Combo Test: Task-Aware Background Processing

Background processes can update Claude task state while the agent continues other work without blocking.

How It Works

The Claude Code task system stores task metadata in JSON files at ~/.claude/tasks/$SESSION/$TASK_ID.json. A background process running in parallel (spawned with & syntax) can write directly to these files after long operations complete. The agent can poll task state via TaskGet and observe updates written by external shells, integrating long-running operations into the task workflow.

Test Setup and Results

The test created task #27 and launched a background shell: (sleep 2 && update task JSON) &. The background process wrote completion metadata and changed the task status to completed while Claude continued. TaskGet immediately reflected the change, validating the pattern.

What This Enables

1. Async build integration: Long-running builds update task status independently
2. External process reporting: Shell scripts or external systems signal completion
3. Workflow composition: Tasks can spawn async work and poll for results without CLI blocking

The pattern is immediate once you have session and task ID:

(
  do_long_operation
  cat > ~/.claude/tasks/$SESSION/$TASK_ID.json << EOF
  {"id": "$TASK_ID", "status": "completed", ...}
  EOF
) &

Caveats

Session ID discovery and collision risk remain open; task ID alone is insufficient. Concurrent writes to the same JSON file lack locking and could corrupt state under heavy parallelism. The approach assumes cooperative, non-malicious background processes. Updates are fire-and-forget; error handling and retry logic remain the caller's responsibility.

Epistemic status: validated by runtime test. Pattern functional; production use requires session context and safe ID generation.