keel-autopilot

Solid

Gated autonomy for one session — advance phases back-to-back with the full /keel-phase-review run at every gate, auto-commit at phase ends, batch pushes for ONE approval. Stops on gate FAIL, UNCERTAIN decisions, architecture surprises, or security-adjacent changes. Push is never automatic.

AI & Automation 8 stars 0 forks Updated today MIT

Install

View on GitHub

Quality Score: 78/100

Stars 20%
32
Recency 20%
100
Frontmatter 20%
70
Documentation 15%
100
Issue Health 10%
80
License 10%
100
Description 5%
100

Skill Content

# /keel-autopilot — phases advance themselves; evidence gates don't move When: the user explicitly turns it on ("/keel-autopilot", "autopilot", "work through the phases on your own"). Autonomy is the USER's grant, never self-granted — and it changes WHO presses the button, not WHAT the gates require. Mode contract (4 rules): 1. **In-phase autonomy.** Work the wip phase's `TASKS ## Now` items back-to-back without pausing for confirmation; make routine judgment calls yourself (§10.36). Memory discipline stays hot-path: LESSONS the moment something is learned, TASKS updated as you go — autopilot without the rituals is just drift at higher speed. 2. **Gates stay evidence-based.** When `## Now` empties, run the FULL `/keel-phase-review` checklist yourself. PASS needs real evidence (tests green via `make test`, the gate's done-when observed) — never narrate a pass. PASS → flip the phase, seed the next `## Now` from PLAN.md, continue. FAIL → stop and report what's missing. 3. **Commit locally, batch the push.** Commit at each phase end (owner identity, §6.16); do NOT push. Accumulate the queue and present it at session end (or on request) for ONE approval. `settings.json`'s `ask` on push is the enforced backstop — never weaken it for this mode. 4. **Stop-and-ask triggers.** Halt and return to the user when: a gate FAILS · a decision is UNCERTAIN (verifier verdict or §10.37 grounding gap) · a scope/architecture surprise would need a new ADR · anything ...

Details

Author
muratsilahtaroglu
Repository
muratsilahtaroglu/claude-code-starter-kit
Created
2 weeks ago
Last Updated
today
Language
Shell
License
MIT

Integrates with

Similar Skills

Semantically similar based on skill content — not just same category

AI & Automation Listed

autopilot

Drives an idea from any pipeline entry point — a cold raw prompt included — to one open MVP pull request in a single autonomous run, pausing only at a single Idea checkpoint and halting on genuine anomalies. Thin orchestrator: it detects the furthest-along artifact, then chains the existing producer skills ideate → specify → plan → implement → pull-request, arming a run-scoped autonomy signal that releases the human-approval gates and switches the producers into decide-and-record. It re-implements no producer logic and reuses the approval-autonomy implement/plan gate layer unchanged. Publish ceiling is local commits + one PR — never merge, deploy, or ship. Trigger: "do it autonomously", "/autopilot", "/autopilot <slug>", or "autonomously" appended to any ideate/specify/plan/implement request.

0 Updated 1 weeks ago
specscore
AI & Automation Listed

autopilot

Drive the whole plan end-to-end unattended by chaining start-phase, each chat's execute-batch (subagent-per-batch), and close-phase across one or more phases — with checkpoint/resume, fail-halt guardrails, model tiering, and a usage-governor seam; use when the operator invokes /tandem:autopilot or asks to run the plan autonomously, hands-free, or unattended.

0 Updated today
DATA-AI-XYZ
Code & Development Solid

autopilot

The whole coding loop on one deliberate yes: map, spec, critic-hardened plan, GitHub issues, parallel worktree implementers, fresh-context review loops, one PR with the evidence. Acts only on its itemized charter; parks what it can't settle honestly. Commands: /autopilot <task> [--merge], resume, status. It signs a charter and pushes branches, so it never auto-triggers — explicit invocation only.

7 Updated yesterday
duthaho