autoloop

name	autoloop
description	Use when running or operating the Autoloop orchestration runtime, including starting new loops, resuming tasks, phase-scoped runs, troubleshooting stuck-looking runs, and interpreting run artifacts. Do not use for unrelated coding tasks.

Autoloop skill

Use this skill for tasks that involve executing Autoloop itself (autoloop or python -m autoloop), reading its run artifacts, or guiding users through Autoloop workflows.

Purpose

Autoloop orchestrates multi-stage producer/verifier loops (plan, implement, test) with durable run state and explicit completion signals.

Use this skill to:

• start a new Autoloop run
• resume or continue an existing task/run
• run a subset of pairs (plan, implement, test)
• analyze run outputs under .autoloop/tasks/...
• diagnose blocked or incomplete verifier outcomes

Canonical commands

From the Autoloop repository root:

autoloop --help

Typical new run:

autoloop \
  --workspace /path/to/target/repo \
  --intent "<user goal>"

Resume task:

autoloop \
  --workspace /path/to/target/repo \
  --task-id <task-id> \
  --resume

Execution policy

• Never kill a running Autoloop process just because output appears idle.
• Only terminate Autoloop if the user explicitly asks, or if there is clear evidence of an unrecoverable local failure.
• If a run appears stalled, inspect logs and artifacts first and report status instead of interrupting execution.

Stalled-looking run checklist

1. Confirm the process is still alive.
2. Check recent updates in run artifacts:
   • .autoloop/tasks/<task-id>/runs/<run-id>/raw_phase_log.md
   • .autoloop/tasks/<task-id>/runs/<run-id>/events.jsonl
3. Verify whether Autoloop is waiting on a verifier decision or clarification.
4. Prefer --resume or follow-up intent over force-killing.

Guardrails

• Keep work scoped to the user-selected workspace (--workspace).
• Preserve git history unless the user explicitly requests otherwise.
• Use --no-git only when git is unavailable or explicitly requested.
• Report exact commands executed and key artifact paths when summarizing outcomes.