name: background-monitor description: Run long-running tasks with autonomous completion notifications. Spawns an isolated monitoring sub-agent that watches your process and notifies you immediately when done. Use for installs, builds, downloads, or any task >2 minutes. Zero bloat - sub-agent auto-deletes after notification.

Background Monitor

Run long tasks with immediate autonomous completion notifications using monitoring sub-agents.

How It Works

1. Start your long task in background → get sessionId
2. Spawn a monitoring sub-agent that polls the process every 30 seconds
3. When complete, sub-agent parses output and announces results
4. Sub-agent auto-deletes (cleanup="delete") - no bloat!

Key benefit: You get notified in seconds, not 30 minutes (heartbeat cycle).

Usage

Pattern: Monitored Task

# Step 1: Start your long task
result = exec(
  command="./install-script.sh",
  background=True,
  timeout=900  # 15 minutes max
)
sessionId = result['sessionId']

# Step 2: Spawn monitor sub-agent
sessions_spawn(
  task=f"""
You are monitoring background process '{sessionId}'.

LOOP UNTIL PROCESS EXITS:
1. Check: process(action='poll', sessionId='{sessionId}')
2. If still running: reply EXACTLY "HEARTBEAT_OK" and wait for next wake
3. If exited: break loop and proceed to announcement

WHEN PROCESS EXITS:
1. Get output: process(action='log', sessionId='{sessionId}')
2. Parse for success/failure and key details:
   - Exit code (0 = success)
   - Important URLs, credentials, errors
   - Installation location
   - Next steps
3. Clean up: process(action='clear', sessionId='{sessionId}')
4. Announce results as your FINAL message (this gets delivered to user):
   ✅/❌ status
   Key details (URLs, credentials, location)
   Next actions if needed
   
Do NOT say anything after the announcement - it must be your last reply!

CRITICAL: Only reply HEARTBEAT_OK when process is STILL RUNNING. When it exits, announce results instead.
  """,
  cleanup="delete",  # Auto-cleanup when done!
  label=f"monitor-{sessionId[:8]}"
)

# Step 3: Tell user
f"Started installation! Monitoring in background - I'll ping you when complete."

Simplified Helper Function

For repeated use, wrap this pattern:

def monitored_exec(command, name="Task", timeout=900):
    """Start a monitored background task with autonomous notification."""
    
    # Start task
    result = exec(command, background=True, timeout=timeout)
    sid = result['sessionId']
    
    # Spawn monitor
    sessions_spawn(
      task=f"""
Monitor process '{sid}' (task: {name}).

LOOP: Poll process(action='poll', sessionId='{sid}').
- If running: reply "HEARTBEAT_OK" and wait
- If exited: STOP replying HEARTBEAT_OK, proceed to announce

WHEN DONE:
1. Get logs: process(action='log', sessionId='{sid}')
2. Parse: exit code, URLs, credentials, errors, locations, next steps
3. Clear: process(action='clear', sessionId='{sid}')
4. Announce as FINAL message: ✅/❌ {name} complete! [details]

CRITICAL: The announcement must be your LAST reply - nothing after it!
      """,
      cleanup="delete",
      label=f"monitor-{sid[:8]}"
    )
    
    return f"Started {name} (session: {sid[:8]}). Monitoring in background!"

Then just:

monitored_exec("./install-romm.sh", name="RomM Installation", timeout=1800)

Examples

Example 1: Installation Script

result = exec(
  command="bash /tmp/install.sh",
  background=True,
  timeout=1800
)

sessions_spawn(
  task=f"""
Monitor process '{result['sessionId']}' for installation completion.
Poll every 30s. When done, parse output for URLs/credentials/errors and notify user.
Clean up and exit.
  """,
  cleanup="delete",
  label="install-monitor"
)

"Installation started! I'll notify you when complete."

Expected notification:

✅ Installation complete!
- Access: http://192.168.1.228:8080
- Credentials: /root/app.creds
- Next: Log in and configure settings

Example 2: Build Process

result = exec("npm run build", background=True, timeout=600, workdir="~/project")

sessions_spawn(
  task=f"""
Monitor build process '{result['sessionId']}'.
When complete, report exit code, build time, and output location.
  """,
  cleanup="delete",
  label="build-monitor"
)

Expected notification:

✅ Build succeeded! (5m 23s)
- Output: dist/
- Bundle size: 2.3 MB

Example 3: Download + Extract

result = exec(
  command="wget https://example.com/file.tar.gz && tar -xzf file.tar.gz",
  background=True,
  timeout=900
)

sessions_spawn(
  task=f"""
Monitor download/extract process '{result['sessionId']}'.
Report file size, extraction location, time taken.
  """,
  cleanup="delete",
  label="download-monitor"
)

Token Efficiency

Sub-agent cost is minimal:

• Heartbeat every 30 seconds = ~2 calls per minute
• For a 10-minute task: ~20 heartbeat calls
• Each heartbeat is tiny (process poll + HEARTBEAT_OK)
• Total: ~1-2K tokens for the entire monitoring session

Compare to: Running the task foreground and blocking your main session.

Configuration

No config needed - works out of the box!

Optional: Adjust sub-agent concurrency if you run many monitors:

{
  "agents": {
    "defaults": {
      "subagents": {
        "maxConcurrent": 8  // default
      }
    }
  }
}

Troubleshooting

Sub-agent not spawning?

• Check: agents_list - is your agent allowed to spawn sub-agents?
• Check concurrency: agents.defaults.subagents.maxConcurrent

Not getting notifications?

• Verify sub-agent was created: sessions_list --kinds isolated
• Check sub-agent status: sessions_history --sessionKey agent:<agentId>:subagent:<uuid>

Process killed before completion?

• Increase timeout parameter in exec()
• Check system resources (OOM killer?)

Lost output?

• Output is kept in memory until cleared
• For huge output, redirect to file: ./script.sh > output.log 2>&1

Advanced: Custom Parsing Logic

For complex output parsing, be specific in the monitor task:

sessions_spawn(
  task=f"""
Monitor process '{sessionId}' for Docker container creation.

When complete:
1. Parse logs for container ID (grep for "Created container")
2. Extract port mapping (look for "0.0.0.0:XXXX->")
3. Check health status
4. Announce: Container ID, access URL, health status
  """,
  cleanup="delete",
  label="docker-monitor"
)

Pattern: Multi-Stage Tasks

For sequential long tasks:

# Stage 1: Download
result = exec("./download.sh", background=True)
sessions_spawn(
  task=f"""
Monitor '{result['sessionId']}' (download stage).
When done, tell user and START NEXT STAGE:
  exec("./process.sh", background=True) + spawn new monitor
  """,
  cleanup="delete",
  label="stage1-monitor"
)

Each stage automatically kicks off the next!

Integration with AGENTS.md

Already integrated! Your AGENTS.md has the pattern documented.

For one-off tasks without the helper function, just use the full pattern inline.

tl;dr: Start task with exec(..., background=True), spawn monitoring sub-agent with sessions_spawn(..., cleanup="delete"). Sub-agent polls, notifies immediately, and auto-deletes. Zero bloat, immediate notifications! 🚀