgn9Au9veRaeJeXb1EqiHiFZU+z+mNB79+w5v9g=="],
+    "@anthropic-ai/claude-agent-sdk-darwin-arm64": ["@anthropic-ai/claude-agent-sdk-darwin-arm64@0.3.154", "", { "os": "darwin", "cpu": "arm64" }, "sha512-oFW3LD5lYrKAU+AKu27Z8hrzqkrh362qQrwi/i3DxGcud9BXUycsXYjShpDj3D3JZu169UzZuSPhx1Wajmbiwg=="],

-    "@anthropic-ai/claude-agent-sdk-darwin-x64": ["@anthropic-ai/claude-agent-sdk-darwin-x64@0.2.116", "", { "os": "darwin", "cpu": "x64" }, "sha512-qC25N0HRM8IXbM4Qi4svH9f51Y6DciDvjLV+oNYnxkdPgDG8p/+b7vQirN7qPxytIQb2TPdoFgUeCsSe7lrQyw=="],
+    "@anthropic-ai/claude-agent-sdk-darwin-x64": ["@anthropic-ai/claude-agent-sdk-darwin-x64@0.3.154", "", { "os": "darwin", "cpu": "x64" }, "sha512-5BgWEueP+cqoctWjZYhCbyltuaV/N2DmKDXD3/69cKaVmJp8XL9OCzlq/HEirA/+Ssjskx6hDUBaOcpuZ3iwQA=="],

-    "@anthropic-ai/claude-agent-sdk-linux-arm64": ["@anthropic-ai/claude-agent-sdk-linux-arm64@0.2.116", "", { "os": "linux", "cpu": "arm64" }, "sha512-MQIcJhhPM+RPJ7kMQdOQarkJ2FlJqOiu953c08YyJOoWdHykd3DIiHws3mf1Mwl/dfFeIyshOVpNND3hyIy5Dg=="],
+    "@anthropic-ai/claude-agent-sdk-linux-arm64": ["@anthropic-ai/claude-agent-sdk-linux-arm64@0.3.154", "", { "os": "linux", "cpu": "arm64" }, "sha512-rRkW4SBL3W7zQvKscCIfIGlmoeuTbMV6dXFbPdmpRGvmYZIs79RpzO6xrGBnnhmm+B7znQ9oHAnffi/2FBgJbA=="],

-    "@anthropic-ai/claude-agent-sdk-linux-arm64-musl": ["@anthropic-ai/claude-agent-sdk-linux-arm64-musl@0.2.116", "", { "os": "linux", "cpu": "arm64" }, "sha512-Dg/T3NkSp35ODiwdhj0KquvC6Xu+DMbyWFNkfepA3bz4oF2SVSgyOPYwVmfoJerzEUnYDldP4YhOxRrhbt0vXA=="],
+    "@anthropic-ai/claude-agent-sdk-linux-arm64-musl": ["@anthropic-ai/claude-agent-sdk-linux-arm64-musl@0.3.154", "", { "os": "linux", "cpu": "arm64" }, "sha512-o2bCQN4Xn3UqCLErC5m4T7u0yYArJYmgFCUFnA6K96DdW2RERvx+gTKXxWuHEBkDO+eMoHLHLxk0u2jGES00Ng=="],

-    "@anthropic-ai/claude-agent-sdk-linux-x64": ["@anthropic-ai/claude-agent-sdk-linux-x64@0.2.116", "", { "os": "linux", "cpu": "x64" }, "sha512-Bww1fzQB+vcF0tRhmCAlwSsN4wR2HgX7pBT9AWuwzJj6DKsVC23N54Ea80lsnM7dTUtUTrGYMTwVUHTWqfYnfQ=="],
+    "@anthropic-ai/claude-agent-sdk-linux-x64": ["@anthropic-ai/claude-agent-sdk-linux-x64@0.3.154", "", { "os": "linux", "cpu": "x64" }, "sha512-GpiFF8Ez6PbM3m0gqtCo/FKM346qyRdP7VhbmJzdnbNKTiiUZ66vDQyEUPZPCG24ZkrG4m96KpRIUwY08rHiNg=="],

-    "@anthropic-ai/claude-agent-sdk-linux-x64-musl": ["@anthropic-ai/claude-agent-sdk-linux-x64-musl@0.2.116", "", { "os": "linux", "cpu": "x64" }, "sha512-LMYxUMa1nK4N9BPRJdcGBAvl9rjTI4ZHo+kfAKrJ3MlfB6VFF1tRIubwsWOaOtkuNazMdAYovsZJg4bdzOBBTQ=="],
+    "@anthropic-ai/claude-agent-sdk-linux-x64-musl": ["@anthropic-ai/claude-agent-sdk-linux-x64-musl@0.3.154", "", { "os": "linux", "cpu": "x64" }, "sha512-zA7S8Lm6O4QBsUpbhiOht8BgiXHOBBFUIo8ZLK6r5wAatK3Q44syWVxICeyCnR6wqfnkf3cugCw27ycS6vVgaA=="],

-    "@anthropic-ai/claude-agent-sdk-win32-arm64": ["@anthropic-ai/claude-agent-sdk-win32-arm64@0.2.116", "", { "os": "win32", "cpu": "arm64" }, "sha512-h0YO1vkTIeUtffQhONrYbNC1pXmk1yjb1xxMEw7bAwucqtFoFpLDWe+q4+RhxaQr8ZOj6LtRE/U3dzPWHOlshA=="],
+    "@anthropic-ai/claude-agent-sdk-win32-arm64": ["@anthropic-ai/claude-agent-sdk-win32-arm64@0.3.154", "", { "os": "win32", "cpu": "arm64" }, "sha512-cDW1YFbU/PJFlrGXhlAGcbkXt80sEO6WtnH8nN8YHXLn5NWduy2q7o/qC6i8XozgvRGf6t/eMoH7IasGIEDhDw=="],

-    "@anthropic-ai/claude-agent-sdk-win32-x64": ["@anthropic-ai/claude-agent-sdk-win32-x64@0.2.116", "", { "os": "win32", "cpu": "x64" }, "sha512-3lllmtDFHgpW0ZM3iNvxsEjblrgRzF9Qm1lxTOtunP3hIn+pA/IkWMtKlN1ixxWiaBguLVQkJ90V6JHsvJJIvw=="],
+    "@anthropic-ai/claude-agent-sdk-win32-x64": ["@anthropic-ai/claude-agent-sdk-win32-x64@0.3.154", "", { "os": "win32", "cpu": "x64" }, "sha512-tSKaIIpL72OPg3WfzZTCIl8OJgcbq4qieu8/fDWjsdeQuari9gQMIuEflFphk9HqNsxpSmDqKi8Sm5mW2V566Q=="],

-    "@anthropic-ai/sdk": ["@anthropic-ai/sdk@0.81.0", "", { "dependencies": { "json-schema-to-ts": "^3.1.1" }, "peerDependencies": { "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["zod"], "bin": { "anthropic-ai-sdk": "bin/cli" } }, "sha512-D4K5PvEV6wPiRtVlVsJHIUhHAmOZ6IT/I9rKlTf84gR7GyyAurPJK7z9BOf/AZqC5d1DhYQGJNKRmV+q8dGhgw=="],
+    "@anthropic-ai/sdk": ["@anthropic-ai/sdk@0.100.0", "", { "dependencies": { "json-schema-to-ts": "^3.1.1", "standardwebhooks": "^1.0.0" }, "peerDependencies": { "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["zod"], "bin": { "anthropic-ai-sdk": "bin/cli" } }, "sha512-cAm3aXm6qAiHIvHxyIIGd6tVmsD2gDqlc2h0R20ijNUzGgVnIN822bit4mKbF6CkuV7qIrLQIPoAepHEpanrQQ=="],

    "@babel/runtime": ["@babel/runtime@7.29.2", "", {}, "sha512-JiDShH45zKHWyGe4ZNVRrCjBz8Nh9TMmZG1kh4QTK8hCBTWBi8Da+i7s1fJw7/lYpM4ccepSNfqzZ/QvABBi5g=="],

@@ -44,6 +45,8 @@

    "@modelcontextprotocol/sdk": ["@modelcontextprotocol/sdk@1.29.0", "", { "dependencies": { "@hono/node-server": "^1.19.9", "ajv": "^8.17.1", "ajv-formats": "^3.0.1", "content-type": "^1.0.5", "cors": "^2.8.5", "cross-spawn": "^7.0.5", "eventsource": "^3.0.2", "eventsource-parser": "^3.0.0", "express": "^5.2.1", "express-rate-limit": "^8.2.1", "hono": "^4.11.4", "jose": "^6.1.3", "json-schema-typed": "^8.0.2", "pkce-challenge": "^5.0.0", "raw-body": "^3.0.0", "zod": "^3.25 || ^4.0", "zod-to-json-schema": "^3.25.1" }, "peerDependencies": { "@cfworker/json-schema": "^4.1.1" }, "optionalPeers": ["@cfworker/json-schema"] }, "sha512-zo37mZA9hJWpULgkRpowewez1y6ML5GsXJPY8FI0tBBCd77HEvza4jDqRKOXgHNn867PVGCyTdzqpz0izu5ZjQ=="],

+    "@stablelib/base64": ["@stablelib/base64@1.0.1", "", {}, "sha512-1bnPQqSxSuc3Ii6MhBysoWCg58j97aUjuCSZrGSmDxNqtytIi0k8utUenAwTZN4V5mXXYGsVUI9zeBqy+jBOSQ=="],
+
    "@types/bun": ["@types/bun@1.3.12", "", { "dependencies": { "bun-types": "1.3.12" } }, "sha512-DBv81elK+/VSwXHDlnH3Qduw+KxkTIWi7TXkAeh24zpi5l0B2kUg9Ga3tb4nJaPcOFswflgi/yAvMVBPrxMB+A=="],

    "@types/node": ["@types/node@22.19.17", "", { "dependencies": { "undici-types": "~6.21.0" } }, "sha512-wGdMcf+vPYM6jikpS/qhg6WiqSV/OhG+jeeHT/KlVqxYfD40iYJf9/AE1uQxVWFvU7MipKRkRv8NSHiCGgPr8Q=="],
@@ -108,6 +111,8 @@

    "fast-deep-equal": ["fast-deep-equal@3.1.3", "", {}, "sha512-f3qQ9oQy9j2AhBe/H9VC91wLmKBCCU/gDOnKNAYG5hswO7BLKj09Hc5HYNz9cGI++xlpDCIgDaitVs03ATR84Q=="],

+    "fast-sha256": ["fast-sha256@1.3.0", "", {}, "sha512-n11RGP/lrWEFI/bWdygLxhI+pVeo1ZYIVwvvPkW7azl/rOy+F3HYRZ2K5zeE9mmkhQppyv9sQFx0JM9UabnpPQ=="],
+
    "fast-uri": ["fast-uri@3.1.0", "", {}, "sha512-iPeeDKJSWf4IEOasVVrknXpaBV0IApz/gp7S2bb7Z4Lljbl2MGJRqInZiUrQwV16cpzw/D3S5j5Julj/gT52AA=="],

    "finalhandler": ["finalhandler@2.1.1", "", { "dependencies": { "debug": "^4.4.0", "encodeurl": "^2.0.0", "escape-html": "^1.0.3", "on-finished": "^2.4.1", "parseurl": "^1.3.3", "statuses": "^2.0.1" } }, "sha512-S8KoZgRZN+a5rNwqTxlZZePjT/4cnm0ROV70LedRHZ0p8u9fRID0hJUZQpkKLzro8LfmC8sx23bY6tVNxv8pQA=="],
@@ -216,6 +221,8 @@

    "side-channel-weakmap": ["side-channel-weakmap@1.0.2", "", { "dependencies": { "call-bound": "^1.0.2", "es-errors": "^1.3.0", "get-intrinsic": "^1.2.5", "object-inspect": "^1.13.3", "side-channel-map": "^1.0.1" } }, "sha512-WPS/HvHQTYnHisLo9McqBHOJk2FkHO/tlpvldyrnem4aeQp4hai3gythswg6p01oSoTl58rcpiFAjF2br2Ak2A=="],

+    "standardwebhooks": ["standardwebhooks@1.0.0", "", { "dependencies": { "@stablelib/base64": "^1.0.0", "fast-sha256": "^1.3.0" } }, "sha512-BbHGOQK9olHPMvQNHWul6MYlrRTAOKn03rOe4A8O3CLWhNf4YHBqq2HJKKC+sfqpxiBY52pNeesD6jIiLDz8jg=="],
+
    "statuses": ["statuses@2.0.2", "", {}, "sha512-DvEy55V3DB7uknRo+4iOGT5fP1slR8wQohVdknigZPMpMstaKJQWhwiYBACJE3Ul2pTnATihhBYnRhZQHGBiRw=="],

    "toidentifier": ["toidentifier@1.0.1", "", {}, "sha512-o5sSPKEkg/DIQNmH43V0/uerLrpzVedkUh8tGNvaeXpfpuwjKenlSox/2O/BTlZUtEe+JG7s5YhEz608PlAHRA=="],
@@ -9,8 +9,9 @@
    "test": "bun test"
  },
  "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.2.116",
-    "@modelcontextprotocol/sdk": "^1.12.1",
+    "@anthropic-ai/claude-agent-sdk": "^0.3.154",
+    "@anthropic-ai/sdk": "^0.100.0",
+    "@modelcontextprotocol/sdk": "^1.29.0",
    "cron-parser": "^5.0.0",
    "zod": "^4.0.0"
  },
@@ -0,0 +1,254 @@
+#!/usr/bin/env bun
+/**
+ * ncl — NanoClaw CLI client (container edition).
+ *
+ * Same interface as the host-side `bin/ncl`. Detects that it's inside a
+ * container (the session DBs exist at /workspace/) and uses a DB transport
+ * instead of the Unix socket transport.
+ *
+ * Writes a cli_request system message to outbound.db, polls inbound.db
+ * for the response. Self-contained — no imports from agent-runner.
+ */
+import { Database } from 'bun:sqlite';
+
+// ---------------------------------------------------------------------------
+// Frame types (mirrors src/cli/frame.ts on the host)
+// ---------------------------------------------------------------------------
+
+type RequestFrame = {
+  id: string;
+  command: string;
+  args: Record<string, unknown>;
+};
+
+type ResponseFrame =
+  | { id: string; ok: true; data: unknown }
+  | { id: string; ok: false; error: { code: string; message: string } };
+
+// ---------------------------------------------------------------------------
+// Paths
+// ---------------------------------------------------------------------------
+
+const INBOUND_DB = '/workspace/inbound.db';
+const OUTBOUND_DB = '/workspace/outbound.db';
+
+// ---------------------------------------------------------------------------
+// DB transport
+// ---------------------------------------------------------------------------
+
+function generateId(): string {
+  return `cli-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
+}
+
+/**
+ * Write a cli_request to outbound.db.
+ *
+ * Uses BEGIN IMMEDIATE to acquire a write lock before reading max(seq),
+ * preventing seq collisions with concurrent agent-runner writes.
+ */
+function writeRequest(req: RequestFrame): void {
+  const db = new Database(OUTBOUND_DB);
+  db.exec('PRAGMA journal_mode = DELETE');
+  db.exec('PRAGMA busy_timeout = 5000');
+
+  const inDb = new Database(INBOUND_DB, { readonly: true });
+  inDb.exec('PRAGMA busy_timeout = 5000');
+
+  try {
+    db.exec('BEGIN IMMEDIATE');
+    const maxOut = (db.prepare('SELECT COALESCE(MAX(seq), 0) AS m FROM messages_out').get() as { m: number }).m;
+    const maxIn = (inDb.prepare('SELECT COALESCE(MAX(seq), 0) AS m FROM messages_in').get() as { m: number }).m;
+    const max = Math.max(maxOut, maxIn);
+    const nextSeq = max % 2 === 0 ? max + 1 : max + 2;
+
+    db.prepare(
+      `INSERT INTO messages_out (id, seq, timestamp, kind, content)
+       VALUES ($id, $seq, datetime('now'), 'system', $content)`,
+    ).run({
+      $id: req.id,
+      $seq: nextSeq,
+      $content: JSON.stringify({
+        action: 'cli_request',
+        requestId: req.id,
+        command: req.command,
+        args: req.args,
+      }),
+    });
+    db.exec('COMMIT');
+  } catch (e) {
+    db.exec('ROLLBACK');
+    throw e;
+  } finally {
+    inDb.close();
+    db.close();
+  }
+}
+
+/**
+ * Poll inbound.db for a cli_response matching our requestId.
+ * Opens a fresh connection each poll (mmap_size=0) for cross-mount visibility.
+ */
+function pollResponse(requestId: string, timeoutMs: number): ResponseFrame | null {
+  const deadline = Date.now() + timeoutMs;
+
+  while (Date.now() < deadline) {
+    const inDb = new Database(INBOUND_DB, { readonly: true });
+    inDb.exec('PRAGMA busy_timeout = 5000');
+    inDb.exec('PRAGMA mmap_size = 0');
+
+    try {
+      const row = inDb
+        .prepare("SELECT id, content FROM messages_in WHERE status = 'pending' AND content LIKE ?")
+        .get(`%"requestId":"${requestId}"%`) as { id: string; content: string } | null;
+
+      if (row) {
+        // Mark as completed via processing_ack so agent-runner skips it
+        const outDb = new Database(OUTBOUND_DB);
+        outDb.exec('PRAGMA journal_mode = DELETE');
+        outDb.exec('PRAGMA busy_timeout = 5000');
+        outDb
+          .prepare(
+            "INSERT OR REPLACE INTO processing_ack (message_id, status, status_changed) VALUES (?, 'completed', datetime('now'))",
+          )
+          .run(row.id);
+        outDb.close();
+
+        const parsed = JSON.parse(row.content);
+        return parsed.frame as ResponseFrame;
+      }
+    } finally {
+      inDb.close();
+    }
+
+    Bun.sleepSync(500);
+  }
+
+  return null;
+}
+
+// ---------------------------------------------------------------------------
+// Arg parsing (mirrors host-side client.ts)
+// ---------------------------------------------------------------------------
+
+function parseArgv(argv: string[]): {
+  command: string;
+  args: Record<string, unknown>;
+  json: boolean;
+} {
+  const positional: string[] = [];
+  const args: Record<string, unknown> = {};
+  let json = false;
+
+  for (let i = 0; i < argv.length; i++) {
+    const a = argv[i];
+    if (a === '--json') {
+      json = true;
+      continue;
+    }
+    if (a.startsWith('--')) {
+      const key = a.slice(2);
+      const next = argv[i + 1];
+      if (next === undefined || next.startsWith('--')) {
+        args[key] = true;
+      } else {
+        args[key] = next;
+        i++;
+      }
+      continue;
+    }
+    positional.push(a);
+  }
+
+  if (positional.length === 0) {
+    process.stderr.write('ncl: missing command\n');
+    printUsage();
+    process.exit(2);
+  }
+
+  // Join all positionals with dashes. The dispatcher trims the last
+  // segment as a target ID if the full name isn't a registered command.
+  const command = positional.join('-');
+
+  return { command, args, json };
+}
+
+function printUsage(): void {
+  process.stdout.write(
+    ['Usage: ncl <command> [--key value ...] [--json]', '', 'Run `ncl help` to list available commands.', ''].join('\n'),
+  );
+}
+
+// ---------------------------------------------------------------------------
+// Formatting (mirrors src/cli/format.ts on the host)
+// ---------------------------------------------------------------------------
+
+function formatHuman(resp: ResponseFrame): string {
+  if (!resp.ok) {
+    return `error (${resp.error.code}): ${resp.error.message}\n`;
+  }
+
+  const data = resp.data;
+  if (!Array.isArray(data) || data.length === 0) {
+    return JSON.stringify(data, null, 2) + '\n';
+  }
+
+  const isFlat = data.every(
+    (r) =>
+      typeof r === 'object' &&
+      r !== null &&
+      !Array.isArray(r) &&
+      Object.values(r as Record<string, unknown>).every((v) => typeof v !== 'object' || v === null),
+  );
+
+  if (!isFlat) return JSON.stringify(data, null, 2) + '\n';
+
+  const keys = Object.keys(data[0] as Record<string, unknown>);
+  const widths = keys.map((k) =>
+    Math.max(k.length, ...data.map((r) => String((r as Record<string, unknown>)[k] ?? '').length)),
+  );
+
+  const header = keys.map((k, i) => k.padEnd(widths[i])).join('  ');
+  const sep = widths.map((w) => '-'.repeat(w)).join('  ');
+  const rows = data.map((r) =>
+    keys
+      .map((k, i) => String((r as Record<string, unknown>)[k] ?? '').padEnd(widths[i]))
+      .join('  '),
+  );
+
+  return [header, sep, ...rows, ''].join('\n');
+}
+
+// ---------------------------------------------------------------------------
+// Main
+// ---------------------------------------------------------------------------
+
+const argv = process.argv.slice(2);
+
+if (argv.length === 0 || argv[0] === '--help' || argv[0] === '-h') {
+  printUsage();
+  process.exit(0);
+}
+
+const { command, args, json } = parseArgv(argv);
+const requestId = generateId();
+const req: RequestFrame = { id: requestId, command, args };
+
+writeRequest(req);
+
+const resp = pollResponse(requestId, 30_000);
+
+if (!resp) {
+  process.stderr.write('ncl: command timed out after 30s\n');
+  process.exit(2);
+}
+
+if (json) {
+  process.stdout.write(JSON.stringify(resp, null, 2) + '\n');
+} else {
+  const output = formatHuman(resp);
+  if (!resp.ok) {
+    process.stderr.write(output);
+    process.exit(1);
+  }
+  process.stdout.write(output);
+}
@@ -0,0 +1,34 @@
+/**
+ * PreCompact hook script — outputs custom compaction instructions to stdout.
+ *
+ * Claude Code captures the stdout of PreCompact shell hooks and passes it
+ * as `customInstructions` to the compaction prompt. This ensures the
+ * compaction summary preserves message routing context that the agent needs
+ * to correctly address responses.
+ *
+ * Invoked by the PreCompact hook in .claude-shared/settings.json:
+ *   "command": "bun /app/src/compact-instructions.ts"
+ */
+import { getAllDestinations } from './destinations.js';
+
+const destinations = getAllDestinations();
+const names = destinations.map((d) => d.name);
+
+const instructions = [
+  'Preserve the following in the compaction summary:',
+  '',
+  '1. For recent messages, keep the full XML structure including all attributes:',
+  '   - <message from="..." sender="..." time="..."> for chat messages',
+  '   - <task from="..." time="..."> for scheduled tasks',
+  '   - <webhook from="..." source="..." event="..."> for webhooks',
+  '   The message content can be summarized if long, but the XML tags and attributes must remain.',
+  '',
+  '2. Preserve the chronological message/reply sequence of recent exchanges.',
+  '   The agent needs to see: who said what, in what order, and from which destination.',
+  '',
+  '3. At the END of the compaction summary, include this verbatim reminder:',
+  '   "You MUST wrap all responses in <message to="name">...</message> blocks.',
+  `   Available destinations: ${names.length > 0 ? names.map((n) => `\`${n}\``).join(', ') : '(none)'}."`,
+];
+
+console.log(instructions.join('\n'));
@@ -16,6 +16,8 @@ export interface RunnerConfig {
  agentGroupId: string;
  maxMessagesPerPrompt: number;
  mcpServers: Record<string, { command: string; args: string[]; env: Record<string, string> }>;
+  model?: string;
+  effort?: string;
 }

