Decomposes vague instructions into phased plans and executes them autonomously with subagents.
npx skills add https://github.com/yan-labs/yan-skills --skill autopilot-yan-labs
Please help me install this Agent Skill. Skill: autopilot Source: https://github.com/yan-labs/yan-skills/tree/main/autopilot Command: npx skills add https://github.com/yan-labs/yan-skills --skill autopilot-yan-labs
High-intent search queries and answers about installing and using this skill.
Invoke autopilot with a broad instruction like "fix this bug and deploy". It classifies the task, builds an XML phase plan with objective gates, then dispatches subagents through investigation, implementation, deployment, E2E testing, and review without asking for confirmation.
Every phase requires an objective pass/fail signal such as a zero exit code, green CI run, or browser-verified behavior. The loop cannot exit until all mandatory phases show completion markers with evidence in the progress.md state file.
Yes. It detects the host platform and adapts: Claude Code uses /loop for iteration, while Codex uses state-only execution with scheduled recovery and /goal only when a persistent goal is explicitly requested.
Failures are classified into types like missing-context, wrong-approach, or environment-issue before any retry. Retries must change strategy; three repeats of the same failure signature stop blind retrying and trigger re-investigation.
The maker-checker split prevents an agent from grading its own work. Verification subagents receive only the diff, deployment URL, and acceptance criteria, never the implementer's reasoning, so their judgment stays independent.
Avoid it for tasks lacking any verifiable completion signal or blocked solely by inaccessible external systems. It also requires a safe bounded next action; if investigation proves none exists, it stops and reports rather than spinning.
MohamedSayedKFiles structured GitHub issues about OMNIGRID framework bugs to MohamedSayedK/omnigrid.
baekenoughCollect and categorize release findings into actionable follow-up recommendations.
jonnathan-lsGuide production deployments with rollback planning and health checks.
crbnosAutomate work item loops through doer, gate, judge, and ledger phases.
