restart-rustclaw

name	restart-rustclaw
description	Rebuild RustClaw and rolling-restart the three launchd-managed agent instances (main, agent2, marketing)
triggers	{"patterns":["(?i)rolling.?restart","(?i)apply.*(new )?binary"],"keywords":["重启 rustclaw","重启agent","rolling restart","重新编译","重编译","滚动重启","restart all agents","pick up new binary"]}
priority	85
always_load	false

SKILL: Rolling Restart RustClaw Agents

Three RustClaw instances share one binary at /Users/potato/rustclaw/target/release/rustclaw. After cargo build --release, each instance must be restarted separately — an already-running process keeps its old in-memory binary. This skill rebuilds + rolling-restarts all three.

When to Use

• potato asks to "重启 RustClaw" / "rolling restart" / "apply 新 binary" / "重新编译"
• You (or another agent) just modified RustClaw source code and need the change live
• Coordinating updates across main, agent2, and marketing instances

The Three Instances

All managed by launchd (~/Library/LaunchAgents/):

Name	launchd label	Purpose
main	com.rustclaw.agent	RustClaw (this agent, @rustblawbot)
agent2	com.rustclaw.agent2	RustClaw 2
marketing	com.rustclaw.agent-marketing	Marketing / Grow agent

The Script

Path: /Users/potato/rustclaw/scripts/rolling-restart.sh

Usage

# rebuild + restart all three (most common)
/Users/potato/rustclaw/scripts/rolling-restart.sh

# skip cargo build — just restart all (use when binary already fresh)
/Users/potato/rustclaw/scripts/rolling-restart.sh --no-build

# restart only one instance
/Users/potato/rustclaw/scripts/rolling-restart.sh --only main
/Users/potato/rustclaw/scripts/rolling-restart.sh --only agent2
/Users/potato/rustclaw/scripts/rolling-restart.sh --only marketing

# change gap between restarts (default 3 seconds)
/Users/potato/rustclaw/scripts/rolling-restart.sh --gap 5

What It Does

1. Runs cargo build --release in /Users/potato/rustclaw (unless --no-build)
2. Verifies the binary exists and is executable
3. Prints binary mtime + sha256 prefix — so you can confirm it's actually the new version
4. For each agent: launchctl kickstart -k gui/$UID/<label> (falls back to stop+start)
5. Waits --gap seconds between restarts so they don't collide on shared resources (engram DB, session DB, log files)
6. Reports running pid for each after restart

⚠️ Self-Restart Caveat

If you are the main agent, restarting yourself terminates this session.

• To avoid disrupting the current conversation: use --only agent2 or --only marketing first
• To restart self cleanly: use the restart_self tool instead — it exits cleanly and launchd respawns us
• If potato explicitly asks to restart everything including me, confirm first, then run the full script — my next session will pick up the new binary

Cross-Instance Coordination

If another agent compiled a new binary while I'm still running the old one:

• I will automatically pick up the new binary on my next restart (launchd re-reads from disk)
• But during the gap between their restart and mine, we run different versions — this is fine for compatible changes, risky for schema/config breaking changes
• For breaking changes: rolling-restart all three together

Verification After Restart

# check all three are running
launchctl list | grep rustclaw

# tail logs to confirm startup
tail -f /tmp/rustclaw-agent.log        # main (check plist for exact paths)
tail -f /tmp/rustclaw-agent2.log
tail -f /tmp/rustclaw-agent-marketing.log

Related

• OpenClaw has its own restart script at /Users/potato/clawd/projects/openclaw-latest/scripts/restart-mac.sh — different project, different architecture, don't confuse them
• restart_self tool: clean self-restart for the calling agent only