 const DEFAULT_MAX_MESSAGES = 10;
@@ -43,6 +45,8 @@ export function loadConfig(): RunnerConfig {
    agentGroupId: (raw.agentGroupId as string) || '',
    maxMessagesPerPrompt: (raw.maxMessagesPerPrompt as number) || DEFAULT_MAX_MESSAGES,
    mcpServers: (raw.mcpServers as RunnerConfig['mcpServers']) || {},
+    model: (raw.model as string) || undefined,
+    effort: (raw.effort as string) || undefined,
  };

  return _config;
@@ -0,0 +1,29 @@
+/**
+ * Per-batch context the poll loop publishes for downstream consumers
+ * (MCP tools, etc.) that don't sit on the poll-loop's call stack.
+ *
+ * Today the only field is `inReplyTo` — the id of the first inbound
+ * message in the batch the agent is currently processing. MCP tools like
+ * `send_message` and `send_file` read this and stamp it onto the outbound
+ * row so the host's a2a return-path routing can correlate replies back to
+ * the originating session.
+ *
+ * This is module-level state on purpose: the agent-runner is single-process
+ * and processes one batch at a time. Poll-loop calls `setCurrentInReplyTo`
+ * before invoking the provider and `clearCurrentInReplyTo` after the batch
+ * completes (or errors out).
+ */
+let currentInReplyTo: string | null = null;
+
+export function setCurrentInReplyTo(id: string | null): void {
+  currentInReplyTo = id;
+}
+
+export function clearCurrentInReplyTo(): void {
+  currentInReplyTo = null;
+}
+
+export function getCurrentInReplyTo(): string | null {
+  return currentInReplyTo;
+}
+
@@ -27,12 +27,46 @@ const DEFAULT_HEARTBEAT_PATH = '/workspace/.heartbeat';
 let _inbound: Database | null = null;
 let _outbound: Database | null = null;
 let _heartbeatPath: string = DEFAULT_HEARTBEAT_PATH;
+let _testMode = false;

-/** Inbound DB — container opens read-only (host is the sole writer). */
+/**
+ * Avoid all cached db reads; open inbound.db read-only with mmap and page cache disabled.
+ *
+ * Use this (not getInboundDb) for readers that need to see host-written rows
+ * promptly — e.g. messages_in polling. Caller must .close() the returned
+ * connection (try/finally).
+ *
+ * Needed for mounts where host writes don't reliably invalidate
+ * SQLite's caches: virtiofs (Colima, Lima, Podman Machine, Apple
+ * Container), NFS.
+ *
+ * Cost is microseconds per query, so safe for universal use.
+ */
+export function openInboundDb(): Database {
+  // In test mode return a thin wrapper over the in-memory singleton.
+  // Callers do try/finally { db.close() } — the wrapper no-ops close()
+  // so the singleton survives for the rest of the test.
+  if (_testMode && _inbound) {
+    const db = _inbound;
+    return { prepare: (sql: string) => db.prepare(sql), exec: (sql: string) => db.exec(sql), close: () => {} } as unknown as Database;
+  }
+  const db = new Database(DEFAULT_INBOUND_PATH, { readonly: true });
+  db.exec('PRAGMA busy_timeout = 5000');
+  db.exec('PRAGMA mmap_size = 0');
+  return db;
+}
+
+/**
+ * Inbound DB — long-lived singleton, OK for tables the host writes once
+ * at spawn and never again (destinations, session_routing). For
+ * messages_in polling — where the host writes continuously and a stale
+ * view causes the pollHandle hang — use `openInboundDb()` instead.
+ */
 export function getInboundDb(): Database {
  if (!_inbound) {
    _inbound = new Database(DEFAULT_INBOUND_PATH, { readonly: true });
    _inbound.exec('PRAGMA busy_timeout = 5000');
+    _inbound.exec('PRAGMA mmap_size = 0');
  }
  return _inbound;
 }
@@ -144,6 +178,7 @@ export function clearStaleProcessingAcks(): void {

 /** For tests — creates in-memory DBs with the session schemas. */
 export function initTestSessionDb(): { inbound: Database; outbound: Database } {
+  _testMode = true;
  _inbound = new Database(':memory:');
  _inbound.exec('PRAGMA foreign_keys = ON');
  _inbound.exec(`
@@ -161,7 +196,8 @@ export function initTestSessionDb(): { inbound: Database; outbound: Database } {
      platform_id    TEXT,
      channel_type   TEXT,
      thread_id      TEXT,
-      content        TEXT NOT NULL
+      content        TEXT NOT NULL,
+      on_wake        INTEGER NOT NULL DEFAULT 0
    );
    CREATE TABLE delivered (
      message_out_id      TEXT PRIMARY KEY,
@@ -220,6 +256,7 @@ export function initTestSessionDb(): { inbound: Database; outbound: Database } {
 export function closeSessionDb(): void {
  _inbound?.close();
  _inbound = null;
+  _testMode = false;
  _outbound?.close();
  _outbound = null;
 }
@@ -8,7 +8,20 @@
 * processing_ack. The host reads processing_ack to sync message lifecycle.
 */
 import { getConfig } from '../config.js';
-import { getInboundDb, getOutboundDb } from './connection.js';
+import { openInboundDb, getOutboundDb } from './connection.js';
+
+// Cache whether inbound.db has the on_wake column (added in v2.0.48).
+// The container opens inbound.db read-only, so it can't ALTER —
+// gracefully degrade when running against an older session DB.
+let _hasOnWake: boolean | null = null;
+function hasOnWakeColumn(db: ReturnType<typeof openInboundDb>): boolean {
+  if (_hasOnWake !== null) return _hasOnWake;
+  const cols = new Set(
+    (db.prepare("PRAGMA table_info('messages_in')").all() as Array<{ name: string }>).map((c) => c.name),
+  );
+  _hasOnWake = cols.has('on_wake');
+  return _hasOnWake;
+}

 export interface MessageInRow {
  id: string;
@@ -49,32 +62,38 @@ function getMaxMessagesPerPrompt(): number {
 * sees the prior context it missed. Host's countDueMessages gates waking on
 * trigger=1 separately (see src/db/session-db.ts).
 */
-export function getPendingMessages(): MessageInRow[] {
-  const inbound = getInboundDb();
+export function getPendingMessages(isFirstPoll = false): MessageInRow[] {
+  const inbound = openInboundDb();
  const outbound = getOutboundDb();

-  const pending = inbound
-    .prepare(
-      `SELECT * FROM messages_in
-       WHERE status = 'pending'
-         AND (process_after IS NULL OR datetime(process_after) <= datetime('now'))
-       ORDER BY seq DESC
-       LIMIT ?`,
-    )
-    .all(getMaxMessagesPerPrompt()) as MessageInRow[];
+  try {
+    const onWakeFilter = hasOnWakeColumn(inbound) ? 'AND (on_wake = 0 OR ?1 = 1)' : '';
+    const pending = inbound
+      .prepare(
+        `SELECT * FROM messages_in
+         WHERE status = 'pending'
+           AND (process_after IS NULL OR datetime(process_after) <= datetime('now'))
+           ${onWakeFilter}
+         ORDER BY seq DESC
+         LIMIT ?2`,
+      )
+      .all(isFirstPoll ? 1 : 0, getMaxMessagesPerPrompt()) as MessageInRow[];

-  if (pending.length === 0) return [];
+    if (pending.length === 0) return [];

-  // Filter out messages already acknowledged in outbound.db
-  const ackedIds = new Set(
-    (outbound.prepare('SELECT message_id FROM processing_ack').all() as Array<{ message_id: string }>).map(
-      (r) => r.message_id,
-    ),
-  );
+    // Filter out messages already acknowledged in outbound.db
+    const ackedIds = new Set(
+      (outbound.prepare('SELECT message_id FROM processing_ack').all() as Array<{ message_id: string }>).map(
+        (r) => r.message_id,
+      ),
+    );

-  // Reverse: we fetched DESC to take the most recent N, but the agent
-  // should see them in chronological order (oldest first).
-  return pending.filter((m) => !ackedIds.has(m.id)).reverse();
+    // Reverse: we fetched DESC to take the most recent N, but the agent
+    // should see them in chronological order (oldest first).
+    return pending.filter((m) => !ackedIds.has(m.id)).reverse();
+  } finally {
+    inbound.close();
+  }
 }

 /** Mark messages as processing — writes to processing_ack in outbound.db. */
@@ -112,7 +131,12 @@ export function markFailed(id: string): void {

 /** Get a message by ID (read from inbound.db). */
 export function getMessageIn(id: string): MessageInRow | undefined {
-  return getInboundDb().prepare('SELECT * FROM messages_in WHERE id = ?').get(id) as MessageInRow | undefined;
+  const inbound = openInboundDb();
+  try {
+    return inbound.prepare('SELECT * FROM messages_in WHERE id = ?').get(id) as MessageInRow | undefined;
+  } finally {
+    inbound.close();
+  }
 }

 /**
@@ -120,19 +144,23 @@ export function getMessageIn(id: string): MessageInRow | undefined {
 * Reads from inbound.db, checks processing_ack to skip already-handled responses.
 */
 export function findQuestionResponse(questionId: string): MessageInRow | undefined {
-  const inbound = getInboundDb();
+  const inbound = openInboundDb();
  const outbound = getOutboundDb();

-  const response = inbound
-    .prepare("SELECT * FROM messages_in WHERE status = 'pending' AND content LIKE ?")
-    .get(`%"questionId":"${questionId}"%`) as MessageInRow | undefined;
+  try {
+    const response = inbound
+      .prepare("SELECT * FROM messages_in WHERE status = 'pending' AND content LIKE ?")
+      .get(`%"questionId":"${questionId}"%`) as MessageInRow | undefined;

-  if (!response) return undefined;
+    if (!response) return undefined;

-  // Check it hasn't been acked already
-  const acked = outbound.prepare('SELECT 1 FROM processing_ack WHERE message_id = ?').get(response.id);
-  if (acked) return undefined;
+    // Check it hasn't been acked already
+    const acked = outbound.prepare('SELECT 1 FROM processing_ack WHERE message_id = ?').get(response.id);
+    if (acked) return undefined;

-  return response;
+    return response;
+  } finally {
+    inbound.close();
+  }
 }

@@ -0,0 +1,100 @@
+import { beforeEach, describe, expect, test } from 'bun:test';
+
+import { getOutboundDb, initTestSessionDb } from './connection.js';
+import {
+  clearContinuation,
+  getContinuation,
+  migrateLegacyContinuation,
+  setContinuation,
+} from './session-state.js';
+
+beforeEach(() => {
+  initTestSessionDb();
+});
+
+function seedLegacy(value: string): void {
+  getOutboundDb()
+    .prepare('INSERT INTO session_state (key, value, updated_at) VALUES (?, ?, ?)')
+    .run('sdk_session_id', value, new Date().toISOString());
+}
+
+describe('session-state — per-provider continuations', () => {
+  test('set/get round-trip, case-insensitive provider key', () => {
+    setContinuation('claude', 'claude-conv-1');
+    expect(getContinuation('claude')).toBe('claude-conv-1');
+    expect(getContinuation('Claude')).toBe('claude-conv-1');
+    expect(getContinuation('CLAUDE')).toBe('claude-conv-1');
+  });
+
+  test('providers are isolated — switching reads the right slot', () => {
+    setContinuation('claude', 'claude-conv-1');
+    setContinuation('codex', 'codex-thread-xyz');
+
+    expect(getContinuation('claude')).toBe('claude-conv-1');
+    expect(getContinuation('codex')).toBe('codex-thread-xyz');
+  });
+
+  test('clearContinuation only affects the specified provider', () => {
+    setContinuation('claude', 'keep-me');
+    setContinuation('codex', 'drop-me');
+
+    clearContinuation('codex');
+
+    expect(getContinuation('claude')).toBe('keep-me');
+    expect(getContinuation('codex')).toBeUndefined();
+  });
+
+  test('unknown provider returns undefined', () => {
+    expect(getContinuation('never-used')).toBeUndefined();
+  });
+});
+
+describe('session-state — legacy migration', () => {
+  test('adopts legacy value into current provider when current is empty', () => {
+    seedLegacy('old-session-id');
+
+    const adopted = migrateLegacyContinuation('claude');
+
+    expect(adopted).toBe('old-session-id');
+    expect(getContinuation('claude')).toBe('old-session-id');
+  });
+
+  test('always deletes legacy row regardless of migration outcome', () => {
+    seedLegacy('old-session-id');
+    setContinuation('claude', 'existing');
+
+    migrateLegacyContinuation('claude');
+
+    // After migration the legacy key must be gone, whether or not it was adopted.
+    // A subsequent migration for a different provider must not see it.
+    const resultAfterSecondCall = migrateLegacyContinuation('codex');
+    expect(resultAfterSecondCall).toBeUndefined();
+  });
+
+  test('prefers existing current-provider slot over legacy', () => {
+    seedLegacy('legacy-value');
+    setContinuation('claude', 'claude-value');
+
+    const result = migrateLegacyContinuation('claude');
+
+    expect(result).toBe('claude-value');
+    expect(getContinuation('claude')).toBe('claude-value');
+  });
+
+  test('no legacy row — returns current provider value (possibly undefined)', () => {
+    expect(migrateLegacyContinuation('claude')).toBeUndefined();
+
+    setContinuation('codex', 'codex-value');
+    expect(migrateLegacyContinuation('codex')).toBe('codex-value');
+  });
+
+  test('migration is idempotent on a second call (legacy already gone)', () => {
+    seedLegacy('once');
+
+    const first = migrateLegacyContinuation('claude');
+    expect(first).toBe('once');
+
+    const second = migrateLegacyContinuation('claude');
+    expect(second).toBe('once');
+  });
+});
@@ -2,12 +2,20 @@
 * Persistent key/value state for the container. Lives in outbound.db
 * (container-owned, already scoped per channel/thread).
 *
- * Primary use: remember the SDK session ID so the agent's conversation
- * resumes across container restarts. Cleared by /clear.
+ * Primary use: remember each provider's opaque continuation id so the
+ * agent's conversation resumes across container restarts. Keyed per
+ * provider because continuations are provider-private — a Claude
+ * conversation id means nothing to Codex and vice versa. Switching
+ * providers is therefore lossless: each provider's last thread stays
+ * on file and resumes cleanly if the user flips back.
 */
 import { getOutboundDb } from './connection.js';

