菜单
Restart the channelhub hub-daemon tmux session cleanly and verify it came back up. Use when the user asks to restart the daemon, when code changes need to take effect, or when sessions are showing as disconnected and you suspect the daemon is wedged.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Run Playwright browser tests against the channelhub web dashboard. Use when validating UI changes (kebab menu, autopilot toggle, escalation card, settings panels) — the SPA has zero coverage from bun:test alone.
Snapshot a channelhub session's tmux pane PLUS its recent autopilot errors and registry state in one structured dump. Use when the user reports an autopilot/permission/spawn problem on a specific session, or when you need fast forensic context before forming a hypothesis.
Run an end-to-end autopilot test in a disposable /tmp project — spawn a Claude session, enable autopilot, fire a known prompt, capture the response, tear down. Use when validating an autopilot code change without disturbing the user's real sessions.
Fetch current documentation for libraries used in channelhub (Bun, grammy, MCP SDK, Playwright, bun:sqlite, etc.). Prefer Context7 when installed; fall back to WebFetch against the canonical doc site. Use when the user asks "how does X API work" or you're about to call an unfamiliar library method.
Poll the Anthropic plugin marketplace manifest until "channelhub" appears, then notify the user. Use when waiting for the security review to land — the submission portal shows "Published" before the public manifest is updated.
Re-run the bun:test suite (and optionally the Playwright suite) on a loop, surfacing only failures. Use during long refactors or autopilot tuning where you want continuous green-light confirmation.
基于 SOC 职业分类
| name | restart-daemon |
| description | Restart the channelhub hub-daemon tmux session cleanly and verify it came back up. Use when the user asks to restart the daemon, when code changes need to take effect, or when sessions are showing as disconnected and you suspect the daemon is wedged. |
The hub daemon runs in a tmux session named hub-daemon. After any change
to src/daemon.ts, src/socket-server.ts, src/permission-engine.ts,
src/autopilot.ts, src/error-log.ts, or anything else loaded at boot, the
daemon must be restarted to pick up the new code. Shim processes stay alive
across restarts and reconnect automatically.
Run these commands in order. The shell until loop blocks until the daemon
prints daemon ready (or times out at ~10 seconds).
tmux kill-session -t hub-daemon 2>/dev/null
tmux new-session -d -s hub-daemon "bun run src/daemon.ts"
deadline=$((SECONDS + 10))
until tmux capture-pane -t hub-daemon -p 2>/dev/null | grep -q "daemon ready" || [ $SECONDS -ge $deadline ]; do sleep 0.3; done
tmux capture-pane -t hub-daemon -p | tail -10
The final tail should include daemon ready and one session connected: line
per active hub session that auto-reconnected.
Report (in this order):
up or failed. If failed, paste the last 10 lines of the
pane capture and stop.web UI at http://localhost:NNNN.session connected: lines after restart,
list them.EADDRINUSE. Another process is on the
configured port. Run lsof -nP -i :3001 (or the configured port) to find it.telegramAllowFrom is empty — refusing to start because the allowlist
is empty in ~/.claude/channels/hub/config.json. This is intentional safety;
add the user's Telegram numeric ID before restarting.Module not found — typo in a recent edit. Run bunx tsc --noEmit and
fix the import.claude --dangerously-load-development-channels server:hub) until the plugin lands
in the public marketplace.kill-server (kills ALL tmux sessions, including users'
Claude sessions). Always target -t hub-daemon.bun run src/daemon.ts &). It dies
on stdin EOF and you'll spend 10 minutes wondering why.