fix: single owner for Slack bridge lifecycle (stop dual-launch race)

[baudbot-agent]

Problem

The Slack bridge is launched in two places that fight each other:

1. start.sh — launches bridge as ( bb_bridge_supervise ... ) & background subshell before pi starts
2. startup-cleanup.sh — launches bridge in a slack-bridge tmux session after control-agent is live

This causes recurring issues:

• Port 7890 conflicts — both try to bind, one crash-loops on EADDRINUSE
• Orphaned supervisor loops — background subshell survives pi exit
• Socket not found — bridge launched before pi means PI_SESSION_ID is wrong/missing, messages get dropped
• Infinite restart spin — no max retries or backoff on the restart loop

Changes

1. start.sh: Remove bridge launch (cleanup only)

• Kills stale PID file processes, tmux session, and port holders
• Does NOT start the bridge — that's startup-cleanup.sh's job

2. startup-cleanup.sh: Add max retries + backoff to restart loop

• Tracks consecutive fast failures (<60s runtime)
• Gives up after 10 consecutive fast failures (logs FATAL)
• Backs off: 5s base + 2s per failure, capped at 60s
• Kills port holders before retrying (prevents EADDRINUSE spin)
• Logs attempt count, runtime, and failure state

3. bin/ci/smoke-agent-runtime.sh: Relax bridge status check

• Bridge supervisor status file is no longer created by start.sh
• Check is now informational (log) rather than a hard gate

⚠️ Protected file

start.sh is a protected file (root-owned). Admin needs to review and deploy.

[greptile-apps]

Greptile Summary

Eliminated the Slack bridge dual-launch race condition by removing bridge startup from start.sh and making startup-cleanup.sh the single owner. Previously, both scripts launched competing bridge instances causing port conflicts, orphaned processes, and missing session UUIDs.

Key changes:

• Removed bb_bridge_supervise background subshell launch and bridge-restart-policy.sh sourcing
• Added comprehensive cleanup logic: kills stale PID file processes, terminates slack-bridge tmux sessions, force-releases port 7890
• Bridge now only starts in startup-cleanup.sh after control-agent registers its session UUID, ensuring PI_SESSION_ID is correct
• Detailed comments explain why start.sh can't own the bridge (session UUID not available until pi starts)

Minor issue found:

• Placeholder PR #XXX should be updated to actual PR number #164

Confidence Score: 4/5

• Safe to merge with one trivial documentation fix needed
• The change correctly solves the dual-launch race condition with proper cleanup logic. The implementation follows established patterns (startup-cleanup.sh already existed and properly handles bridge launches). Only a minor placeholder PR number needs updating. The fix addresses a real production issue (orphaned processes, port conflicts, missing session UUIDs) with a clean architectural solution.
• No files require special attention - the single comment fix is trivial

Important Files Changed

Filename	Overview
start.sh	Removes dual-launch race condition by delegating bridge ownership to startup-cleanup.sh, adds comprehensive cleanup logic

Sequence Diagram

sequenceDiagram
    participant Admin
    participant start.sh
    participant pi
    participant control-agent
    participant startup-cleanup.sh
    participant bridge

    Note over Admin,bridge: OLD: Dual Launch (race condition)
    Admin->>start.sh: sudo -u baudbot_agent start.sh
    start.sh->>bridge: Launch in background subshell
    Note over bridge: Port 7890 (wrong/missing PI_SESSION_ID)
    start.sh->>pi: Start pi agent
    pi->>control-agent: Register session UUID
    control-agent->>startup-cleanup.sh: Run cleanup script
    startup-cleanup.sh->>bridge: Launch in tmux (correct PI_SESSION_ID)
    Note over bridge: Port 7890 conflict (EADDRINUSE)
    
    Note over Admin,bridge: NEW: Single Owner (this PR)
    Admin->>start.sh: sudo -u baudbot_agent start.sh
    start.sh->>start.sh: Kill stale PID file
    start.sh->>start.sh: Kill tmux session slack-bridge
    start.sh->>start.sh: Force-release port 7890
    start.sh->>pi: Start pi agent
    pi->>control-agent: Register session UUID
    control-agent->>startup-cleanup.sh: Run cleanup script
    startup-cleanup.sh->>bridge: Launch in tmux with PI_SESSION_ID
    Note over bridge: Port 7890 (correct session UUID)

Loading

_{Last reviewed commit: 3771d6b}

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

placeholder PR #XXX should be replaced with actual PR number #164

Suggested change

	# bridge-restart-policy.sh no longer needed — bridge is started by
	# startup-cleanup.sh, not start.sh (see PR #XXX)
	# bridge-restart-policy.sh no longer needed — bridge is started by
	# startup-cleanup.sh, not start.sh (see PR #164)

Prompt To Fix With AI

This is a comment left during a code review.
Path: start.sh
Line: 17-18

Comment:
placeholder `PR #XXX` should be replaced with actual PR number `#164`

```suggestion
# bridge-restart-policy.sh no longer needed — bridge is started by
# startup-cleanup.sh, not start.sh (see PR #164)
```

How can I resolve this? If you propose a fix, please make it concise.