-const SDK_SESSION_KEY = 'sdk_session_id';
+const LEGACY_KEY = 'sdk_session_id';
+
+function continuationKey(providerName: string): string {
+  return `continuation:${providerName.toLowerCase()}`;
+}

 function getValue(key: string): string | undefined {
  const row = getOutboundDb()
@@ -18,9 +26,7 @@ function getValue(key: string): string | undefined {

 function setValue(key: string, value: string): void {
  getOutboundDb()
-    .prepare(
-      'INSERT OR REPLACE INTO session_state (key, value, updated_at) VALUES (?, ?, ?)',
-    )
+    .prepare('INSERT OR REPLACE INTO session_state (key, value, updated_at) VALUES (?, ?, ?)')
    .run(key, value, new Date().toISOString());
 }

@@ -28,14 +34,46 @@ function deleteValue(key: string): void {
  getOutboundDb().prepare('DELETE FROM session_state WHERE key = ?').run(key);
 }

-export function getStoredSessionId(): string | undefined {
-  return getValue(SDK_SESSION_KEY);
+/**
+ * One-time migration of the pre-per-provider continuation row.
+ *
+ * Before this was keyed per provider, continuations lived under the
+ * single key `sdk_session_id`. On container start, if that legacy row
+ * exists and the current provider has no continuation of its own, adopt
+ * the legacy value into the current provider's slot (best-guess — the
+ * legacy row was written by whatever provider ran last). The legacy row
+ * is always deleted so future provider flips never re-read a stale id
+ * through the wrong lens.
+ *
+ * Returns the continuation the caller should use at startup (either the
+ * current provider's existing value, the adopted legacy value, or
+ * undefined).
+ */
+export function migrateLegacyContinuation(providerName: string): string | undefined {
+  const legacy = getValue(LEGACY_KEY);
+  const currentKey = continuationKey(providerName);
+  const current = getValue(currentKey);
+
+  if (legacy === undefined) return current;
+
+  // Always drop the legacy row so no future provider reads it.
+  deleteValue(LEGACY_KEY);
+
+  // Prefer the current provider's own slot if one already exists.
+  if (current !== undefined) return current;
+
+  setValue(currentKey, legacy);
+  return legacy;
 }

-export function setStoredSessionId(sessionId: string): void {
-  setValue(SDK_SESSION_KEY, sessionId);
+export function getContinuation(providerName: string): string | undefined {
+  return getValue(continuationKey(providerName));
 }

-export function clearStoredSessionId(): void {
-  deleteValue(SDK_SESSION_KEY);
+export function setContinuation(providerName: string, id: string): void {
+  setValue(continuationKey(providerName), id);
+}
+
+export function clearContinuation(providerName: string): void {
+  deleteValue(continuationKey(providerName));
 }
@@ -0,0 +1,63 @@
+import { afterEach, beforeEach, describe, expect, it } from 'bun:test';
+
+import { closeSessionDb, getInboundDb, initTestSessionDb } from './db/connection.js';
+import { buildSystemPromptAddendum } from './destinations.js';
+
+beforeEach(() => {
+  initTestSessionDb();
+});
+
+afterEach(() => {
+  closeSessionDb();
+});
+
+function seedDestination(name: string, displayName: string, channelType: string, platformId: string): void {
+  getInboundDb()
+    .prepare(
+      `INSERT INTO destinations (name, display_name, type, channel_type, platform_id, agent_group_id)
+       VALUES (?, ?, 'channel', ?, ?, NULL)`,
+    )
+    .run(name, displayName, channelType, platformId);
+}
+
+describe('buildSystemPromptAddendum — multi-destination routing guidance', () => {
+  it('includes default-routing nudge when there are >1 destinations', () => {
+    seedDestination('casa', 'Casa', 'whatsapp', 'group-1@g.us');
+    seedDestination('whatsapp-mg-17780', 'whatsapp-mg-17780', 'whatsapp', 'phone-2@s.whatsapp.net');
+
+    const prompt = buildSystemPromptAddendum('Casa');
+
+    expect(prompt).toContain('default to addressing the destination it came `from`');
+    expect(prompt).toContain('from="name"');
+    expect(prompt).toContain('`casa`');
+    expect(prompt).toContain('`whatsapp-mg-17780`');
+  });
+
+  it('describes message wrapping for a single destination', () => {
+    seedDestination('casa', 'Casa', 'whatsapp', 'group-1@g.us');
+
+    const prompt = buildSystemPromptAddendum('Casa');
+
+    expect(prompt).toContain('Wrap each delivered message');
+    expect(prompt).toContain('<message to="name">');
+    expect(prompt).toContain('`casa`');
+  });
+
+  it('handles the no-destination case without crashing', () => {
+    const prompt = buildSystemPromptAddendum('Casa');
+
+    expect(prompt).toContain('no configured destinations');
+    expect(prompt).not.toContain('default to addressing');
+  });
+
+  it('includes default-routing and wrapping instructions for single destination', () => {
+    seedDestination('casa', 'Casa', 'whatsapp', 'group-1@g.us');
+
+    const prompt = buildSystemPromptAddendum('Casa');
+
+    expect(prompt).toContain('Wrap each delivered message');
+    expect(prompt).toContain('<message to="name">');
+    expect(prompt).toContain('default to addressing the destination it came `from`');
+    expect(prompt).toContain('`casa`');
+  });
+});
@@ -102,34 +102,29 @@ function buildDestinationsSection(): string {
    ].join('\n');
  }

-  // Single-destination shortcut: the agent just writes its response normally.
+  const lines = ['## Sending messages', ''];
  if (all.length === 1) {
    const d = all[0];
    const label = d.displayName && d.displayName !== d.name ? ` (${d.displayName})` : '';
-    return [
-      '## Sending messages',
-      '',
-      `Your messages are delivered to \`${d.name}\`${label}. Just write your response directly — no special wrapping needed.`,
-      '',
-      'To mark something as scratchpad (logged but not sent), wrap it in `<internal>...</internal>`.',
-      '',
-      'To send a message mid-response (e.g., an acknowledgment before a long task), call the `send_message` MCP tool.',
-    ].join('\n');
+    lines.push(`Your destination is \`${d.name}\`${label}.`);
+  } else {
+    lines.push('You can send messages to the following destinations:', '');
+    for (const d of all) {
+      const label = d.displayName && d.displayName !== d.name ? ` (${d.displayName})` : '';
+      lines.push(`- \`${d.name}\`${label}`);
+    }
  }
-
-  const lines = ['## Sending messages', '', 'You can send messages to the following destinations:', ''];
-  for (const d of all) {
-    const label = d.displayName && d.displayName !== d.name ? ` (${d.displayName})` : '';
-    lines.push(`- \`${d.name}\`${label}`);
-  }
-  lines.push('');
-  lines.push('To send a message, wrap it in a `<message to="name">...</message>` block.');
-  lines.push('You can include multiple `<message>` blocks in one response to send to multiple destinations.');
-  lines.push('Text outside of `<message>` blocks is scratchpad — logged but not sent anywhere.');
-  lines.push('Use `<internal>...</internal>` to make scratchpad intent explicit.');
  lines.push('');
  lines.push(
-    'To send a message mid-response (e.g., an acknowledgment before a long task), call the `send_message` MCP tool with the `to` parameter set to a destination name.',
+    'Wrap each delivered message in a `<message to="name">…</message>` block; include several blocks in one response to address several destinations. `<internal>…</internal>` marks thinking you don\'t want sent.',
+  );
+  lines.push('');
+  lines.push(
+    'When replying to an incoming message, default to addressing the destination it came `from` (every inbound `<message>` tag carries a `from="name"` attribute). Pick a different destination when the request asks for it (e.g., "tell Laura that…").',
+  );
+  lines.push('');
+  lines.push(
+    'The `send_message` MCP tool is the same delivery, available mid-turn — handy for a quick acknowledgment ("on it") before a slow tool call. Each `send_message` call and each final-response `<message>` block lands as its own message in the conversation, so they read as a sequence rather than as one combined reply.',
  );
  return lines.join('\n');
 }
@@ -51,14 +51,43 @@ describe('context timezone header', () => {
    expect(result).toContain(`<context timezone="${TIMEZONE}"`);
  });

-  it('header comes before the <messages> block', () => {
+  it('header comes before the first <message> block when multiple are present', () => {
    insertMessage('m1', 'chat', { sender: 'Alice', text: 'one' });
    insertMessage('m2', 'chat', { sender: 'Bob', text: 'two' });
    const result = formatMessages(getPendingMessages());
    const ctxIdx = result.indexOf('<context');
-    const msgsIdx = result.indexOf('<messages>');
+    const firstMsgIdx = result.indexOf('<message ');
    expect(ctxIdx).toBeGreaterThanOrEqual(0);
-    expect(msgsIdx).toBeGreaterThan(ctxIdx);
+    expect(firstMsgIdx).toBeGreaterThan(ctxIdx);
+  });
+});
+
+describe('multi-message chat batches', () => {
+  // Regression guard for #2555: an outer `<messages>` envelope around
+  // multiple chat messages caused the Claude Agent SDK to emit a synthetic
+  // `No response requested.` stub instead of calling the API. Each
+  // `<message>` block is self-contained; concatenating them is enough.
+  it('does NOT wrap multiple chat messages in an outer <messages> envelope', () => {
+    insertMessage('m1', 'chat', { sender: 'Alice', text: 'one' });
+    insertMessage('m2', 'chat', { sender: 'Bob', text: 'two' });
+    const result = formatMessages(getPendingMessages());
+    expect(result).not.toContain('<messages>');
+    expect(result).not.toContain('</messages>');
+  });
+
+  it('emits one <message> block per inbound row, in order', () => {
+    insertMessage('m1', 'chat', { sender: 'Alice', text: 'first' });
+    insertMessage('m2', 'chat', { sender: 'Bob', text: 'second' });
+    insertMessage('m3', 'chat', { sender: 'Carol', text: 'third' });
+    const result = formatMessages(getPendingMessages());
+    const matches = result.match(/<message [^>]*>/g) ?? [];
+    expect(matches.length).toBe(3);
+    const firstIdx = result.indexOf('first');
+    const secondIdx = result.indexOf('second');
+    const thirdIdx = result.indexOf('third');
+    expect(firstIdx).toBeGreaterThan(0);
+    expect(secondIdx).toBeGreaterThan(firstIdx);
+    expect(thirdIdx).toBeGreaterThan(secondIdx);
  });
 });

@@ -11,7 +11,7 @@ import { TIMEZONE, formatLocalTime } from './timezone.js';
 */
 export type CommandCategory = 'admin' | 'filtered' | 'passthrough' | 'none';

-const ADMIN_COMMANDS = new Set(['/remote-control', '/clear', '/compact', '/context', '/cost', '/files']);
+const ADMIN_COMMANDS = new Set(['/remote-control', '/clear', '/compact', '/context', '/cost', '/files', '/upload-trace']);
 const FILTERED_COMMANDS = new Set(['/help', '/login', '/logout', '/doctor', '/config', '/start']);

 export interface CommandInfo {
@@ -66,6 +66,18 @@ export function isClearCommand(msg: MessageInRow): boolean {
  return text.toLowerCase().startsWith('/clear');
 }

+/**
+ * True for any chat that needs the outer loop's command path: /clear plus
+ * admin/passthrough slash commands the SDK can only dispatch when they are
+ * a query's first input. Used by the follow-up poller to bail out and let
+ * the outer loop reopen the query.
+ */
+export function isRunnerCommand(msg: MessageInRow): boolean {
+  if (msg.kind !== 'chat' && msg.kind !== 'chat-sdk') return false;
+  const cat = categorizeMessage(msg).category;
+  return cat === 'admin' || cat === 'passthrough';
+}
+
 // eslint-disable-next-line @typescript-eslint/no-explicit-any
 function extractSenderId(msg: MessageInRow, content: any): string | null {
  const raw: string | null = content?.senderId || content?.author?.userId || null;
@@ -143,16 +155,15 @@ export function formatMessages(messages: MessageInRow[]): string {
 }

 function formatChatMessages(messages: MessageInRow[]): string {
-  if (messages.length === 1) {
-    return formatSingleChat(messages[0]);
-  }
-
-  const lines = ['<messages>'];
-  for (const msg of messages) {
-    lines.push(formatSingleChat(msg));
-  }
-  lines.push('</messages>');
-  return lines.join('\n');
+  // Each `<message id="..." from="...">...</message>` block is self-contained;
+  // concatenating them reads to the agent as a sequence of distinct messages.
+  // Earlier revisions wrapped multi-message batches in an outer `<messages>`
+  // envelope, but the Claude Agent SDK responded to that shape with a
+  // synthetic stub (`model: "<synthetic>"`, `content: "No response
+  // requested."`) instead of calling the API — see #2555 for the full trace.
+  // The fix is simply to drop the wrapper; the single-message path (which
+  // already worked) is now just the N=1 case of the same code.
+  return messages.map(formatSingleChat).join('\n');
 }

 function formatSingleChat(msg: MessageInRow): string {
@@ -165,40 +176,49 @@ function formatSingleChat(msg: MessageInRow): string {
  const replyPrefix = formatReplyContext(content.replyTo);
  const attachmentsSuffix = formatAttachments(content.attachments);

-  // Look up the destination name for the origin (reverse map lookup).
-  // If not found, fall back to a raw channel:platform_id marker so nothing
-  // gets silently dropped — this should only happen if the destination was
-  // removed between when the message was received and when it's being processed.
-  const fromDest = findByRouting(msg.channel_type, msg.platform_id);
-  const fromAttr = fromDest
-    ? ` from="${escapeXml(fromDest.name)}"`
-    : msg.channel_type || msg.platform_id
-      ? ` from="unknown:${escapeXml(msg.channel_type || '')}:${escapeXml(msg.platform_id || '')}"`
-      : '';
+  const fromAttr = originAttr(msg);

  return `<message${idAttr}${fromAttr} sender="${escapeXml(sender)}" time="${escapeXml(time)}"${replyAttr}>${replyPrefix}${escapeXml(text)}${attachmentsSuffix}</message>`;
 }

+/**
+ * Build a ` from="destination_name"` attribute string from a message's routing
+ * fields. Shared by all formatters so the agent always knows where a message
+ * originated — critical for explicit addressing.
+ */
+function originAttr(msg: MessageInRow): string {
+  const fromDest = findByRouting(msg.channel_type, msg.platform_id);
+  if (fromDest) return ` from="${escapeXml(fromDest.name)}"`;
+  if (msg.channel_type || msg.platform_id) {
+    return ` from="unknown:${escapeXml(msg.channel_type || '')}:${escapeXml(msg.platform_id || '')}"`;
+  }
+  return '';
+}
+
 function formatTaskMessage(msg: MessageInRow): string {
  const content = parseContent(msg.content);
-  const parts = ['[SCHEDULED TASK]'];
+  const from = originAttr(msg);
+  const time = formatLocalTime(msg.timestamp, TIMEZONE);
+  const parts: string[] = [];
  if (content.scriptOutput) {
-    parts.push('', 'Script output:', JSON.stringify(content.scriptOutput, null, 2));
+    parts.push('Script output:', JSON.stringify(content.scriptOutput, null, 2), '');
  }
-  parts.push('', 'Instructions:', content.prompt || '');
-  return parts.join('\n');
+  parts.push('Instructions:', content.prompt || '');
+  return `<task${from} time="${escapeXml(time)}">${parts.join('\n')}</task>`;
 }

 function formatWebhookMessage(msg: MessageInRow): string {
  const content = parseContent(msg.content);
  const source = content.source || 'unknown';
  const event = content.event || 'unknown';
-  return `[WEBHOOK: ${source}/${event}]\n\n${JSON.stringify(content.payload || content, null, 2)}`;
+  const from = originAttr(msg);
+  return `<webhook${from} source="${escapeXml(source)}" event="${escapeXml(event)}">${JSON.stringify(content.payload || content, null, 2)}</webhook>`;
 }

 function formatSystemMessage(msg: MessageInRow): string {
  const content = parseContent(msg.content);
-  return `[SYSTEM RESPONSE]\n\nAction: ${content.action || 'unknown'}\nStatus: ${content.status || 'unknown'}\nResult: ${JSON.stringify(content.result || null)}`;
+  const from = originAttr(msg);
+  return `<system_response${from} action="${escapeXml(content.action || 'unknown')}" status="${escapeXml(content.status || 'unknown')}">${JSON.stringify(content.result || null)}</system_response>`;
 }

 /**
@@ -91,10 +91,13 @@ async function main(): Promise<void> {
    mcpServers,
    env: { ...process.env },
    additionalDirectories: additionalDirectories.length > 0 ? additionalDirectories : undefined,
+    model: config.model,
+    effort: config.effort,
  });

  await runPollLoop({
    provider,
+    providerName,
    cwd: CWD,
    systemContext: { instructions },
  });
@@ -3,6 +3,7 @@ import { describe, it, expect, beforeEach, afterEach } from 'bun:test';
 import { initTestSessionDb, closeSessionDb, getInboundDb, getOutboundDb } from './db/connection.js';
 import { getUndeliveredMessages } from './db/messages-out.js';
 import { getPendingMessages } from './db/messages-in.js';
+import { getContinuation, setContinuation } from './db/session-state.js';
 import { MockProvider } from './providers/mock.js';
 import { runPollLoop } from './poll-loop.js';

@@ -74,6 +75,163 @@ describe('poll loop integration', () => {
    await loopPromise.catch(() => {});
  });

