Stats
Actions
Tags
From shannon
Refusal-driven retry loop around /shannon:cook. Preserves Iron Rule + refusal-discipline pattern.
How this command is triggered — by the user, by Claude, or both
Slash command
/shannon:autopilot <task> [--max-attempts N]The summary Claude sees in its command listing — used to decide when to auto-load this command
# /shannon:autopilot Fully autonomous execution. Wraps `/shannon:cook` in a refusal-driven retry loop. ## Inputs - `<task>` — task description or plan path - `--max-attempts N` — retry cap (default 3) ## Behavior For attempt 1..N: 1. Invoke `/shannon:cook <task>` (which spawns `Task: shannon:executor` with embedded validation). 2. After cook returns, check `Skill: completion-gate` verdict: - `COMPLETE` → exit success. - `REFUSED` → read `REFUSAL.md`, parse cited blockers, build remediation prompt for next attempt. 3. If attempt == max-attempts AND still REFUSED → emit final REFUS...
Fully autonomous execution. Wraps /shannon:cook in a refusal-driven retry loop.
<task> — task description or plan path--max-attempts N — retry cap (default 3)For attempt 1..N:
/shannon:cook <task> (which spawns Task: shannon:executor with embedded validation).Skill: completion-gate verdict:
COMPLETE → exit success.REFUSED → read REFUSAL.md, parse cited blockers, build remediation prompt for next attempt.plans/reports/autopilot-<run-id>-REFUSAL.md; exit failure.Refusal is a feature, not a bug. No override flag. No force-complete (per Skill: refusal-discipline).
Skill: completion-gateSkill: refusal-disciplineSkill: autopilot-runner (this command IS the autopilot runner)Task: shannon:executor (delegated via cook)completion-gate returns COMPLETE/shannon:autopilot "Add SSO with Okta to admin panel"
/shannon:autopilot plans/260528-feature-y/ --max-attempts 5
npx claudepluginhub krzemienski/shannon --plugin shannon/autopilotRuns autonomous security hunting on a target — scope check, recon, surface ranking, exploit testing, validation, and report drafting with configurable checkpoint modes.
/autopilotRuns autonomous autopilot mode: iterates team-exec -> team-verify -> team-fix loops until acceptance criteria pass or blocked, producing loop tables, status, and checkpoints.
/autopilotRuns an autonomous platform improvement cycle: recon, triage, fix, report. Default mode performs recon and triage; --fix enables automated fixes with human-gated deploys.
/autopilotOperates a gated autopilot for AI-driven workflows: start runs, view inbox of pending human signatures, approve or reject cases, and inspect run traces. Also supports list and status subcommands.
/autopilotOrchestrates multi-session autonomous workflows with configurable kill-switches, session limits, and headless mode.
/autopilotExecutes autonomous phase loop: chains planning, execution, verification, review of code changes; pauses at human decisions; offers PR after final review.