feat(providers): generalize provider output substitutions

Adds a per-provider `errorSubstitutions` array on AgentProvider — each
entry is a `(name, test, replace)` triple. The poll-loop iterates the
provider's rules (in declaration order) before delivering result text
or error text to the user; the first matching `test` regex wins and
its `replace` string is sent in place of the raw output. If no rule
matches, the original text passes through unchanged — there is no
fallback message.

Replaces the single-purpose `isAuthRequired` / `authRequiredMessage`
shape with a list of typed rules so providers can declare any number
of swap-this-for-that mappings (auth banners, rate-limit hints,
context-too-long errors, etc.). New rule types just add an entry.

ClaudeProvider ships one rule today: `auth-required`, matching Claude
Code's "Not logged in · Please run /login" / "Invalid API key …"
banners with a host-aware remediation (re-run setup or run `claude`
in the project directory).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.claude/skills/add-atomic-chat-tool/SKILL.md

@@ -0,0 +1,243 @@
+---
+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
+
+```bash
+launchctl kickstart -k gui/$(id -u)/com.nanoclaw  # macOS
+# Linux: systemctl --user restart nanoclaw
+```
+
+## 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.

.claude/skills/add-atomic-chat-tool/atomic-chat-mcp-stdio.ts

@@ -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);

.claude/skills/add-codex/SKILL.md

@@ -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.

.claude/skills/add-gcal-tool/SKILL.md

@@ -0,0 +1,210 @@
+---
+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 && \
+grep -q "mcp__calendar__\*" container/agent-runner/src/providers/claude.ts && \
+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}"
+```
+
+### Add tools to allowlist
+
+Edit `container/agent-runner/src/providers/claude.ts`. Add `'mcp__calendar__*'` to `TOOL_ALLOWLIST` after `'mcp__nanoclaw__*'` (or after `'mcp__gmail__*'` if present).
+
+### Rebuild the container image
+
+```bash
+./container/build.sh
+```
+
+## Phase 3: Wire Per-Agent-Group
+
+For each agent group, merge into `groups/<folder>/container.json`:
+
+```jsonc
+{
+  "mcpServers": {
+    "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"
+      }
+    }
+  },
+  "additionalMounts": [
+    {
+      "hostPath": "/home/<user>/.calendar-mcp",
+      "containerPath": ".calendar-mcp",
+      "readonly": false
+    }
+  ]
+}
+```
+
+Substitute `<user>` with `echo $HOME`. `containerPath` is relative (mount-security rejects absolute paths — additional mounts land at `/workspace/extra/<relative>`).
+
+**Same-group-as-gmail tip:** if this group already has the gmail MCP + `.gmail-mcp` mount, **merge, don't replace** — both entries coexist in `mcpServers` and `additionalMounts`.
+
+## Phase 4: Build and Restart
+
+```bash
+pnpm run build
+systemctl --user restart nanoclaw   # Linux
+# launchctl kickstart -k gui/$(id -u)/com.nanoclaw   # macOS
+```
+
+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" → `mcp__calendar__*` missing from `TOOL_ALLOWLIST`, or image cache stale (`./container/build.sh` again).
+
+## Removal
+
+1. Delete `"calendar"` from `mcpServers` and the `.calendar-mcp` mount from `additionalMounts` in each group's `container.json`.
+2. Remove `'mcp__calendar__*'` from `TOOL_ALLOWLIST`.
+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 nanoclaw`.
+5. Optional: `rm -rf ~/.calendar-mcp/` and `onecli apps disconnect --provider google-calendar`.
+
+## 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.