+  it('should resolve thread_id per-destination, not from global routing', async () => {
+    // Seed a second destination
+    getInboundDb()
+      .prepare(
+        `INSERT INTO destinations (name, display_name, type, channel_type, platform_id, agent_group_id)
+         VALUES ('slack-test', 'Slack Test', 'channel', 'slack', 'chan-2', NULL)`,
+      )
+      .run();
+
+    // Insert messages from each destination with distinct thread IDs
+    insertMessage('m-discord', { sender: 'Alice', text: 'from discord' }, { platformId: 'chan-1', channelType: 'discord', threadId: 'discord-thread-1' });
+    insertMessage('m-slack', { sender: 'Bob', text: 'from slack' }, { platformId: 'chan-2', channelType: 'slack', threadId: 'slack-thread-99' });
+
+    // Agent replies to both destinations
+    const provider = new MockProvider({}, () =>
+      '<message to="discord-test">reply-d</message><message to="slack-test">reply-s</message>',
+    );
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length >= 2, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    const discordOut = out.find((m) => m.platform_id === 'chan-1');
+    const slackOut = out.find((m) => m.platform_id === 'chan-2');
+
+    expect(discordOut).toBeDefined();
+    expect(discordOut!.thread_id).toBe('discord-thread-1');
+    expect(discordOut!.in_reply_to).toBe('m-discord');
+
+    expect(slackOut).toBeDefined();
+    expect(slackOut!.thread_id).toBe('slack-thread-99');
+    expect(slackOut!.in_reply_to).toBe('m-slack');
+
+    await loopPromise.catch(() => {});
+  });
+
+  it('bare text produces no outbound messages (scratchpad only)', async () => {
+    insertMessage('m1', { sender: 'Alice', text: 'hello' }, { platformId: 'chan-1', channelType: 'discord' });
+
+    // Agent responds with bare text — no <message to="..."> wrapping
+    const provider = new MockProvider({}, () => 'I am thinking about this...');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    // Wait long enough for the poll loop to process
+    await sleep(1000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(0);
+
+    await loopPromise.catch(() => {});
+  });
+
+  it('unknown destination is dropped, valid destination is sent', async () => {
+    insertMessage('m1', { sender: 'Alice', text: 'hi' }, { platformId: 'chan-1', channelType: 'discord' });
+
+    const provider = new MockProvider(
+      {},
+      () => '<message to="nonexistent">dropped</message><message to="discord-test">delivered</message>',
+    );
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    // Only the valid destination should produce output
+    expect(out).toHaveLength(1);
+    expect(JSON.parse(out[0].content).text).toBe('delivered');
+    expect(out[0].platform_id).toBe('chan-1');
+
+    await loopPromise.catch(() => {});
+  });
+
+  it('multiple <message> blocks each produce an outbound message', async () => {
+    getInboundDb()
+      .prepare(
+        `INSERT INTO destinations (name, display_name, type, channel_type, platform_id, agent_group_id)
+         VALUES ('slack-test', 'Slack Test', 'channel', 'slack', 'chan-2', NULL)`,
+      )
+      .run();
+
+    insertMessage('m1', { sender: 'Alice', text: 'broadcast' }, { platformId: 'chan-1', channelType: 'discord' });
+
+    const provider = new MockProvider(
+      {},
+      () => '<message to="discord-test">for discord</message><message to="slack-test">for slack</message>',
+    );
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length >= 2, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(2);
+    const discord = out.find((m) => m.platform_id === 'chan-1');
+    const slack = out.find((m) => m.platform_id === 'chan-2');
+    expect(discord).toBeDefined();
+    expect(JSON.parse(discord!.content).text).toBe('for discord');
+    expect(slack).toBeDefined();
+    expect(JSON.parse(slack!.content).text).toBe('for slack');
+
+    await loopPromise.catch(() => {});
+  });
+
+  it('sends null thread_id when no prior inbound from destination', async () => {
+    // Seed a second destination that has NO inbound messages
+    getInboundDb()
+      .prepare(
+        `INSERT INTO destinations (name, display_name, type, channel_type, platform_id, agent_group_id)
+         VALUES ('slack-new', 'Slack New', 'channel', 'slack', 'chan-new', NULL)`,
+      )
+      .run();
+
+    // Only insert a message from discord — slack-new has never sent anything
+    insertMessage('m1', { sender: 'Alice', text: 'tell slack' }, { platformId: 'chan-1', channelType: 'discord', threadId: 'discord-thread' });
+
+    const provider = new MockProvider({}, () => '<message to="slack-new">hello slack</message>');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(out[0].platform_id).toBe('chan-new');
+    expect(out[0].thread_id).toBeNull();
+
+    await loopPromise.catch(() => {});
+  });
+
+  it('resolves most recent thread_id when destination has multiple inbound messages', async () => {
+    // Two messages from same destination, different threads
+    insertMessage('m-old', { sender: 'Alice', text: 'old' }, { platformId: 'chan-1', channelType: 'discord', threadId: 'thread-old' });
+    insertMessage('m-new', { sender: 'Alice', text: 'new' }, { platformId: 'chan-1', channelType: 'discord', threadId: 'thread-new' });
+
+    const provider = new MockProvider({}, () => '<message to="discord-test">reply</message>');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(out[0].thread_id).toBe('thread-new');
+    expect(out[0].in_reply_to).toBe('m-new');
+
+    await loopPromise.catch(() => {});
+  });
+
  it('should process messages arriving after loop starts', async () => {
    const provider = new MockProvider({}, () => '<message to="discord-test">Processed</message>');
    const controller = new AbortController();
@@ -91,6 +249,52 @@ describe('poll loop integration', () => {

    await loopPromise.catch(() => {});
  });
+
+  it('internal tags between message blocks are stripped from scratchpad', async () => {
+    insertMessage('m1', { sender: 'Alice', text: 'hi' }, { platformId: 'chan-1', channelType: 'discord' });
+
+    const provider = new MockProvider(
+      {},
+      () => '<internal>thinking about this...</internal><message to="discord-test">answer</message><internal>done thinking</internal>',
+    );
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(JSON.parse(out[0].content).text).toBe('answer');
+
+    await loopPromise.catch(() => {});
+  });
+
+  it('handles mixed task + chat batch with correct origin metadata', async () => {
+    // Seed destination for routing lookup
+    insertMessage('m-chat', { sender: 'Alice', text: 'check this' }, { platformId: 'chan-1', channelType: 'discord' });
+    // Task with same routing — simulates a scheduled task in a channel session
+    getInboundDb()
+      .prepare(
+        `INSERT INTO messages_in (id, kind, timestamp, status, platform_id, channel_type, content)
+         VALUES ('t-task', 'task', datetime('now'), 'pending', 'chan-1', 'discord', ?)`,
+      )
+      .run(JSON.stringify({ prompt: 'daily check' }));
+
+    const provider = new MockProvider({}, () => '<message to="discord-test">done</message>');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(out[0].platform_id).toBe('chan-1');
+
+    await loopPromise.catch(() => {});
+  });
+
 });

 // Helper: run poll loop until aborted or timeout
@@ -98,6 +302,7 @@ async function runPollLoopWithTimeout(provider: MockProvider, signal: AbortSigna
  return Promise.race([
    runPollLoop({
      provider,
+      providerName: 'mock',
      cwd: '/tmp',
    }),
    new Promise<void>((_, reject) => {
@@ -118,3 +323,142 @@ async function waitFor(condition: () => boolean, timeoutMs: number): Promise<voi
 function sleep(ms: number): Promise<void> {
  return new Promise((resolve) => setTimeout(resolve, ms));
 }
+
+describe('poll loop — provider error recovery', () => {
+  it('writes error to outbound and continues loop on provider throw', async () => {
+    insertMessage('m1', { sender: 'Alice', text: 'trigger error' }, { platformId: 'chan-1', channelType: 'discord' });
+
+    const provider = new ThrowingProvider('API rate limit exceeded');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider as unknown as MockProvider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(JSON.parse(out[0].content).text).toContain('Error:');
+    expect(JSON.parse(out[0].content).text).toContain('API rate limit exceeded');
+
+    // Input message should be marked completed despite the error
+    const pending = getPendingMessages();
+    expect(pending).toHaveLength(0);
+
+    await loopPromise.catch(() => {});
+  });
+});
+
+describe('poll loop — stale session recovery', () => {
+  it('clears continuation when provider reports session invalid', async () => {
+    // Pre-seed a continuation so the local variable in runPollLoop is set.
+    // Without this, the `if (continuation && isSessionInvalid)` check skips.
+    setContinuation('mock', 'pre-existing-session');
+
+    insertMessage('m1', { sender: 'Alice', text: 'stale session' }, { platformId: 'chan-1', channelType: 'discord' });
+
+    const provider = new InvalidSessionProvider();
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider as unknown as MockProvider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    // Error was written to outbound
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(JSON.parse(out[0].content).text).toContain('Error:');
+
+    // Continuation was cleared (isSessionInvalid returned true)
+    expect(getContinuation('mock')).toBeUndefined();
+
+    await loopPromise.catch(() => {});
+  });
+});
+
+describe('poll loop — /clear command', () => {
+  it('clears session, writes confirmation, skips query', async () => {
+    // Seed a continuation so we can verify it gets cleared
+    setContinuation('mock', 'existing-session-id');
+    expect(getContinuation('mock')).toBe('existing-session-id');
+
+    // Insert a /clear command
+    getInboundDb()
+      .prepare(
+        `INSERT INTO messages_in (id, kind, timestamp, status, platform_id, channel_type, content)
+         VALUES ('m-clear', 'chat', datetime('now'), 'pending', 'chan-1', 'discord', ?)`,
+      )
+      .run(JSON.stringify({ text: '/clear' }));
+
+    const provider = new MockProvider({}, () => '<message to="discord-test">should not run</message>');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 2000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 2000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(JSON.parse(out[0].content).text).toBe('Session cleared.');
+
+    // Continuation was cleared
+    expect(getContinuation('mock')).toBeUndefined();
+
+    // Command message was completed
+    const pending = getPendingMessages();
+    expect(pending).toHaveLength(0);
+
+    await loopPromise.catch(() => {});
+  });
+});
+
+/**
+ * Provider that throws on every query, simulating API failures.
+ */
+class ThrowingProvider {
+  readonly supportsNativeSlashCommands = false;
+  private errorMessage: string;
+
+  constructor(errorMessage: string) {
+    this.errorMessage = errorMessage;
+  }
+
+  isSessionInvalid(): boolean {
+    return false;
+  }
+
+  query(_input: { prompt: string; cwd: string }) {
+    const errorMessage = this.errorMessage;
+    return {
+      push() {},
+      end() {},
+      abort() {},
+      events: (async function* () {
+        throw new Error(errorMessage);
+      })(),
+    };
+  }
+}
+
+/**
+ * Provider that throws with an error that triggers isSessionInvalid.
+ * First emits an init event (setting continuation), then throws.
+ */
+class InvalidSessionProvider {
+  readonly supportsNativeSlashCommands = false;
+
+  isSessionInvalid(): boolean {
+    return true;
+  }
+
+  query(_input: { prompt: string; cwd: string }) {
+    return {
+      push() {},
+      end() {},
+      abort() {},
+      events: (async function* () {
+        yield { type: 'init' as const, continuation: 'doomed-session' };
+        throw new Error('session not found');
+      })(),
+    };
+  }
+}
@@ -0,0 +1,83 @@
+## Admin CLI (`ncl`)
+
+The `ncl` command is available at `/usr/local/bin/ncl`. It lets you query and modify NanoClaw's central configuration.
+
+### Usage
+
+```
+ncl <resource> <verb> [--flags]
+ncl <resource> help
+ncl help
+```
+
+### Scope
+
+Your CLI access may be scoped. Run `ncl help` to see which resources are available and whether args are auto-filled. Under `group` scope (the default), `--id` and group-related args are auto-filled to your agent group — you don't need to pass them.
+
+### Resources
+
+Run `ncl help` for the full list. Common resources:
+
+| Resource | Verbs | What it is |
+|----------|-------|------------|
+| groups | list, get, create, update, delete, restart, config get/update, config add-mcp-server/remove-mcp-server, config add-package/remove-package | Agent groups (workspace, personality, container config) |
+| sessions | list, get | Active sessions (read-only) |
+| destinations | list, add, remove | Where an agent group can send messages |
+| members | list, add, remove | Unprivileged access gate for an agent group |
+
+Additional resources (available under `global` scope only): messaging-groups, wirings, users, roles, user-dms, dropped-messages, approvals.
+
+### When to use
+
+- **Looking up your own config** — `ncl groups get` or `ncl groups config get` to see your container config.
+- **Restarting your container** — `ncl groups restart` (with optional `--rebuild` and `--message`).
+- **Checking who's in your group** — `ncl members list`.
+- **Seeing your destinations** — `ncl destinations list`.
+- **Answering questions about the system** — query `ncl` rather than guessing.
+
+### Access rules
+
+Read commands (list, get) are open. Write commands (create, update, delete, restart, config update, add, remove) require admin approval — the request is held until an admin approves it.
+
+### Approval flow
+
+Write commands require admin approval. Here's what happens:
+
+1. You run the command (e.g. `ncl groups config update --model claude-sonnet-4-5-20250514`).
+2. The command returns immediately with an `approval-pending` response — it has **not** been executed yet.
+3. An admin or owner gets a notification showing exactly what you requested, with approve/reject options.
+4. Once the admin responds:
+   - **Approved:** the command executes and the result is delivered back to you as a system message in this conversation.
+   - **Rejected:** you get a system message saying the request was rejected.
+
+You don't need to poll or retry — the result arrives automatically.
+
+### Examples
+
+```bash
+# Read commands (no approval needed)
+ncl groups get
+ncl groups config get
+ncl sessions list
+ncl destinations list
+ncl members list
+
+# Write commands (approval required)
+ncl groups restart
+ncl groups restart --rebuild --message "Config updated."
+ncl groups config update --model claude-sonnet-4-5-20250514
+ncl groups config add-mcp-server --name rss --command npx --args '["some-rss-mcp"]'
+ncl groups config add-package --npm some-package
+ncl members add --user telegram:jane
+```
+
+### Important
+
+Config changes via `ncl groups config update` do not take effect until `ncl groups restart`. Run `ncl groups config help` for details.
+
+### Tips
+
+- Use `ncl <resource> help` to see all available fields, types, enums, and which fields are auto-filled.
+- Flags use `--hyphen-case` (e.g. `--agent-group-id`), mapped to `underscore_case` DB columns automatically.
+- `list` supports filtering by any non-auto column. Default limit is 200 rows; override with `--limit N`.
+- Write commands return `approval-pending` immediately — don't treat this as an error. Wait for the system message with the result.
@@ -1,6 +1,6 @@
 ## Sending messages

-Your final response is delivered via the `## Sending messages` rules in your runtime system prompt (single-destination: just write; multi-destination: use `<message to="name">...</message>` blocks). See that section for the current destination list.
+**Every response** must be wrapped in `<message to="name">...</message>` blocks — even if you only have one destination. Bare text outside of `<message>` blocks is scratchpad (logged but never sent). See the `## Sending messages` section in your runtime system prompt for the current destination list and names.

 ### Mid-turn updates (`send_message`)

@@ -0,0 +1,50 @@
+/**
+ * Tests for the core MCP tools' interaction with the per-batch routing
+ * context. The agent-runner sets a current `inReplyTo` at the top of each
+ * batch in poll-loop, and outbound writes from MCP tools (send_message,
+ * send_file) must pick it up so a2a return-path routing on the host can
+ * correlate replies back to the originating session.
+ */
+import { describe, it, expect, beforeEach, afterEach } from 'bun:test';
+
+import { initTestSessionDb, closeSessionDb, getInboundDb } from '../db/connection.js';
+import { getUndeliveredMessages } from '../db/messages-out.js';
+import { setCurrentInReplyTo, clearCurrentInReplyTo } from '../current-batch.js';
+import { sendMessage } from './core.js';
+
+beforeEach(() => {
+  initTestSessionDb();
+  // Seed a peer agent destination
+  getInboundDb()
+    .prepare(
+      `INSERT INTO destinations (name, display_name, type, channel_type, platform_id, agent_group_id)
+       VALUES ('peer', 'Peer', 'agent', NULL, NULL, 'ag-peer')`,
+    )
+    .run();
+});
+
+afterEach(() => {
+  clearCurrentInReplyTo();
+  closeSessionDb();
+});
+
+describe('send_message MCP tool — in_reply_to plumbing', () => {
+  it('stamps current batch in_reply_to on outbound rows', async () => {
+    setCurrentInReplyTo('inbound-msg-1');
+
+    await sendMessage.handler({ to: 'peer', text: 'hello' });
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(out[0].in_reply_to).toBe('inbound-msg-1');
+  });
+
+  it('writes null when no batch is active', async () => {
+    // No setCurrentInReplyTo before this call — simulates ad-hoc / out-of-batch invocation.
+    await sendMessage.handler({ to: 'peer', text: 'hello' });
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    expect(out[0].in_reply_to).toBeNull();
+  });
+});
@@ -9,6 +9,7 @@
 import fs from 'fs';
 import path from 'path';

+import { getCurrentInReplyTo } from '../current-batch.js';
 import { findByName, getAllDestinations } from '../destinations.js';
 import { getMessageIdBySeq, getRoutingBySeq, writeMessageOut } from '../db/messages-out.js';
 import { getSessionRouting } from '../db/session-routing.js';
@@ -50,9 +51,7 @@ function destinationList(): string {
 */
 function resolveRouting(
  to: string | undefined,
-):
-  | { channel_type: string; platform_id: string; thread_id: string | null; resolvedName: string }
-  | { error: string } {
+): { channel_type: string; platform_id: string; thread_id: string | null; resolvedName: string } | { error: string } {
  if (!to) {
    // Default: reply to whatever thread/channel this session is bound to.
    const session = getSessionRouting();
@@ -82,9 +81,7 @@ function resolveRouting(
    // preserve the thread_id so replies land in the correct thread.
    const session = getSessionRouting();
    const threadId =
-      session.channel_type === dest.channelType && session.platform_id === dest.platformId
-        ? session.thread_id
-        : null;
+      session.channel_type === dest.channelType && session.platform_id === dest.platformId ? session.thread_id : null;
    return {
      channel_type: dest.channelType!,
      platform_id: dest.platformId!,
@@ -98,12 +95,14 @@ function resolveRouting(
 export const sendMessage: McpToolDefinition = {
  tool: {
    name: 'send_message',
-    description:
-      'Send a message to a named destination. If you have only one destination, you can omit `to`.',
+    description: 'Send a message to a named destination. If you have only one destination, you can omit `to`.',
    inputSchema: {
      type: 'object' as const,
      properties: {
-        to: { type: 'string', description: 'Destination name (e.g., "family", "worker-1"). Optional if you have only one destination.' },
+        to: {
+          type: 'string',
+          description: 'Destination name (e.g., "family", "worker-1"). Optional if you have only one destination.',
+        },
        text: { type: 'string', description: 'Message content' },
      },
      required: ['text'],
@@ -119,6 +118,7 @@ export const sendMessage: McpToolDefinition = {
    const id = generateId();
    const seq = writeMessageOut({
      id,
+      in_reply_to: getCurrentInReplyTo(),
      kind: 'chat',
      platform_id: routing.platform_id,
      channel_type: routing.channel_type,
@@ -165,6 +165,7 @@ export const sendFile: McpToolDefinition = {

    writeMessageOut({
      id,
+      in_reply_to: getCurrentInReplyTo(),
      kind: 'chat',
      platform_id: routing.platform_id,
      channel_type: routing.channel_type,
@@ -89,6 +89,9 @@ export const scheduleTask: McpToolDefinition = {
        script,
        processAfter,
        recurrence,
+        platformId: r.platform_id,
+        channelType: r.channel_type,
+        threadId: r.thread_id,
      }),
    });

