メニュー
Build TypeScript and restart the ClawDad service. Use when the user says "restart", "rebuild", "bounce", or after making code changes that need to take effect. Does NOT pull code — use /update for that.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Browse the web for any task — research topics, read articles, interact with web apps, fill forms, take screenshots, extract data, and test web pages. Use whenever a browser would be useful, not just when the user explicitly asks.
How to make authenticated API calls to external services. Covers api.sh, cred-exec, auth patterns, available credentials, cloning repos, SSH forwarding, and troubleshooting 401s. Read this before calling any external API or using a CLI tool that needs auth.
Rich content blocks for the ClawDad web UI. Use when responding to web channels (folder starts with "web_") to emit structured blocks — code, cards, tables, stats, progress bars, diffs, alerts, action buttons, files, forms — alongside standard markdown.
Create and manage agent groups and multi-agent teams. Use when the user wants to create a new group, create a team, list groups, delete a group, add/remove agents from a team, or troubleshoot group configuration.
File a forward-looking GitHub issue (enhancement, design debt, follow-up) against dfederspiel/clawdad with the established structured-proposal template. Use when work surfaces a follow-up that isn't a confirmed bug — proposed redesigns, deferred cleanup, dependent issues, capability gaps. NOT for filing bugs (use /report-bug) or for filing without context. Triggers on "file followup", "file a followup", "/file-followup", or "file as an issue" when the topic is forward-looking rather than a defect.
Audit the context layers for a group or agent. Reports which CLAUDE.md files are in play, lists loaded skills with line counts, and flags any skill over the ~150-line compaction threshold where content may be silently truncated. Use when debugging agent drift, before shipping a new skill, or when an agent seems to ignore rules you thought were loaded. Triggers on "context audit", "audit context", "check compaction", "why is the agent ignoring X", or "/context-audit".
| name | restart |
| description | Build TypeScript and restart the ClawDad service. Use when the user says "restart", "rebuild", "bounce", or after making code changes that need to take effect. Does NOT pull code — use /update for that. |
Build and restart the running service. No git operations.
npm run build
If the build fails, show the error and stop. Don't restart with a broken build.
systemctl --user list-units --type=service --state=running 2>/dev/null | grep -oE '(com-nanoclaw[^ ]+|nanoclaw)\.service' | head -1
If a service unit was found:
systemctl --user restart "$UNIT_NAME"
sleep 3
systemctl --user status "$UNIT_NAME" --no-pager | head -5
If no service unit is found (dev mode):
echo "No systemd service found. Stop any running instance and start manually:"
echo " node dist/index.js"
curl -sf http://localhost:3456/api/health | python3 -c "import sys,json; d=json.load(sys.stdin); print(f\"docker={d['docker']['status']} proxy={d['credential_proxy']['status']} image={d['container_image']['status']}\")"
If changes were made to container/agent-runner/src/ or container/ files, the host-side TypeScript build is not enough — the Docker image also needs rebuilding:
CONTAINER_RUNTIME=docker ./container/build.sh
The build script automatically clears warm pool source caches (data/sessions/*/agent-runner-src/). Without a container rebuild, agents will still run the old code.
When to rebuild: Any change to files under container/ (agent-runner, MCP server, skills, Dockerfile). Not needed for changes to src/, web/, or groups/ — those are host-side or volume-mounted.
NEVER use kill on the node process directly. The service has Restart=always — killing it just causes systemd to respawn the old build, creating zombie processes and port conflicts. Always use systemctl --user restart.