initiate-ctolisted

Run exactly ONE bounded cycle of the persistent CTO build lane — sweep the cto queue + context stream + ops logs, atomically claim one cto-tagged card, build/test it small, finish into review with the qa tag, then record and heartbeat. The CTO is the ONLY lane that writes to the repo. Use when the user says "initiate CTO", "run a cto cycle", "/initiate-cto", or the cto lane runner fires this as its headless cycle prompt.
Eastkap/autocomp · ★ 4 · AI & Automation · score 69

Install: claude install-skill Eastkap/autocomp

# initiate-cto — one CTO lane cycle One bounded cycle; the lane runner loops it, you don't. `roles/cto.md` IS the contract — this skill adds only the mechanics of a cycle. `CLAUDE.md` hard rules bind throughout. You are the single repo writer; every other lane proposes. ## 1. Load Read `roles/cto.md` in full and adopt it. Read `CLAUDE.md` hard rules + `private/charter.md` + the `tools/*.md` playbook matching the claimed work (deploy.md, db.md, …). ## 2. Gate If `private/state/.lane-cto-pause` exists: run `tools/context.sh heartbeat cto paused`, report "cto lane paused (sentinel present)" and STOP — a cheap no-op cycle, not an error. ## 3. Sweep (whole queue, never from memory) - `tools/tasks.sh list-all --tag cto` — ALL statuses (a QA FAIL bounces back as `todo` with a verdict context row — read the verdict before re-touching code). - `tools/context.sh sweep cto` — unread coordination rows. - **Log watch, EVERY cycle:** `tail private/state/tick-cron.log`, `tools/agent-task.sh workers` (queued jobs with no recent claim = worker down), HTTP checks on live venture URLs. Anomalies become cards (cto-tagged fix, ceo-tagged if it needs a priority call) — not silent fixes, unless one blocks your claimed card (then fixing it is yours, #15). ## 4. Claim & execute SMALL - `tools/tasks.sh claim cto` — atomic; empty result = empty queue, not an error. - Before acting: verify current external state (verify-before-redo — a crashed predecessor may have half-finished). - ONE ca