@@ -22,4 +22,4 @@ Use **`add_mcp_server`** to add an MCP server to your configuration. Browse avai
 add_mcp_server({ name: "memory", command: "pnpm", args: ["dlx", "@modelcontextprotocol/server-memory"] })
 ```

-Do not ask the user to give you credentials. Credentials are managed by the user in the OneCLI agent vault. Add a "placeholder" string instead of the credential, and ask the user to add the credential to the vault. You can make a test request before the secret is added and the vault proxy will respond with the local url of the vault dashboard on the user's machine and a link to a form for adding that specific credential.
+Do not ask the user to give you credentials or tell them how to create credentials (OAuth, API keys, etc.) — NEVER fabricate credential setup instructions. Credentials are handled by the OneCLI gateway. Use `"onecli-managed"` as the placeholder value for any credential env vars or config fields. After the MCP server is installed and the container restarts, load `/onecli-gateway` for the full credential-handling flow (connect URLs, stubs, error recovery).
@@ -4,6 +4,7 @@ import { initTestSessionDb, closeSessionDb, getInboundDb, getOutboundDb } from '
 import { getPendingMessages, markCompleted } from './db/messages-in.js';
 import { getUndeliveredMessages } from './db/messages-out.js';
 import { formatMessages, extractRouting } from './formatter.js';
+import { isCorruptionError } from './poll-loop.js';
 import { MockProvider } from './providers/mock.js';

 beforeEach(() => {
@@ -14,13 +15,18 @@ afterEach(() => {
  closeSessionDb();
 });

-function insertMessage(id: string, kind: string, content: object, opts?: { processAfter?: string; trigger?: 0 | 1 }) {
+function insertMessage(
+  id: string,
+  kind: string,
+  content: object,
+  opts?: { processAfter?: string; trigger?: 0 | 1; onWake?: 0 | 1 },
+) {
  getInboundDb()
    .prepare(
-      `INSERT INTO messages_in (id, kind, timestamp, status, process_after, trigger, content)
-     VALUES (?, ?, datetime('now'), 'pending', ?, ?, ?)`,
+      `INSERT INTO messages_in (id, kind, timestamp, status, process_after, trigger, on_wake, content)
+     VALUES (?, ?, datetime('now'), 'pending', ?, ?, ?, ?)`,
    )
-    .run(id, kind, opts?.processAfter ?? null, opts?.trigger ?? 1, JSON.stringify(content));
+    .run(id, kind, opts?.processAfter ?? null, opts?.trigger ?? 1, opts?.onWake ?? 0, JSON.stringify(content));
 }

 describe('formatter', () => {
@@ -32,13 +38,15 @@ describe('formatter', () => {
    expect(prompt).toContain('Hello world');
  });

-  it('should format multiple chat messages as XML block', () => {
+  it('should format multiple chat messages as distinct <message> blocks', () => {
    insertMessage('m1', 'chat', { sender: 'John', text: 'Hello' });
    insertMessage('m2', 'chat', { sender: 'Jane', text: 'Hi there' });
    const messages = getPendingMessages();
    const prompt = formatMessages(messages);
-    expect(prompt).toContain('<messages>');
-    expect(prompt).toContain('</messages>');
+    // The <messages> envelope was dropped in fe2e881b (#2556) so the SDK calls
+    // the API; each message is now its own self-contained <message> block.
+    expect(prompt).not.toContain('<messages>');
+    expect(prompt.match(/<message /g) ?? []).toHaveLength(2);
    expect(prompt).toContain('sender="John"');
    expect(prompt).toContain('sender="Jane"');
  });
@@ -47,7 +55,7 @@ describe('formatter', () => {
    insertMessage('m1', 'task', { prompt: 'Review open PRs' });
    const messages = getPendingMessages();
    const prompt = formatMessages(messages);
-    expect(prompt).toContain('[SCHEDULED TASK]');
+    expect(prompt).toContain('<task');
    expect(prompt).toContain('Review open PRs');
  });

@@ -55,15 +63,17 @@ describe('formatter', () => {
    insertMessage('m1', 'webhook', { source: 'github', event: 'push', payload: { ref: 'main' } });
    const messages = getPendingMessages();
    const prompt = formatMessages(messages);
-    expect(prompt).toContain('[WEBHOOK: github/push]');
+    expect(prompt).toContain('<webhook');
+    expect(prompt).toContain('source="github"');
+    expect(prompt).toContain('event="push"');
  });

  it('should format system messages', () => {
    insertMessage('m1', 'system', { action: 'register_group', status: 'success', result: { id: 'ag-1' } });
    const messages = getPendingMessages();
    const prompt = formatMessages(messages);
-    expect(prompt).toContain('[SYSTEM RESPONSE]');
-    expect(prompt).toContain('register_group');
+    expect(prompt).toContain('<system_response');
+    expect(prompt).toContain('action="register_group"');
  });

  it('should handle mixed kinds', () => {
@@ -72,7 +82,7 @@ describe('formatter', () => {
    const messages = getPendingMessages();
    const prompt = formatMessages(messages);
    expect(prompt).toContain('sender="John"');
-    expect(prompt).toContain('[SYSTEM RESPONSE]');
+    expect(prompt).toContain('<system_response');
  });

  it('should escape XML in content', () => {
@@ -129,6 +139,58 @@ describe('accumulate gate (trigger column)', () => {
  });
 });

+describe('on_wake filtering', () => {
+  it('first poll returns on_wake=1 messages', () => {
+    insertMessage('m1', 'chat', { sender: 'system', text: 'Resuming.' }, { onWake: 1 });
+    const messages = getPendingMessages(true);
+    expect(messages).toHaveLength(1);
+    expect(messages[0].id).toBe('m1');
+  });
+
+  it('subsequent polls skip on_wake=1 messages', () => {
+    insertMessage('m1', 'chat', { sender: 'system', text: 'Resuming.' }, { onWake: 1 });
+    const messages = getPendingMessages(false);
+    expect(messages).toHaveLength(0);
+  });
+
+  it('normal messages returned regardless of isFirstPoll', () => {
+    insertMessage('m1', 'chat', { sender: 'A', text: 'hello' });
+    expect(getPendingMessages(true)).toHaveLength(1);
+
+    // Reset: mark completed so we can re-test with a fresh message
+    markCompleted(['m1']);
+    insertMessage('m2', 'chat', { sender: 'A', text: 'hello again' });
+    expect(getPendingMessages(false)).toHaveLength(1);
+  });
+
+  it('mixed batch: first poll returns both normal and on_wake messages', () => {
+    insertMessage('m1', 'chat', { sender: 'A', text: 'user msg' });
+    insertMessage('m2', 'chat', { sender: 'system', text: 'Resuming.' }, { onWake: 1 });
+    const messages = getPendingMessages(true);
+    expect(messages).toHaveLength(2);
+    expect(messages.map((m) => m.id).sort()).toEqual(['m1', 'm2']);
+  });
+
+  it('mixed batch: subsequent poll returns only normal messages', () => {
+    insertMessage('m1', 'chat', { sender: 'A', text: 'user msg' });
+    insertMessage('m2', 'chat', { sender: 'system', text: 'Resuming.' }, { onWake: 1 });
+    const messages = getPendingMessages(false);
+    expect(messages).toHaveLength(1);
+    expect(messages[0].id).toBe('m1');
+  });
+
+  it('on_wake defaults to 0 for inserts without explicit value', () => {
+    getInboundDb()
+      .prepare(
+        `INSERT INTO messages_in (id, kind, timestamp, status, content)
+         VALUES ('m1', 'chat', datetime('now'), 'pending', '{"text":"hi"}')`,
+      )
+      .run();
+    // Should be returned even on non-first poll (on_wake=0)
+    expect(getPendingMessages(false)).toHaveLength(1);
+  });
+});
+
 describe('routing', () => {
  it('should extract routing from messages', () => {
    getInboundDb()
@@ -147,6 +209,76 @@ describe('routing', () => {
  });
 });

+describe('origin metadata (from= attribute)', () => {
+  function seedDestination(name: string, channelType: string, platformId: string): void {
+    getInboundDb()
+      .prepare(
+        `INSERT INTO destinations (name, display_name, type, channel_type, platform_id, agent_group_id)
+         VALUES (?, ?, 'channel', ?, ?, NULL)`,
+      )
+      .run(name, name, channelType, platformId);
+  }
+
+  function insertWithRouting(id: string, kind: string, content: object, channelType: string | null, platformId: string | null): void {
+    getInboundDb()
+      .prepare(
+        `INSERT INTO messages_in (id, kind, timestamp, status, platform_id, channel_type, content)
+         VALUES (?, ?, datetime('now'), 'pending', ?, ?, ?)`,
+      )
+      .run(id, kind, platformId, channelType, JSON.stringify(content));
+  }
+
+  it('chat message includes from= when destination matches', () => {
+    seedDestination('discord-main', 'discord', 'chan-1');
+    insertWithRouting('m1', 'chat', { sender: 'Alice', text: 'hi' }, 'discord', 'chan-1');
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).toContain('from="discord-main"');
+  });
+
+  it('chat message falls back to raw routing when no destination matches', () => {
+    insertWithRouting('m1', 'chat', { sender: 'Alice', text: 'hi' }, 'telegram', 'chat-999');
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).toContain('from="unknown:telegram:chat-999"');
+  });
+
+  it('chat message omits from= when routing is null', () => {
+    insertMessage('m1', 'chat', { sender: 'Alice', text: 'hi' });
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).not.toContain('from=');
+  });
+
+  it('task message includes from= when destination matches', () => {
+    seedDestination('slack-ops', 'slack', 'C-OPS');
+    insertWithRouting('t1', 'task', { prompt: 'check status' }, 'slack', 'C-OPS');
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).toContain('<task');
+    expect(prompt).toContain('from="slack-ops"');
+  });
+
+  it('task message omits from= when routing is null', () => {
+    insertMessage('t1', 'task', { prompt: 'check status' });
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).toContain('<task');
+    expect(prompt).not.toContain('from=');
+  });
+
+  it('webhook message includes from= when destination matches', () => {
+    seedDestination('github-ch', 'github', 'repo-1');
+    insertWithRouting('w1', 'webhook', { source: 'github', event: 'push', payload: {} }, 'github', 'repo-1');
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).toContain('<webhook');
+    expect(prompt).toContain('from="github-ch"');
+  });
+
+  it('system message includes from= when destination matches', () => {
+    seedDestination('discord-main', 'discord', 'chan-1');
+    insertWithRouting('s1', 'system', { action: 'test', status: 'ok', result: null }, 'discord', 'chan-1');
+    const prompt = formatMessages(getPendingMessages());
+    expect(prompt).toContain('<system_response');
+    expect(prompt).toContain('from="discord-main"');
+  });
+});
+
 describe('mock provider', () => {
  it('should produce init + result events', async () => {
    const provider = new MockProvider({}, (prompt) => `Echo: ${prompt}`);
@@ -246,3 +378,20 @@ describe('end-to-end with mock provider', () => {
    expect(outMessages[0].in_reply_to).toBe('m1');
  });
 });
+
+describe('isCorruptionError', () => {
+  it('matches the Docker Desktop macOS torn-read symptom', () => {
+    expect(isCorruptionError('database disk image is malformed')).toBe(true);
+  });
+
+  it('matches wrapped SQLite corruption codes', () => {
+    expect(isCorruptionError('SqliteError: SQLITE_CORRUPT_VTAB: ...')).toBe(true);
+    expect(isCorruptionError('file is not a database')).toBe(true);
+  });
+
+  it('returns false for unrelated errors', () => {
+    expect(isCorruptionError('database is locked')).toBe(false);
+    expect(isCorruptionError('no such table: messages_in')).toBe(false);
+    expect(isCorruptionError('')).toBe(false);
+  });
+});
@@ -1,14 +1,48 @@
 import { findByName, getAllDestinations, type DestinationEntry } from './destinations.js';
 import { getPendingMessages, markProcessing, markCompleted, type MessageInRow } from './db/messages-in.js';
 import { writeMessageOut } from './db/messages-out.js';
-import { touchHeartbeat, clearStaleProcessingAcks } from './db/connection.js';
-import { getStoredSessionId, setStoredSessionId, clearStoredSessionId } from './db/session-state.js';
-import { formatMessages, extractRouting, categorizeMessage, isClearCommand, stripInternalTags, type RoutingContext } from './formatter.js';
+import { getInboundDb, touchHeartbeat, clearStaleProcessingAcks } from './db/connection.js';
+import { clearContinuation, migrateLegacyContinuation, setContinuation } from './db/session-state.js';
+import { clearCurrentInReplyTo, setCurrentInReplyTo } from './current-batch.js';
+import {
+  formatMessages,
+  extractRouting,
+  categorizeMessage,
+  isClearCommand,
+  isRunnerCommand,
+  stripInternalTags,
+  type RoutingContext,
+} from './formatter.js';
+import { isUploadTraceCommand, uploadTrace } from './upload-trace.js';
 import type { AgentProvider, AgentQuery, ProviderEvent } from './providers/types.js';

 const POLL_INTERVAL_MS = 1000;
 const ACTIVE_POLL_INTERVAL_MS = 500;

+/**
+ * Number of consecutive `database disk image is malformed` errors after which
+ * the follow-up poll gives up and exits the process. At ACTIVE_POLL_INTERVAL_MS
+ * = 500ms this is roughly 5 seconds — long enough to dodge a transient torn
+ * read during a host write, short enough to recover quickly from a poisoned
+ * page cache (host-sweep then respawns with a fresh mount).
+ */
+const CORRUPTION_STREAK_EXIT = 10;
+
+/**
+ * True for SQLite errors that indicate a corrupt READ view — almost always a
+ * cross-mount page-cache coherency issue on Docker Desktop macOS rather than
+ * actual file damage (host-side integrity_check passes). Reopening the DB
+ * handle inside this process does NOT recover; only a fresh container mount
+ * does. Caller's job is to exit so host-sweep respawns the container.
+ */
+export function isCorruptionError(msg: string): boolean {
+  return (
+    msg.includes('database disk image is malformed') ||
+    msg.includes('SQLITE_CORRUPT') ||
+    msg.includes('file is not a database')
+  );
+}
+
 function log(msg: string): void {
  console.error(`[poll-loop] ${msg}`);
 }
@@ -19,6 +53,12 @@ function generateId(): string {

 export interface PollLoopConfig {
  provider: AgentProvider;
+  /**
+   * Name of the provider (e.g. "claude", "codex", "opencode"). Used to key
+   * the stored continuation per-provider so flipping providers doesn't
+   * resurrect a stale id from a different backend.
+   */
+  providerName: string;
  cwd: string;
  systemContext?: {
    instructions?: string;
@@ -39,8 +79,22 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
  // Resume the agent's prior session from a previous container run if one
  // was persisted. The continuation is opaque to the poll-loop — the
  // provider decides how to use it (Claude resumes a .jsonl transcript,
-  // other providers may reload a thread ID, etc.).
-  let continuation: string | undefined = getStoredSessionId();
+  // other providers may reload a thread ID, etc.). Keyed per-provider so
+  // a Codex thread id never gets handed to Claude or vice versa.
+  let continuation: string | undefined = migrateLegacyContinuation(config.providerName);
+
+  // Before resuming, drop a session whose on-disk transcript has grown too
+  // large/old to cold-resume within the host's idle ceiling. Without this a
+  // long-lived hub keeps trying to reload an ever-growing .jsonl, hangs the
+  // first turn, and gets killed before it can reply (then repeats forever).
+  if (continuation) {
+    const rotateReason = config.provider.maybeRotateContinuation?.(continuation, config.cwd);
+    if (rotateReason) {
+      log(`Rotating session — ${rotateReason}; starting fresh`);
+      clearContinuation(config.providerName);
+      continuation = undefined;
+    }
+  }

  if (continuation) {
    log(`Resuming agent session ${continuation}`);
@@ -51,9 +105,11 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
  clearStaleProcessingAcks();

  let pollCount = 0;
+  let isFirstPoll = true;
  while (true) {
    // Skip system messages — they're responses for MCP tools (e.g., ask_user_question)
-    const messages = getPendingMessages().filter((m) => m.kind !== 'system');
+    const messages = getPendingMessages(isFirstPoll).filter((m) => m.kind !== 'system');
+    isFirstPoll = false;
    pollCount++;

    // Periodic heartbeat so we know the loop is alive
@@ -94,7 +150,7 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
      if ((msg.kind === 'chat' || msg.kind === 'chat-sdk') && isClearCommand(msg)) {
        log('Clearing session (resetting continuation)');
        continuation = undefined;
-        clearStoredSessionId();
+        clearContinuation(config.providerName);
        writeMessageOut({
          id: generateId(),
          kind: 'chat',
@@ -106,6 +162,19 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
        commandIds.push(msg.id);
        continue;
      }
+      if ((msg.kind === 'chat' || msg.kind === 'chat-sdk') && isUploadTraceCommand(msg)) {
+        log('Uploading session trace to Hugging Face');
+        writeMessageOut({
+          id: generateId(),
+          kind: 'chat',
+          platform_id: routing.platformId,
+          channel_type: routing.channelType,
+          thread_id: routing.threadId,
+          content: JSON.stringify({ text: uploadTrace() }),
+        });
+        commandIds.push(msg.id);
+        continue;
+      }
      normalMessages.push(msg);
    }

@@ -159,11 +228,14 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
    // Process the query while concurrently polling for new messages
    const skippedSet = new Set(skipped);
    const processingIds = ids.filter((id) => !commandIds.includes(id) && !skippedSet.has(id));
+    // Publish the batch's in_reply_to so MCP tools (send_message, send_file)
+    // can stamp it on outbound rows — needed for a2a return-path routing.
+    setCurrentInReplyTo(routing.inReplyTo);
    try {
-      const result = await processQuery(query, routing, processingIds);
+      const result = await processQuery(query, routing, processingIds, config.providerName);
      if (result.continuation && result.continuation !== continuation) {
        continuation = result.continuation;
-        setStoredSessionId(continuation);
+        setContinuation(config.providerName, continuation);
      }
    } catch (err) {
      const errMsg = err instanceof Error ? err.message : String(err);
@@ -175,7 +247,7 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
      if (continuation && config.provider.isSessionInvalid(err)) {
        log(`Stale session detected (${continuation}) — clearing for next retry`);
        continuation = undefined;
-        clearStoredSessionId();
+        clearContinuation(config.providerName);
      }

      // Write error response so the user knows something went wrong
@@ -187,6 +259,8 @@ export async function runPollLoop(config: PollLoopConfig): Promise<void> {
        thread_id: routing.threadId,
        content: JSON.stringify({ text: `Error: ${errMsg}` }),
      });
+    } finally {
+      clearCurrentInReplyTo();
    }

    // Ensure completed even if processQuery ended without a result event
@@ -238,41 +312,124 @@ async function processQuery(
  query: AgentQuery,
  routing: RoutingContext,
  initialBatchIds: string[],
+  providerName: string,
 ): Promise<QueryResult> {
  let queryContinuation: string | undefined;
  let done = false;
+  let unwrappedNudged = false;

  // Concurrent polling: push follow-ups into the active query as they arrive.
-  // We do NOT force-end the stream on silence — keeping the query open is
-  // strictly cheaper than close+reopen (no cold prompt cache, no reconnect).
+  // We do NOT force-end the stream on silence — keeping the query open avoids
+  // re-spawning the SDK subprocess (~few seconds) and re-loading the .jsonl
+  // transcript on every turn. The Anthropic prompt cache is server-side with
+  // a 5-min TTL keyed on prefix hash, so stream lifecycle does NOT affect
+  // cache lifetime — close+reopen within 5 min still gets cache hits.
  // Stream liveness is decided host-side via the heartbeat file + processing
  // claim age (see src/host-sweep.ts); if something is truly stuck, the host
  // will kill the container and messages get reset to pending.
+  let pollInFlight = false;
+  let endedForCommand = false;
+  let corruptionStreak = 0;
  const pollHandle = setInterval(() => {
-    if (done) return;
+    if (done || pollInFlight || endedForCommand) return;
+    pollInFlight = true;

-    // Skip system messages (MCP tool responses) and /clear (needs fresh query).
-    // Thread routing is the router's concern — if a message landed in this
-    // session, the agent should see it. Per-thread sessions already isolate
-    // threads into separate containers; shared sessions intentionally merge
-    // everything. Filtering on thread_id here caused deadlocks when the
-    // initial batch and follow-ups had mismatched thread_ids (e.g. a
-    // host-generated welcome trigger with null thread vs a Discord DM reply).
-    const newMessages = getPendingMessages().filter((m) => {
-      if (m.kind === 'system') return false;
-      if ((m.kind === 'chat' || m.kind === 'chat-sdk') && isClearCommand(m)) return false;
-      return true;
-    });
-    if (newMessages.length > 0) {
-      const newIds = newMessages.map((m) => m.id);
-      markProcessing(newIds);
+    void (async () => {
+      try {
+        const pending = getPendingMessages();

-      const prompt = formatMessages(newMessages);
-      log(`Pushing ${newMessages.length} follow-up message(s) into active query`);
-      query.push(prompt);
+        // Slash commands need a fresh query: /clear resets the SDK's
+        // resume id (fixed at sdkQuery() time); admin/passthrough commands
+        // (/compact, /cost, …) only dispatch when they're the first input
+        // of a query — pushed mid-stream they arrive as plain text and
+        // the SDK never runs them. End the stream and leave the rows
+        // pending; the outer loop handles them on next iteration via the
+        // canonical command path + formatMessagesWithCommands.
+        if (pending.some((m) => isRunnerCommand(m))) {
+          log('Pending slash command — ending stream so outer loop can process');
+          endedForCommand = true;
+          query.end();
+          return;
+        }

-      markCompleted(newIds);
-    }
+        // Skip system messages (MCP tool responses).
+        // Thread routing is the router's concern — if a message landed in this
+        // session, the agent should see it. Per-thread sessions already isolate
+        // threads into separate containers; shared sessions intentionally merge
+        // everything. Filtering on thread_id here caused deadlocks when the
+        // initial batch and follow-ups had mismatched thread_ids (e.g. a
+        // host-generated welcome trigger with null thread vs a Discord DM reply).
+        const newMessages = pending.filter((m) => m.kind !== 'system');
+        if (newMessages.length === 0) return;
+
+        const newIds = newMessages.map((m) => m.id);
+        markProcessing(newIds);
+
+        // Run pre-task scripts on follow-ups too — without this, a task that
+        // arrives during an active query (e.g. a */10 monitoring cron) bypasses
+        // its script gate and always wakes the agent, defeating the gate.
+        // Mirrors the initial-batch hook above.
+        let keep = newMessages;
+        let skipped: string[] = [];
+        // MODULE-HOOK:scheduling-pre-task-followup:start
+        const { applyPreTaskScripts } = await import('./scheduling/task-script.js');
+        const preTask = await applyPreTaskScripts(newMessages);
+        keep = preTask.keep;
+        skipped = preTask.skipped;
+        if (skipped.length > 0) {
+          markCompleted(skipped);
+          log(`Pre-task script skipped ${skipped.length} follow-up task(s): ${skipped.join(', ')}`);
+        }
+        // MODULE-HOOK:scheduling-pre-task-followup:end
+
+        if (keep.length === 0) return;
+        // Re-check done — the outer query may have finished while the script
+        // was awaited. Pushing into a closed stream is wasted work; the
+        // claimed messages get released by the host's processing-claim sweep.
+        if (done) return;
+
+        const keptIds = keep.map((m) => m.id);
+        const prompt = formatMessages(keep);
+        log(`Pushing ${keep.length} follow-up message(s) into active query`);
+        unwrappedNudged = false;
+        query.push(prompt);
+        markCompleted(keptIds);
+      } catch (err) {
+        // Without this catch the rejection escapes the void IIFE and Node
+        // terminates the container on unhandled-rejection. The initial-batch
+        // path is wrapped by processQuery's outer try/catch; the follow-up
+        // path is not, so it needs its own.
+        const errMsg = err instanceof Error ? err.message : String(err);
+        log(`Follow-up poll error: ${errMsg}`);
+
+        // Detect SQLite cross-mount corruption (Docker Desktop macOS virtiofs /
+        // gRPC-FUSE coherency bug — the kernel page cache for the inbound.db
+        // bind mount can latch a torn snapshot mid-host-write, after which
+        // every fresh openInboundDb() in this process sees the same broken
+        // view. Reopening inside the container does NOT recover; only a fresh
+        // container mount does. Exit so the host sweep respawns us.
+        if (isCorruptionError(errMsg)) {
+          corruptionStreak += 1;
+          if (corruptionStreak >= CORRUPTION_STREAK_EXIT) {
+            log(
+              `Follow-up poll: ${corruptionStreak} consecutive '${errMsg}' errors — ` +
+                `inbound.db page cache is poisoned. Exiting so host respawns with a fresh mount.`,
+            );
+            // Stop touching the heartbeat so host-sweep stale detection fires
+            // promptly even if exit() races with in-flight async work.
+            done = true;
+            clearInterval(pollHandle);
+            // Defer exit one tick so this log line flushes through Docker's
+            // log driver before the process dies.
+            setTimeout(() => process.exit(75), 100);
+          }
+        } else {
+          corruptionStreak = 0;
+        }
+      } finally {
+        pollInFlight = false;
+      }
+    })();
  }, ACTIVE_POLL_INTERVAL_MS);

  try {
@@ -288,7 +445,7 @@ async function processQuery(
        // container died between `init` and `result`, the SDK session was
        // effectively orphaned and the next message started a blank
        // Claude session with no prior context.
-        setStoredSessionId(event.continuation);
+        setContinuation(providerName, event.continuation);
      } else if (event.type === 'result') {
        // A result — with or without text — means the turn is done. Mark
        // the initial batch completed now so the host sweep doesn't see
@@ -298,7 +455,18 @@ async function processQuery(
        // at all — either way the turn is finished.
        markCompleted(initialBatchIds);
        if (event.text) {
-          dispatchResultText(event.text, routing);
+          const { hasUnwrapped } = dispatchResultText(event.text, routing);
+          if (hasUnwrapped && !unwrappedNudged) {
+            unwrappedNudged = true;
+            const destinations = getAllDestinations();
+            const names = destinations.map((d) => d.name).join(', ');
+            query.push(
+              `<system>Your response was not delivered — it was not wrapped in <message to="name">...</message> blocks. ` +
+                `All output must be wrapped: use <message to="name"> for content to send, or <internal> for scratchpad. ` +
+                `Your destinations: ${names}. ` +
+                `Please re-send your response with the correct wrapping.</system>`,
+            );
+          }
        }
      }
    }
@@ -319,7 +487,9 @@ function handleEvent(event: ProviderEvent, _routing: RoutingContext): void {
      log(`Result: ${event.text ? event.text.slice(0, 200) : '(empty)'}`);
      break;
    case 'error':
-      log(`Error: ${event.message} (retryable: ${event.retryable}${event.classification ? `, ${event.classification}` : ''})`);
+      log(
+        `Error: ${event.message} (retryable: ${event.retryable}${event.classification ? `, ${event.classification}` : ''})`,
+      );
      break;
    case 'progress':
      log(`Progress: ${event.message}`);
@@ -330,16 +500,12 @@ function handleEvent(event: ProviderEvent, _routing: RoutingContext): void {
 /**
 * Parse the agent's final text for <message to="name">...</message> blocks
 * and dispatch each one to its resolved destination. Text outside of blocks
- * (including <internal>...</internal>) is normally scratchpad — logged but
- * not sent.
+ * (including <internal>...</internal>) is scratchpad — logged but not sent.
 *
- * Single-destination shortcut: if the agent has exactly one configured
- * destination AND the output contains zero <message> blocks, the entire
- * cleaned text (with <internal> tags stripped) is sent to that destination.
- * This preserves the simple case of one user on one channel — the agent
- * doesn't need to know about wrapping syntax at all.
+ * The agent must always wrap output in <message to="name">...</message>
+ * blocks, even with a single destination. Bare text is scratchpad only.
 */
-function dispatchResultText(text: string, routing: RoutingContext): void {
+function dispatchResultText(text: string, routing: RoutingContext): { sent: number; hasUnwrapped: boolean } {
  const MESSAGE_RE = /<message\s+to="([^"]+)"\s*>([\s\S]*?)<\/message>/g;

  let match: RegExpExecArray | null;
@@ -370,56 +536,60 @@ function dispatchResultText(text: string, routing: RoutingContext): void {

  const scratchpad = stripInternalTags(scratchpadParts.join(''));

-  // Single-destination shortcut: the agent wrote plain text — send to
-  // the session's originating channel (from session_routing) if available,
-  // otherwise fall back to the single destination.
-  if (sent === 0 && scratchpad) {
-    if (routing.channelType && routing.platformId) {
-      // Reply to the channel/thread the message came from
-      writeMessageOut({
-        id: generateId(),
-        in_reply_to: routing.inReplyTo,
-        kind: 'chat',
-        platform_id: routing.platformId,
-        channel_type: routing.channelType,
-        thread_id: routing.threadId,
-        content: JSON.stringify({ text: scratchpad }),
-      });
-      return;
-    }
-    const all = getAllDestinations();
-    if (all.length === 1) {
-      sendToDestination(all[0], scratchpad, routing);
-      return;
-    }
-  }
-
  if (scratchpad) {
    log(`[scratchpad] ${scratchpad.slice(0, 500)}${scratchpad.length > 500 ? '…' : ''}`);
  }

-  if (sent === 0 && text.trim()) {
+  const hasUnwrapped = sent === 0 && !!scratchpad;
+  if (hasUnwrapped) {
    log(`WARNING: agent output had no <message to="..."> blocks — nothing was sent`);
  }
+  return { sent, hasUnwrapped };
 }

 function sendToDestination(dest: DestinationEntry, body: string, routing: RoutingContext): void {
  const platformId = dest.type === 'channel' ? dest.platformId! : dest.agentGroupId!;
  const channelType = dest.type === 'channel' ? dest.channelType! : 'agent';
-  // Inherit thread_id from the inbound routing context so replies land in the
-  // same thread the conversation is in. For non-threaded adapters the router
-  // strips thread_id at ingest, so this will already be null.
+  // Resolve thread_id per-destination from the most recent inbound message
+  // that came from this same channel+platform. In agent-shared sessions,
+  // different destinations have different thread contexts — using a single
+  // routing.threadId would stamp one channel's thread onto another.
+  const destRouting = resolveDestinationThread(channelType, platformId);
  writeMessageOut({
    id: generateId(),
-    in_reply_to: routing.inReplyTo,
+    in_reply_to: destRouting?.inReplyTo ?? routing.inReplyTo,
    kind: 'chat',
    platform_id: platformId,
    channel_type: channelType,
-    thread_id: routing.threadId,
+    thread_id: destRouting?.threadId ?? null,
    content: JSON.stringify({ text: body }),
  });
 }

+/**
+ * Find the thread_id and message id from the most recent inbound message
+ * matching the given channel+platform. Returns null if no match found.
+ */
+function resolveDestinationThread(
+  channelType: string,
+  platformId: string,
+): { threadId: string | null; inReplyTo: string | null } | null {
+  try {
+    const db = getInboundDb();
+    const row = db
+      .prepare(
+        `SELECT thread_id, id FROM messages_in
+         WHERE channel_type = ? AND platform_id = ?
+         ORDER BY seq DESC LIMIT 1`,
+      )
+      .get(channelType, platformId) as { thread_id: string | null; id: string } | undefined;
+    if (row) return { threadId: row.thread_id, inReplyTo: row.id };
+  } catch (err) {
+    log(`resolveDestinationThread error: ${err instanceof Error ? err.message : String(err)}`);
+  }
+  return null;
+}
+
 function sleep(ms: number): Promise<void> {
  return new Promise((resolve) => setTimeout(resolve, ms));
 }
@@ -0,0 +1,89 @@
+import { describe, it, expect, beforeEach, afterEach } from 'bun:test';
+import fs from 'fs';
+import os from 'os';
+import path from 'path';
+
+import { ClaudeProvider } from './claude.js';
+
+// maybeRotateContinuation guards the cold-resume failure mode: a long-lived
+// session whose on-disk transcript has grown so large (or old) that the SDK
+// can't reload it before the host's idle ceiling kills the container.
+
+let tmp: string;
+let prevHome: string | undefined;
+let prevConv: string | undefined;
+let prevBytes: string | undefined;
+let prevDays: string | undefined;
+
+const PROJECT_DIR = '-workspace-agent';
+const CWD = '/workspace/agent';
+
+function writeTranscript(sessionId: string, bytes: number, firstTs?: string): string {
+  const dir = path.join(tmp, '.claude', 'projects', PROJECT_DIR);
+  fs.mkdirSync(dir, { recursive: true });
+  const p = path.join(dir, `${sessionId}.jsonl`);
+  const first =
+    JSON.stringify({
+      type: 'user',
+      timestamp: firstTs ?? new Date().toISOString(),
+      message: { role: 'user', content: 'hello' },
+    }) + '\n';
+  const filler = 'x'.repeat(Math.max(0, bytes - first.length));
+  fs.writeFileSync(p, first + filler);
+  return p;
+}
+
+beforeEach(() => {
+  tmp = fs.mkdtempSync(path.join(os.tmpdir(), 'claude-rotate-'));
+  prevHome = process.env.HOME;
+  prevConv = process.env.NANOCLAW_CONVERSATIONS_DIR;
+  prevBytes = process.env.CLAUDE_TRANSCRIPT_ROTATE_BYTES;
+  prevDays = process.env.CLAUDE_TRANSCRIPT_ROTATE_AGE_DAYS;
+  process.env.HOME = tmp;
+  delete process.env.CLAUDE_CONFIG_DIR;
+  process.env.NANOCLAW_CONVERSATIONS_DIR = path.join(tmp, 'conversations');
+});
+
+afterEach(() => {
+  const restore = (k: string, v: string | undefined) => (v === undefined ? delete process.env[k] : (process.env[k] = v));
+  restore('HOME', prevHome);
+  restore('NANOCLAW_CONVERSATIONS_DIR', prevConv);
+  restore('CLAUDE_TRANSCRIPT_ROTATE_BYTES', prevBytes);
+  restore('CLAUDE_TRANSCRIPT_ROTATE_AGE_DAYS', prevDays);
+  fs.rmSync(tmp, { recursive: true, force: true });
+});
+
+describe('ClaudeProvider.maybeRotateContinuation', () => {
+  it('keeps a small, recent transcript (returns null, leaves file in place)', () => {
+    process.env.CLAUDE_TRANSCRIPT_ROTATE_BYTES = String(1024 * 1024);
+    const p = writeTranscript('sess-small', 4096);
+    const provider = new ClaudeProvider();
