Skill: task-alive-updates

Purpose

Keep long-running agent tasks visibly alive by emitting periodic heartbeat status updates to stdout. Prevents silent timeouts and gives the operator confidence the task is still progressing.

When to Use

• Any task expected to run longer than 2 minutes
• Background agent loops (e.g., parallel-agents, agent-harness)
• CI-style workflows where no output = assumed dead

How It Works

1. Heartbeat interval: Every 30 seconds of inactivity, emit a status line.
2. Status line format:

   [ALIVE] <timestamp> | step: <current_step> | elapsed: <Xs> | status: <message>

3. Completion signal: On finish, emit:

   [DONE] <timestamp> | elapsed: <Xs> | result: <success|failure>

Usage

In a shell script / agent harness

# Source the heartbeat helper
source .claude/skills/task-alive-updates/heartbeat.sh

# Start heartbeat (fires every 30s in background)
heartbeat_start "running implementation plan"

# ... do your work ...

heartbeat_stop "success"

In Claude task descriptions

Prepend your task prompt with:

<task-alive-updates interval="30" />
Emit [ALIVE] lines every 30 seconds while working on: ...

Implementation Rules

• NEVER suppress stdout during long operations
• Always call heartbeat_stop in a trap/finally pattern
• The heartbeat message should reflect the current step name, not a generic string
• Timestamps must be ISO-8601

Integration with parallel-agents

When used inside parallel-agents, each worker emits its own heartbeat prefixed with its worker ID:

[ALIVE][worker-2] 2025-01-15T10:23:45Z | step: test-suite | elapsed: 47s

Files

• SKILL.md — this file
• heartbeat.sh — bash helper (source this in shell-based agents)

Example Output

[ALIVE] 2025-01-15T10:21:00Z | step: installing-deps | elapsed: 30s | status: npm install running
[ALIVE] 2025-01-15T10:21:30Z | step: running-tests   | elapsed: 60s | status: 142/300 tests passed
[DONE]  2025-01-15T10:22:15Z | elapsed: 105s          | result: success