Menu
Use only when the user explicitly allows bounded unattended execution of an approved Alloy task.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | alloy-autopilot |
| description | Use only when the user explicitly allows bounded unattended execution of an approved Alloy task. |
Autopilot chains /execute and /verify under strict stop conditions. It does not add a scheduler, daemon, or extra persistent agent.
.alloy/tasks/<task-id>/plan.md exists and is approved, or the user approves it in the current message./autopilot <task-id> [--risk low|med|high] [--max-iters N] [--resume]
plan.md and progress.md./execute task at a time.alloy-tdd and alloy-debug as required.progress.md./verify after implementation tasks are complete.Low risk: small scope, known tests, no secrets, no production services.
Medium risk: multi-file changes or generated artifacts. Require review before final DONE.
High risk: migrations, auth, billing, data loss, production infra, or secrets. Do not run unattended unless the user explicitly overrides and provides verification boundaries.
Return status, iterations used, files changed, commands run, gate state, remaining risk, and next safe command.
Use when encountering a bug, regression, failing command, or unexpected behavior before attempting a fix.
Use before planning when requirements are ambiguous, large, user-visible, or likely to cause rework unless decisions are locked.
Use when implementing an approved Alloy plan. Maintains progress.md, enforces TDD/debug discipline, and stops on blocked gates.
Use when a task needs broad source understanding before planning or execution.
Use when turning context or a clear user request into an implementation-ready plan with acceptance criteria, files, commands, gates, risks, and handoff.
Automated end-to-end QA — read acceptance criteria (Azure DevOps or manual), generate Playwright test plan, execute with screenshots/video/GIF, produce acceptance report. Uses Playwright CLI for deterministic, reportable execution.