+    expect(provider.maybeRotateContinuation('sess-small', CWD)).toBeNull();
+    expect(fs.existsSync(p)).toBe(true);
+  });
+
+  it('rotates an oversized transcript (returns reason, moves the .jsonl aside)', () => {
+    process.env.CLAUDE_TRANSCRIPT_ROTATE_BYTES = String(64 * 1024);
+    const p = writeTranscript('sess-big', 200 * 1024);
+    const provider = new ClaudeProvider();
+    const reason = provider.maybeRotateContinuation('sess-big', CWD);
+    expect(reason).toContain('MB');
+    expect(fs.existsSync(p)).toBe(false); // original moved out of the resume path
+    const dir = path.dirname(p);
+    expect(fs.readdirSync(dir).some((f) => f.startsWith('sess-big.jsonl.rotated-'))).toBe(true);
+  });
+
+  it('rotates an aged transcript even when small', () => {
+    process.env.CLAUDE_TRANSCRIPT_ROTATE_BYTES = String(1024 * 1024);
+    process.env.CLAUDE_TRANSCRIPT_ROTATE_AGE_DAYS = '7';
+    const old = new Date(Date.now() - 10 * 86400_000).toISOString();
+    writeTranscript('sess-old', 2048, old);
+    const provider = new ClaudeProvider();
+    expect(provider.maybeRotateContinuation('sess-old', CWD)).toContain('d');
+  });
+
+  it('returns null for an unknown session id', () => {
+    const provider = new ClaudeProvider();
+    expect(provider.maybeRotateContinuation('does-not-exist', CWD)).toBeNull();
+  });
+});
@@ -1,4 +1,5 @@
 import fs from 'fs';
+import os from 'os';
 import path from 'path';

 import { query as sdkQuery, type HookCallback, type PreCompactHookInput } from '@anthropic-ai/claude-agent-sdk';
@@ -34,7 +35,11 @@ const SDK_DISALLOWED_TOOLS = [
  'ExitWorktree',
 ];

-// Tool allowlist for NanoClaw agent containers
+// Tool allowlist for NanoClaw agent containers. MCP-tool entries are derived
+// at the call site from the registered `mcpServers` map so that any server
+// added via `add_mcp_server` (or wired in container.json directly) is
+// reachable to the agent — without this, the SDK's allowedTools filter
+// silently drops every MCP namespace not listed here.
 const TOOL_ALLOWLIST = [
  'Bash',
  'Read',
@@ -54,9 +59,15 @@ const TOOL_ALLOWLIST = [
  'ToolSearch',
  'Skill',
  'NotebookEdit',
-  'mcp__nanoclaw__*',
 ];

+// MCP server names are sanitized by the SDK when forming tool prefixes:
+// any character outside [A-Za-z0-9_-] becomes '_'. Mirror that here so our
+// allowlist patterns match what the SDK actually exposes.
+function mcpAllowPattern(serverName: string): string {
+  return `mcp__${serverName.replace(/[^a-zA-Z0-9_-]/g, '_')}__*`;
+}
+
 interface SDKUserMessage {
  type: 'user';
  message: { role: 'user'; content: string };
@@ -178,56 +189,137 @@ const postToolUseHook: HookCallback = async () => {
  return { continue: true };
 };

+/**
+ * Read a Claude transcript .jsonl, render a markdown summary, and drop it into
+ * the agent's `conversations/` folder so context survives a compaction or a
+ * session rotation. Best-effort: returns false (and logs) on any failure.
+ */
+function archiveTranscriptFile(transcriptPath: string | undefined, sessionId: string | undefined, assistantName?: string): boolean {
+  if (!transcriptPath || !fs.existsSync(transcriptPath)) {
+    log('No transcript found for archiving');
+    return false;
+  }
+
+  try {
+    const content = fs.readFileSync(transcriptPath, 'utf-8');
+    const messages = parseTranscript(content);
+    if (messages.length === 0) return false;
+
+    // Try to get summary from sessions index
+    let summary: string | undefined;
+    const indexPath = path.join(path.dirname(transcriptPath), 'sessions-index.json');
+    if (fs.existsSync(indexPath)) {
+      try {
+        const index = JSON.parse(fs.readFileSync(indexPath, 'utf-8'));
+        summary = index.entries?.find((e: { sessionId: string; summary?: string }) => e.sessionId === sessionId)?.summary;
+      } catch {
+        /* ignore */
+      }
+    }
+
+    const name = summary
+      ? summary.toLowerCase().replace(/[^a-z0-9]+/g, '-').replace(/^-+|-+$/g, '').slice(0, 50)
+      : `conversation-${new Date().getHours().toString().padStart(2, '0')}${new Date().getMinutes().toString().padStart(2, '0')}`;
+
+    const conversationsDir = process.env.NANOCLAW_CONVERSATIONS_DIR || '/workspace/agent/conversations';
+    fs.mkdirSync(conversationsDir, { recursive: true });
+    const filename = `${new Date().toISOString().split('T')[0]}-${name}.md`;
+    fs.writeFileSync(path.join(conversationsDir, filename), formatTranscriptMarkdown(messages, summary, assistantName));
+    log(`Archived conversation to ${filename}`);
+    return true;
+  } catch (err) {
+    log(`Failed to archive transcript: ${err instanceof Error ? err.message : String(err)}`);
+    return false;
+  }
+}
+
 function createPreCompactHook(assistantName?: string): HookCallback {
  return async (input) => {
    const preCompact = input as PreCompactHookInput;
-    const { transcript_path: transcriptPath, session_id: sessionId } = preCompact;
-
-    if (!transcriptPath || !fs.existsSync(transcriptPath)) {
-      log('No transcript found for archiving');
-      return {};
-    }
-
-    try {
-      const content = fs.readFileSync(transcriptPath, 'utf-8');
-      const messages = parseTranscript(content);
-      if (messages.length === 0) return {};
-
-      // Try to get summary from sessions index
-      let summary: string | undefined;
-      const indexPath = path.join(path.dirname(transcriptPath), 'sessions-index.json');
-      if (fs.existsSync(indexPath)) {
-        try {
-          const index = JSON.parse(fs.readFileSync(indexPath, 'utf-8'));
-          summary = index.entries?.find((e: { sessionId: string; summary?: string }) => e.sessionId === sessionId)?.summary;
-        } catch {
-          /* ignore */
-        }
-      }
-
-      const name = summary
-        ? summary.toLowerCase().replace(/[^a-z0-9]+/g, '-').replace(/^-+|-+$/g, '').slice(0, 50)
-        : `conversation-${new Date().getHours().toString().padStart(2, '0')}${new Date().getMinutes().toString().padStart(2, '0')}`;
-
-      const conversationsDir = '/workspace/agent/conversations';
-      fs.mkdirSync(conversationsDir, { recursive: true });
-      const filename = `${new Date().toISOString().split('T')[0]}-${name}.md`;
-      fs.writeFileSync(path.join(conversationsDir, filename), formatTranscriptMarkdown(messages, summary, assistantName));
-      log(`Archived conversation to ${filename}`);
-    } catch (err) {
-      log(`Failed to archive transcript: ${err instanceof Error ? err.message : String(err)}`);
-    }
+    archiveTranscriptFile(preCompact.transcript_path, preCompact.session_id, assistantName);
    return {};
  };
 }

+// ── Continuation rotation (cold-resume guard) ──
+
+/**
+ * Resume cost is dominated by transcript size. Past this many bytes a fresh
+ * cold container can't reload the .jsonl before the host's 30-min idle ceiling
+ * fires, so the session is dropped and started clean. Operator-overridable.
+ */
+function transcriptRotateBytes(): number {
+  return Number(process.env.CLAUDE_TRANSCRIPT_ROTATE_BYTES) || 12 * 1024 * 1024;
+}
+
+/**
+ * Secondary age trigger, measured from the transcript's first entry. 0 (or a
+ * non-positive value) disables the age check; size alone then governs.
+ */
+function transcriptRotateAgeMs(): number {
+  const raw = process.env.CLAUDE_TRANSCRIPT_ROTATE_AGE_DAYS;
+  if (raw === undefined || raw.trim() === '') return 14 * 86_400_000;
+  const days = Number(raw);
+  if (!Number.isFinite(days)) return 14 * 86_400_000;
+  // Explicit non-positive override disables the age check; size alone governs.
+  return days > 0 ? days * 86_400_000 : Infinity;
+}
+
+function claudeProjectsDir(): string {
+  const base = process.env.CLAUDE_CONFIG_DIR || path.join(process.env.HOME || os.homedir(), '.claude');
+  return path.join(base, 'projects');
+}
+
+/**
+ * Locate the .jsonl backing a session id. The SDK names project dirs by a
+ * mangled cwd; rather than reproduce that convention we scan project dirs for
+ * `<sessionId>.jsonl` (session ids are UUIDs, so this is unambiguous).
+ */
+function findTranscriptPath(sessionId: string): string | null {
+  const projects = claudeProjectsDir();
+  let dirs: string[];
+  try {
+    dirs = fs.readdirSync(projects);
+  } catch {
+    return null;
+  }
+  for (const dir of dirs) {
+    const candidate = path.join(projects, dir, `${sessionId}.jsonl`);
+    if (fs.existsSync(candidate)) return candidate;
+  }
+  return null;
+}
+
+/** Epoch-ms of the first transcript entry, or null if unreadable. */
+function transcriptStartMs(transcriptPath: string): number | null {
+  try {
+    const fd = fs.openSync(transcriptPath, 'r');
+    try {
+      const buf = Buffer.alloc(4096);
+      const n = fs.readSync(fd, buf, 0, buf.length, 0);
+      const firstLine = buf.toString('utf-8', 0, n).split('\n', 1)[0];
+      const ts = JSON.parse(firstLine)?.timestamp;
+      const ms = ts ? Date.parse(ts) : NaN;
+      return Number.isNaN(ms) ? null : ms;
+    } finally {
+      fs.closeSync(fd);
+    }
+  } catch {
+    return null;
+  }
+}
+
 // ── Provider ──

 /**
 * Claude Code auto-compacts context at this window (tokens). Kept here so
 * the generic bootstrap doesn't need to know about Claude-specific env vars.
+ *
+ * Operator override: set CLAUDE_CODE_AUTO_COMPACT_WINDOW in the host env to
+ * raise or lower the threshold without editing source — useful when running
+ * with a 1M-context model variant or when emergency-tuning a deployment.
 */
-const CLAUDE_CODE_AUTO_COMPACT_WINDOW = '165000';
+const CLAUDE_CODE_AUTO_COMPACT_WINDOW = process.env.CLAUDE_CODE_AUTO_COMPACT_WINDOW || '165000';

 /**
 * Stale-session detection. Matches Claude Code's error text when a
@@ -243,11 +335,15 @@ export class ClaudeProvider implements AgentProvider {
  private mcpServers: Record<string, McpServerConfig>;
  private env: Record<string, string | undefined>;
  private additionalDirectories?: string[];
+  private model?: string;
+  private effort?: string;

  constructor(options: ProviderOptions = {}) {
    this.assistantName = options.assistantName;
    this.mcpServers = options.mcpServers ?? {};
    this.additionalDirectories = options.additionalDirectories;
+    this.model = options.model;
+    this.effort = options.effort;
    this.env = {
      ...(options.env ?? {}),
      CLAUDE_CODE_AUTO_COMPACT_WINDOW,
@@ -259,6 +355,41 @@ export class ClaudeProvider implements AgentProvider {
    return STALE_SESSION_RE.test(msg);
  }

+  maybeRotateContinuation(continuation: string): string | null {
+    const transcriptPath = findTranscriptPath(continuation);
+    if (!transcriptPath) return null;
+
+    let size: number;
+    try {
+      size = fs.statSync(transcriptPath).size;
+    } catch {
+      return null;
+    }
+
+    const maxBytes = transcriptRotateBytes();
+    const startMs = transcriptStartMs(transcriptPath);
+    const ageMs = startMs === null ? 0 : Date.now() - startMs;
+    const maxAgeMs = transcriptRotateAgeMs();
+
+    let reason: string | null = null;
+    if (size > maxBytes) {
+      reason = `transcript ${(size / 1_048_576).toFixed(1)}MB > ${(maxBytes / 1_048_576).toFixed(0)}MB cap`;
+    } else if (startMs !== null && ageMs > maxAgeMs) {
+      reason = `transcript ${(ageMs / 86_400_000).toFixed(1)}d old > ${(maxAgeMs / 86_400_000).toFixed(0)}d cap`;
+    }
+    if (!reason) return null;
+
+    // Preserve a readable summary, then move the heavy .jsonl out of the
+    // resume path so the SDK starts a fresh session and the disk is reclaimed.
+    archiveTranscriptFile(transcriptPath, continuation, this.assistantName);
+    try {
+      fs.renameSync(transcriptPath, `${transcriptPath}.rotated-${Date.now()}`);
+    } catch (err) {
+      log(`Failed to move rotated transcript aside: ${err instanceof Error ? err.message : String(err)}`);
+    }
+    return reason;
+  }
+
  query(input: QueryInput): AgentQuery {
    const stream = new MessageStream();
    stream.push(input.prompt);
@@ -273,12 +404,18 @@ export class ClaudeProvider implements AgentProvider {
        resume: input.continuation,
        pathToClaudeCodeExecutable: '/pnpm/claude',
        systemPrompt: instructions ? { type: 'preset' as const, preset: 'claude_code' as const, append: instructions } : undefined,
-        allowedTools: TOOL_ALLOWLIST,
+        allowedTools: [
+          ...TOOL_ALLOWLIST,
+          ...Object.keys(this.mcpServers).map(mcpAllowPattern),
+        ],
        disallowedTools: SDK_DISALLOWED_TOOLS,
        env: this.env,
+        model: this.model,
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        effort: this.effort as any,
        permissionMode: 'bypassPermissions',
        allowDangerouslySkipPermissions: true,
-        settingSources: ['project', 'user'],
+        settingSources: ['project', 'user', 'local'],
        mcpServers: this.mcpServers,
        hooks: {
          PreToolUse: [{ hooks: [preToolUseHook] }],
@@ -14,6 +14,21 @@ export interface AgentProvider {
   * (missing transcript, unknown session, etc.) and should be cleared.
   */
  isSessionInvalid(err: unknown): boolean;
+
+  /**
+   * Optional pre-resume maintenance. Given the stored continuation token,
+   * decide whether its backing transcript has grown too large or too old to
+   * resume cheaply. Return a non-null reason string to tell the caller to drop
+   * the continuation and start a fresh session (the provider archives any
+   * recoverable summary first); return null to keep resuming.
+   *
+   * Guards the cold-resume failure mode: a long-lived hub session accumulates
+   * days of history — including base64 image blocks the agent Read — and the
+   * SDK reloads the whole .jsonl on every resume. Past a threshold the first
+   * turn alone can exceed the host's idle ceiling, so the container is killed
+   * before it ever replies. Providers without an on-disk transcript omit this.
+   */
+  maybeRotateContinuation?(continuation: string, cwd: string): string | null;
 }

 /**
@@ -25,6 +40,16 @@ export interface ProviderOptions {
  mcpServers?: Record<string, McpServerConfig>;
  env?: Record<string, string | undefined>;
  additionalDirectories?: string[];
+  /**
+   * Model alias (`sonnet`, `opus`, `haiku`) or full model ID. Passed through
+   * to the underlying SDK. If omitted, the SDK default is used.
+   */
+  model?: string;
+  /**
+   * Reasoning effort (`'low' | 'medium' | 'high' | 'xhigh' | 'max'`). Passed
+   * through to the underlying SDK. If omitted, the SDK default is used.
+   */
+  effort?: string;
 }

 export interface QueryInput {
@@ -0,0 +1,84 @@
+import { describe, it, expect, beforeEach, afterEach } from 'bun:test';
+
+import { initTestSessionDb, closeSessionDb, getInboundDb } from './db/connection.js';
+import { getUndeliveredMessages } from './db/messages-out.js';
+import { getPendingMessages } from './db/messages-in.js';
+import type { MessageInRow } from './db/messages-in.js';
+import { MockProvider } from './providers/mock.js';
+import { runPollLoop } from './poll-loop.js';
+import { isUploadTraceCommand } from './upload-trace.js';
+
+beforeEach(() => {
+  initTestSessionDb();
+});
+
+afterEach(() => {
+  closeSessionDb();
+});
+
+describe('isUploadTraceCommand', () => {
+  const make = (text: unknown) => ({ content: JSON.stringify({ text }) }) as MessageInRow;
+
+  it('matches /upload-trace (case-insensitive, with args)', () => {
+    expect(isUploadTraceCommand(make('/upload-trace'))).toBe(true);
+    expect(isUploadTraceCommand(make('/UPLOAD-TRACE'))).toBe(true);
+    expect(isUploadTraceCommand(make('  /upload-trace now '))).toBe(true);
+  });
+
+  it('does not match other text or commands', () => {
+    expect(isUploadTraceCommand(make('hello'))).toBe(false);
+    expect(isUploadTraceCommand(make('/upload'))).toBe(false);
+    expect(isUploadTraceCommand(make('/clear'))).toBe(false);
+    expect(isUploadTraceCommand({ content: 'not json' } as MessageInRow)).toBe(false);
+  });
+});
+
+describe('poll loop — /upload-trace command', () => {
+  it('handles the command in the runner, writes a status, skips query', async () => {
+    getInboundDb()
+      .prepare(
+        `INSERT INTO messages_in (id, kind, timestamp, status, platform_id, channel_type, content)
+         VALUES ('m-upload-trace', 'chat', datetime('now'), 'pending', 'chan-1', 'discord', ?)`,
+      )
+      .run(JSON.stringify({ text: '/upload-trace' }));
+
+    // If the provider were ever queried it would emit this — asserting its
+    // absence proves the runner intercepted /upload-trace instead of the LLM.
+    const provider = new MockProvider({}, () => '<message to="discord-test">should not run</message>');
+    const controller = new AbortController();
+    const loopPromise = runPollLoopWithTimeout(provider, controller.signal, 5000);
+
+    await waitFor(() => getUndeliveredMessages().length > 0, 5000);
+    controller.abort();
+
+    const out = getUndeliveredMessages();
+    expect(out).toHaveLength(1);
+    // A status line from uploadTrace() — never the provider's reply.
+    const text = JSON.parse(out[0].content).text as string;
+    expect(text.length).toBeGreaterThan(0);
+    expect(text).not.toBe('should not run');
+
+    // Command message was completed (not left pending).
+    expect(getPendingMessages()).toHaveLength(0);
+
+    await loopPromise.catch(() => {});
+  });
+});
+
+async function runPollLoopWithTimeout(provider: MockProvider, signal: AbortSignal, timeoutMs: number): Promise<void> {
+  return Promise.race([
+    runPollLoop({ provider, providerName: 'mock', cwd: '/tmp' }),
+    new Promise<void>((_, reject) => {
+      signal.addEventListener('abort', () => reject(new Error('aborted')));
+    }),
+    new Promise<void>((_, reject) => setTimeout(() => reject(new Error('timeout')), timeoutMs)),
+  ]);
+}
+
+async function waitFor(condition: () => boolean, timeoutMs: number): Promise<void> {
+  const start = Date.now();
+  while (!condition()) {
+    if (Date.now() - start > timeoutMs) throw new Error('waitFor timeout');
+    await new Promise((resolve) => setTimeout(resolve, 50));
+  }
+}
@@ -0,0 +1,172 @@
+import { spawnSync } from 'node:child_process';
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
+
+import type { MessageInRow } from './db/messages-in.js';
+
+/**
+ * `/upload-trace` command: upload this session's Claude Code transcript to the user's
+ * own private `{hf_user}/nanoclaw-traces` dataset, browsable in the HF Agent
+ * Trace Viewer. The transcript the Claude provider keeps under
+ * `~/.claude/projects/<dir>/<sessionId>.jsonl` is already in the format the
+ * viewer auto-detects, so this just locates the newest one and pushes it.
+ *
+ * Auth is the OneCLI gateway's job: curl goes out through the injected
+ * HTTPS_PROXY, which adds the user's HF token. We never see the raw token, and
+ * a 401 from `whoami` is our "not signed in" signal.
+ */
+
+/**
+ * Narrow check for /upload-trace — the runner handles this command directly
+ * (no LLM turn). Admin-gated by the host router before it reaches the container.
+ */
+export function isUploadTraceCommand(msg: MessageInRow): boolean {
+  let text = '';
+  try {
+    text = (JSON.parse(msg.content)?.text ?? '').trim();
+  } catch {
+    return false; // non-JSON content is never a command
+  }
+  return text.toLowerCase().startsWith('/upload-trace');
+}
+
+/** Newest Claude Code transcript jsonl (the current session). */
+function newestTranscript(): string | null {
+  const projects = path.join(os.homedir(), '.claude', 'projects');
+  let best: { p: string; m: number } | null = null;
+  let dirs: string[];
+  try {
+    dirs = fs.readdirSync(projects);
+  } catch {
+    return null;
+  }
+  for (const dir of dirs) {
+    let files: string[];
+    try {
+      files = fs.readdirSync(path.join(projects, dir));
+    } catch {
+      continue;
+    }
+    for (const f of files) {
+      if (!f.endsWith('.jsonl')) continue;
+      const p = path.join(projects, dir, f);
+      const m = fs.statSync(p).mtimeMs;
+      if (!best || m > best.m) best = { p, m };
+    }
+  }
+  return best?.p ?? null;
+}
+
+function curl(args: string[], input?: string): { ok: boolean; out: string } {
+  const r = spawnSync('curl', args, { input, encoding: 'utf-8' });
+  return { ok: r.status === 0, out: (r.stdout ?? '') + (r.stderr ?? '') };
+}
+
+/**
+ * Setup instructions for when whoami fails. `body` is the gateway's error
+ * JSON (when the request was proxied through OneCLI). We surface the URL it
+ * hands back — `secret_url` for an unknown host (HF's case), `connect_url`
+ * for an OAuth app, `manage_url` when the secret exists but this agent lacks
+ * access — so the link always points at the right gateway (local or hosted).
+ */
+function notSignedInMessage(body: string): string {
+  let setupUrl: string | undefined;
+  try {
+    const e = JSON.parse(body) as { secret_url?: string; connect_url?: string; manage_url?: string };
+    if (e.secret_url) {
+      // The pre-filled `path` defaults to the failing request path
+      // (/api/whoami-v2), which scopes the secret to that one endpoint. Blank
+      // it so the secret matches all of huggingface.co — the upload endpoints
+      // included, not just whoami.
+      setupUrl = e.secret_url.replace(/([?&]path=)[^&]*/, '$1');
+    } else {
+      setupUrl = e.connect_url ?? e.manage_url;
+    }
+  } catch {
+    /* non-JSON body (e.g. HF's own error, or no gateway) — generic fallback */
+  }
+  const lines = [
+    "Can't upload — no Hugging Face token is available to this agent. To set it up:",
+    '',
+    '1. Create a token with WRITE access at https://huggingface.co/settings/tokens',
+    '   (New token → type "Write" → copy it).',
+    '',
+    setupUrl
+      ? `2. Add it to OneCLI here: ${setupUrl}`
+      : '2. Add it to the OneCLI vault as a secret with host pattern  huggingface.co',
+    '',
+    'Then run /upload-trace again.',
+  ];
+  return lines.join('\n');
+}
+
+/** Returns a user-facing status line. Never throws. */
+export function uploadTrace(): string {
+  const file = newestTranscript();
+  if (!file) return 'No transcript to upload for this session yet.';
+
+  // whoami, capturing the body + HTTP status (no -f, so the gateway's error
+  // JSON survives a 401). When no token is available the OneCLI gateway
+  // returns a setup URL pre-filled for *this* gateway — so we never hardcode
+  // local-vs-hosted dashboard links, and never have to know which it is.
+  const who = curl(['-s', '-w', '\n%{http_code}', 'https://huggingface.co/api/whoami-v2']);
+  const nl = who.out.lastIndexOf('\n');
+  const body = nl === -1 ? '' : who.out.slice(0, nl);
+  const status = nl === -1 ? who.out.trim() : who.out.slice(nl + 1).trim();
+
+  if (status !== '200') {
+    return notSignedInMessage(body);
+  }
+  let user: string | undefined;
+  try {
+    user = JSON.parse(body)?.name;
+  } catch {
+    /* fall through */
+  }
+  if (!user) return 'Could not resolve your Hugging Face username.';
+
+  const repo = `${user}/nanoclaw-traces`;
+  // Idempotent create — ignore failure (already exists / no-op). The
+  // Content-Type header is required: without it curl sends form-encoding and
+  // the Hub rejects the body with 400 (expected string at "name").
+  curl([
+    '-sf',
+    '-X',
+    'POST',
+    'https://huggingface.co/api/repos/create',
+    '-H',
+    'Content-Type: application/json',
+    '-d',
+    JSON.stringify({ type: 'dataset', name: 'nanoclaw-traces', private: true }),
+  ]);
+
+  const content = fs.readFileSync(file).toString('base64');
+  const repoPath = `sessions/${path.basename(file)}`;
+  const ndjson =
+    JSON.stringify({ key: 'header', value: { summary: 'add session trace' } }) +
+    '\n' +
+    JSON.stringify({
+      key: 'file',
+      value: { path: repoPath, encoding: 'base64', content },
+    }) +
+    '\n';
+
+  const commit = curl(
+    [
+      '-sf',
+      '-X',
+      'POST',
+      `https://huggingface.co/api/datasets/${repo}/commit/main`,
+      '-H',
+      'Content-Type: application/x-ndjson',
+      '--data-binary',
+      '@-',
+    ],
+    ndjson,
+  );
+  if (!commit.ok) {
+    return 'Upload to Hugging Face failed (the transcript may be too large for an inline commit).';
+  }
+  return `Uploaded → https://huggingface.co/datasets/${repo}/blob/main/${repoPath}`;
+}
@@ -9,9 +9,15 @@
 set -e

 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
 cd "$SCRIPT_DIR"

-IMAGE_NAME="nanoclaw-agent"
+# Derive the image name from the project root so two NanoClaw installs on the
+# same host don't overwrite each other's `nanoclaw-agent:latest` tag. Matches
+# setup/lib/install-slug.sh + src/install-slug.ts.
+# shellcheck source=../setup/lib/install-slug.sh
+source "$PROJECT_ROOT/setup/lib/install-slug.sh"
+IMAGE_NAME="$(container_image_base)"
 TAG="${1:-latest}"
 CONTAINER_RUNTIME="${CONTAINER_RUNTIME:-docker}"

@@ -0,0 +1,85 @@
+---
+name: onecli-gateway
+description: >-
+  OneCLI Gateway: transparent HTTPS proxy that injects stored credentials
+  into outbound calls. You MUST use this skill when the user asks you to
+  read emails, check calendar, access GitHub repos, create issues, check
+  Stripe payments, or interact with ANY external service or API. Do NOT
+  use browser extensions or OAuth CLI tools. Make HTTP requests directly;
+  the gateway injects credentials automatically.
+compatibility: Requires HTTPS_PROXY set in environment (automatic when launched via `onecli run`)
+metadata:
+  author: onecli
+  version: "0.5.0"
+---
+
+# OneCLI Gateway
+
+Your outbound HTTPS traffic is transparently proxied through the OneCLI
+gateway, which injects stored credentials at the proxy boundary. You never
+see or handle credential values directly.
+
+## How to Access External Services
+
+You have direct HTTP access to external APIs. OAuth apps (Gmail, GitHub,
+Google Calendar, Google Drive, etc.) and API key services are all available
+through the gateway. Just make the request directly; the gateway injects
+credentials if the app is connected. If not, it returns an error with a
+connect URL you can present to the user.
+
+## Making Requests
+
+Call the real API URL. The gateway intercepts the request and injects
+credentials automatically.
+
+```bash
+curl -s "https://gmail.googleapis.com/gmail/v1/users/me/messages?maxResults=5"
+curl -s "https://api.github.com/user/repos?per_page=10"
+curl -s "https://api.stripe.com/v1/charges?limit=5"
+```
+
+Standard HTTP clients (curl, fetch, requests, axios, Go net/http, git) all
+honor the `HTTPS_PROXY` environment variable automatically. You do not need
+to set any auth headers.
+
+## Credential Stubs for MCP Servers
+
+Some MCP servers need local credential files to start. Stubs for connected
+apps are pre-written automatically. Files containing `"onecli-managed"`
+values are managed by OneCLI — do NOT modify or delete them.
+
+If an MCP server won't start due to missing credentials, create stubs
+**before** starting it. Use `"onecli-managed"` as the placeholder for all
+secret values, with file permissions `0600`. See the guide at:
+https://www.onecli.sh/docs/guides/credential-stubs/general-app
+
+## When a Request Fails
+
+If you get a 401, 403, or a gateway error (e.g., `app_not_connected`):
+
+**Step 1 — Show the user a connect link.** Use the `connect_url` from the
+error response:
+
+> To connect [service], open this link:
+> [connect_url from the error response]
+
+If there is no `connect_url` in the error, tell the user to open the
+OneCLI dashboard and connect the service there.
+
+**Step 2 — Retry after the user connects.** Let the user know you will
+retry once they have connected. When they confirm, retry the original
+request. If the retry still fails, ask if they need help with the setup.
+
+## Rules
+
+- **Never** say "I don't have access to X" without first making the HTTP
+  request through the proxy.
+- **Never** use browser extensions, gcloud, or manual auth flows. The
+  gateway handles credentials for you.
+- **Never** ask the user for API keys or tokens directly. Direct them to
+  connect the service in the OneCLI dashboard.
+- **Never** suggest the user open Gmail/Calendar/GitHub in their browser
+  when they ask you to read or interact with those services. You have API
+  access. Use it.
+- If the gateway returns a policy error (403 with a JSON body), respect
+  the block. Do not retry or circumvent it.
@@ -0,0 +1,7 @@
+# Credentials & External Services
+
+Your HTTP requests go through the OneCLI proxy, which injects real credentials automatically. Just call any API directly (Gmail, GitHub, Slack, etc.) — the proxy adds auth before it reaches the service.
+
+Use any method: curl, Python, a CLI tool, whatever fits. If a tool checks for credentials locally, pass any placeholder value — the proxy replaces it with real credentials at request time.
+
+If you get a `401`/`403`/`app_not_connected`, the error response contains a `connect_url` — you MUST show it to the user as a bare URL on its own line (no angle brackets, no markdown link syntax) so they can click to connect. Run `/onecli-gateway` for the full error-handling flow. Never ask the user for API keys or tokens.
@@ -9,7 +9,7 @@ You've just been connected to a new user. This your time to shine and make a str

 ## What to do

-1. Send a short, warm greeting using `send_message`
+1. Send a short, warm greeting
 2. State your name (from your system prompt / CLAUDE.md)
 3. Signal that you're capable of a lot — but don't list everything upfront. Be intriguing, not encyclopedic
 4. Ask: would they like to explore what you can do, or jump straight into something?
@@ -0,0 +1,61 @@
+---
+name: whatsapp-formatting
+description: Format messages for WhatsApp, including mentions that render as real WhatsApp tags. Use when responding in a WhatsApp conversation (platform_id / chatJid ends with @s.whatsapp.net or @g.us).
+---
+
+# WhatsApp Message Formatting
+
+WhatsApp uses its own lightweight markup and a phone-number-based mention syntax. The host's WhatsApp adapter (Baileys) handles markdown conversion automatically, but **mentions are only protocol-level mentions if you use the right syntax** — otherwise they render as plain text and don't notify the recipient.
+
+## How to detect WhatsApp context
+
+You're in a WhatsApp conversation when any of these are true:
+- The chat JID / platform id ends with `@s.whatsapp.net` (1-on-1 DM)
+- The chat JID / platform id ends with `@g.us` (group)
+- Your inbound message metadata has `chatJid` matching the above
+
+## Mentions — the important part
+
+To tag a user so their name appears **bold and clickable** in WhatsApp and they get a push notification, write the `@` followed by their phone number digits (no `+`, no spaces, no display name):
+
+```
+@15551234567 can you confirm?
+```
+
+The adapter scans your outgoing text for `@<digits>` (5–15 digits, optional leading `+` is stripped) and tells WhatsApp to render them as real mention tags.
+
+**The sender's phone JID is always in your inbound message metadata.** When a user writes to you, inbound `content.sender` looks like `15551234567@s.whatsapp.net`. The part before the `@` is exactly what you put after `@` when tagging them back.
+
+### Wrong vs right
+
+| You write | What recipients see |
+|-----------|---------------------|
+| `@Adam can you...` | Plain text `@Adam`. No tag, no notification. |
+| `@15551234567 can you...` | Bold/blue **@Adam** (or whatever name they're saved as), notification fires. |
+| `@+15551234567 ...` | Same as above — adapter strips the `+`. |
+
+### Picking who to tag
+
+- In a DM, there's no real need to tag the recipient (they already see every message), but tagging still works if you want emphasis.
+- In a group, look at the `participants` / inbound `content.sender` to find the JID of the person you mean. Don't guess from display names — pushNames can collide and are not reliable.
+- If you don't know the JID, just refer to the person by name in plain prose. Don't write `@<name>` — it won't tag and it will look like a tag that failed.
+
+## Text styles
+
+WhatsApp uses single-character delimiters, *not* doubled like standard Markdown.
+
+| Style | Syntax | Renders as |
+|-------|--------|------------|
+| Bold | `*bold*` | **bold** |
+| Italic | `_italic_` | *italic* |
+| Strikethrough | `~strike~` | ~strike~ |
+| Monospace | `` `code` `` | `code` |
+| Block monospace | ```` ```block``` ```` | preformatted block |
+
+The adapter converts standard Markdown (`**bold**`, `[link](url)`, `# heading`) to the WhatsApp-native form automatically, so you don't have to think about it — but be aware that single asterisks become italics, not bold.
+
+## What not to do
+
+- Don't write `<@U123>` (that's Slack), `<@!123>` (Discord), or any other channel's mention syntax.
+- Don't paste a full JID like `@15551234567@s.whatsapp.net` in the text — only the digits before the JID's `@` go after your `@`.
+- Don't try to tag display names. WhatsApp has no display-name-based mention API.
@@ -0,0 +1,19 @@
+## WhatsApp mentions — always use phone digits
+
+When you are replying in a WhatsApp conversation (the inbound message's `chatJid` ends with `@s.whatsapp.net` for a DM or `@g.us` for a group), and you want to tag a person so their name appears **bold and clickable** with a push notification, write `@` followed by their phone-number digits — never the display name.
+
+**The sender's phone JID is in your inbound message metadata** at `content.sender` (e.g. `15551234567@s.whatsapp.net`). The part before the `@` is exactly what you put after `@` when tagging them.
+
+| You write | What recipients see |
+|-----------|---------------------|
+| `@Adam, can you...` | Plain text. No tag, no notification. |
+| `@15551234567, can you...` | Bold/blue **@Adam** (whatever name they're saved as), notification fires. |
+| `@+15551234567 ...` | Same as above — the adapter strips the `+` automatically. |
+
+The host adapter scans your outbound text for `@<5–15 digits>` (with optional leading `+`) and tells WhatsApp to render those as real mention tags. If the digits aren't in the text, the tag doesn't render — no exceptions.
+
+### In groups
+
+Tag the person you're addressing using their JID from inbound metadata (look at the most recent message from them). Don't guess — pushNames collide and aren't reliable.
+
+If you don't know someone's JID, refer to them by name in plain prose. Do not write `@<displayname>` hoping it works.
@@ -2,7 +2,7 @@

 ## Structure

-**`qwibitai/nanoclaw`** (upstream) — core engine with skill definitions (`.claude/skills/`). No channel code on `main`.
+**`nanocoai/nanoclaw`** (upstream) — core engine with skill definitions (`.claude/skills/`). No channel code on `main`.

 **Channel forks** (`nanoclaw-whatsapp`, `nanoclaw-telegram`, `nanoclaw-slack`, etc.) — each fork = upstream + one channel's code applied. Users clone upstream, then merge a fork into their clone to add a channel.

@@ -10,7 +10,7 @@ Access layer: `src/db/`. Authoritative schema reference: `src/db/schema.ts` (com

 ### 1.1 `agent_groups`

-Agent workspaces. Each maps 1:1 to a `groups/<folder>/` directory containing `CLAUDE.md`, skills, and `container.json`. Container config lives on disk, not in the DB.
+Agent workspaces. Each maps 1:1 to a `groups/<folder>/` directory containing `CLAUDE.md` and skills. Container config lives in `container_configs` (see §1.x below); a `container.json` file is materialized at spawn time for the container runner to read.

