daemon-restart

name	daemon-restart
description	Restart the cookiedclaw Claude Code session via the gateway's `restart_runtime` MCP tool. Use after installing a new skill / plugin / MCP server (anything that's only discovered at startup), or when the agent itself asks to be reborn. Requires `/cookiedclaw:setup` to have been run first.
allowed-tools	Bash(systemctl --user is-active cookiedclaw-gateway) Bash(systemctl --user status cookiedclaw-gateway --no-pager) mcp__cookiedclaw__restart_runtime

Restart cookiedclaw

Restarts the cookiedclaw Claude Code session. The current process (and you, the agent inside it) gets terminated by the gateway's supervisor, then a fresh CC starts with the same plugin, channel, and identity files.

Pre-flight

Confirm the daemon mode is set up:

systemctl --user is-active cookiedclaw-gateway

If this prints anything other than active, the gateway isn't running and there's nothing to restart. Tell the user to run /cookiedclaw:setup first.

Warn the user, then restart

Send a Telegram message warning the user, because the next thing that happens is your own death:

🔄 Restarting cookiedclaw — I'll be back in a few seconds. Conversation continues via --continue; identity files (IDENTITY/USER/SOUL) reread on boot.

Then call the gateway's restart_runtime MCP tool:

mcp__cookiedclaw__restart_runtime { "reason": "skill/plugin reload via /cookiedclaw:daemon-restart" }

The tool returns immediately. The gateway delays the actual SIGTERM by ~2 seconds so this skill can return and the warning message above can flush to Telegram before the channel server goes down with the rest of CC. End your turn after issuing the tool call — don't try to send another message; everything past this point happens in the next session.

After restart

The fresh CC process will:

1. Reload all skills, plugins, and MCP servers (this is the whole point)
2. Reconnect to the gateway over MCP, which republishes the bot menu
3. Reload IDENTITY.md / USER.md / SOUL.md from the workspace
4. Resume the previous conversation via claude --continue (baked into the launcher) — recent context survives the restart

The user will not see a "back online" message automatically. If continuity matters, the user pings the bot; the new session picks up the DM and continues the prior thread.

Notes

• This skill restarts only the CC child, not the gateway itself. The Telegram bot, MCP-over-HTTP server, and progress endpoint stay up across the restart — you'll just briefly have an empty liveServers set until the new CC reconnects.
• If the gateway itself is what you need to restart (e.g. to pick up a new gateway binary), that's a systemctl --user restart cookiedclaw-gateway job, not this skill.
• If the restart fails (no liveServers after ~30s), surface systemctl --user status cookiedclaw-gateway --no-pager so the user sees the journal tail.