.claude/skills/add-gmail-tool/SKILL.md

@@ -0,0 +1,229 @@
+---
+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:
+
+```bash
+onecli secrets list     # find Gmail secret IDs (OneCLI creates one per connected app)
+onecli agents set-secrets --id <agent-id> --secret-ids <gmail-secret-id>
+```
+
+## Phase 2: Apply Code Changes
+
+### Check if already applied
+
+```bash
+grep -q 'GMAIL_MCP_VERSION' container/Dockerfile && \
+grep -q "mcp__gmail__\*" container/agent-runner/src/providers/claude.ts && \
+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`.
+
+### Add tools to allowlist
+
+Edit `container/agent-runner/src/providers/claude.ts`. Find `'mcp__nanoclaw__*',` in `TOOL_ALLOWLIST` and add `'mcp__gmail__*',` after it.
+
+### 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), edit `groups/<folder>/container.json` to add the mount and MCP server.
+
+Merge these into the group's `container.json`:
+
+```jsonc
+{
+  "mcpServers": {
+    "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"
+      }
+    }
+  },
+  "additionalMounts": [
+    {
+      "hostPath": "/home/<user>/.gmail-mcp",
+      "containerPath": ".gmail-mcp",
+      "readonly": false
+    }
+  ]
+}
+```
+
+Substitute `<user>` with the host user's home (use `echo $HOME`, don't assume `~` will expand — `container-runner.ts` does expand `~` via `expandPath`, but an explicit absolute path is clearer and matches what `/manage-mounts` writes).
+
+**Why the container path is relative:** `mount-security` rejects absolute `containerPath` values. Additional mounts are prefixed with `/workspace/extra/`, so `containerPath: ".gmail-mcp"` lands at `/workspace/extra/.gmail-mcp`. The MCP server's `GMAIL_OAUTH_PATH` / `GMAIL_CREDENTIALS_PATH` env vars point at that absolute location inside the container.
+
+## Phase 4: Build and Restart
+
+```bash
+pnpm run build
+systemctl --user restart nanoclaw  # Linux
+# launchctl kickstart -k gui/$(id -u)/com.nanoclaw   # macOS
+```
+
+## 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" → `mcp__gmail__*` wasn't added to `TOOL_ALLOWLIST`, or the agent-runner wasn't rebuilt (image cache — run `./container/build.sh` again with `--no-cache` if suspicious).
+
+## Removal
+
+1. Delete the `"gmail"` entry from `mcpServers` and the `.gmail-mcp` entry from `additionalMounts` in each group's `container.json`.
+2. Remove `'mcp__gmail__*'` from `TOOL_ALLOWLIST` in `container/agent-runner/src/providers/claude.ts`.
+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 nanoclaw`.
+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`.
+
+## 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/qwibitai/nanoclaw/issues/1500) (proxy Gmail/Calendar OAuth tokens through credential proxy) for the Gmail side.
+- **Related PRs:** [#1810](https://github.com/qwibitai/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.

.claude/skills/add-opencode/SKILL.md

@@ -208,7 +208,7 @@ 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.
 

.claude/skills/add-signal/REMOVE.md

@@ -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`.)

.claude/skills/add-signal/SKILL.md

@@ -0,0 +1,318 @@
+---
+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.
+
+```bash
+# macOS
+launchctl unload ~/Library/LaunchAgents/com.nanoclaw.plist
+signal-cli -a +1YOURNUMBER updateProfile --name "YourBotName"
+# optionally: --avatar /path/to/avatar.jpg
+launchctl load ~/Library/LaunchAgents/com.nanoclaw.plist
+
+# Linux
+systemctl --user stop nanoclaw
+signal-cli -a +1YOURNUMBER updateProfile --name "YourBotName"
+systemctl --user start nanoclaw
+```
+
+### 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
+
+```bash
+# macOS
+launchctl kickstart -k gui/$(id -u)/com.nanoclaw
+
+# Linux
+systemctl --user restart nanoclaw
+```
+
+## 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
+sqlite3 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=$(date -u +"%Y-%m-%dT%H:%M:%S.000Z")
+sqlite3 data/v2.db "
+INSERT OR IGNORE INTO messaging_group_agents
+  (id, messaging_group_id, agent_group_id, session_mode, priority, created_at)
+VALUES
+  ('mga-'||hex(randomblob(8)), 'mg-GROUPID', 'ag-AGENTID', 'isolated', 0, '$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=$(date -u +"%Y-%m-%dT%H:%M:%S.000Z")
+sqlite3 data/v2.db "
+INSERT OR REPLACE INTO user_roles (user_id, role, agent_group_id, granted_by, granted_at)
+  VALUES ('signal:UUID', 'owner', NULL, 'system', '$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: `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='signal'"`
+3. Service running: `launchctl print gui/$(id -u)/com.nanoclaw` (macOS) / `systemctl --user status nanoclaw` (Linux)
+
+### 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.