 ```sql
 CREATE TABLE agent_groups (
@@ -294,6 +294,32 @@ CREATE TABLE schema_version (
 );
 ```

+### 1.15 `container_configs`
+
+Per-agent-group container runtime config. Source of truth for provider, model, packages, MCP servers, mounts, CLI scope, etc. Materialized to `groups/<folder>/container.json` at spawn time.
+
+```sql
+CREATE TABLE container_configs (
+  agent_group_id         TEXT PRIMARY KEY REFERENCES agent_groups(id) ON DELETE CASCADE,
+  provider               TEXT,
+  model                  TEXT,
+  effort                 TEXT,
+  image_tag              TEXT,
+  assistant_name         TEXT,
+  max_messages_per_prompt INTEGER,
+  skills                 TEXT NOT NULL DEFAULT '"all"',
+  mcp_servers            TEXT NOT NULL DEFAULT '{}',
+  packages_apt           TEXT NOT NULL DEFAULT '[]',
+  packages_npm           TEXT NOT NULL DEFAULT '[]',
+  additional_mounts      TEXT NOT NULL DEFAULT '[]',
+  cli_scope              TEXT NOT NULL DEFAULT 'group',   -- disabled | group | global
+  updated_at             TEXT NOT NULL
+);
+```
+
+- **Readers:** `src/container-config.ts`, `src/container-runner.ts`, `src/cli/dispatch.ts` (scope enforcement), `src/claude-md-compose.ts`
+- **Writers:** `src/db/container-configs.ts`, `src/modules/self-mod/apply.ts`, `src/backfill-container-configs.ts`
+
 ---

 ## 2. Migration system
@@ -313,6 +339,8 @@ Migrations live in `src/db/migrations/`, one file per migration. Runner: `runMig
 | 007 | `007-pending-approvals-title-options.ts` | `ALTER TABLE pending_approvals` add `title`, `options_json` (retrofits DBs created between 003 and 007) |
 | 008 | `008-dropped-messages.ts` | `unregistered_senders` |
 | 009 | `009-drop-pending-credentials.ts` | Drop the defunct `pending_credentials` table |
+| 014 | `014-container-configs.ts` | `container_configs` — per-agent-group container runtime config |
+| 015 | `015-cli-scope.ts` | `ALTER TABLE container_configs ADD COLUMN cli_scope` |

 Numbers 005 and 006 are intentionally absent — migrations were renumbered during early development.

@@ -33,19 +33,22 @@ Every message landing in the session: user chat, scheduled task, recurring task,

 ```sql
 CREATE TABLE messages_in (
-  id            TEXT PRIMARY KEY,
-  seq           INTEGER UNIQUE,           -- EVEN only (host assigns) — see §3
-  kind          TEXT NOT NULL,
-  timestamp     TEXT NOT NULL,
-  status        TEXT DEFAULT 'pending',   -- pending|completed|failed|paused
-  process_after TEXT,
-  recurrence    TEXT,                     -- cron expr for recurring
-  series_id     TEXT,                     -- groups occurrences of a recurring task
-  tries         INTEGER DEFAULT 0,
-  platform_id   TEXT,
-  channel_type  TEXT,
-  thread_id     TEXT,
-  content       TEXT NOT NULL             -- JSON; shape depends on kind
+  id             TEXT PRIMARY KEY,
+  seq            INTEGER UNIQUE,           -- EVEN only (host assigns) — see §3
+  kind           TEXT NOT NULL,
+  timestamp      TEXT NOT NULL,
+  status         TEXT DEFAULT 'pending',   -- pending|completed|failed|paused
+  process_after  TEXT,
+  recurrence     TEXT,                     -- cron expr for recurring
+  series_id      TEXT,                     -- groups occurrences of a recurring task
+  tries          INTEGER DEFAULT 0,
+  trigger        INTEGER NOT NULL DEFAULT 1, -- 0 = context only (don't wake), 1 = wake agent
+  platform_id    TEXT,
+  channel_type   TEXT,
+  thread_id      TEXT,
+  content        TEXT NOT NULL,            -- JSON; shape depends on kind
+  source_session_id TEXT,                  -- agent-to-agent return path
+  on_wake        INTEGER NOT NULL DEFAULT 0 -- 1 = only deliver on container's first poll
 );
 CREATE INDEX idx_messages_in_series ON messages_in(series_id);
 ```
--- a/Show More
+++ b/Show More