full-stop
/full-stop
Dopina zmianę od weryfikacji po gotowy do scalenia pull request: przegląd przez inny model, poprawki CI i eskalacje, nie scalając sam.
- Autor
- mistakeknot ↗
- Repozytorium
- clavain ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 13.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Dopina zmianę od weryfikacji po gotowy do scalenia pull request: przegląd przez inny model, poprawki CI i eskalacje, nie scalając sam.
Wyzwalacze z opisu
Oryginalny opis (EN)
Use when told "full stop", or when asked to own a change end to end up to a merge-ready pull request. Verify, cross-lab review, open the PR ready, register the watcher, fix CI, escalate, then stop. Never merges.
Jak działa
- Hard limits
- Process
Podgląd SKILL.md
--- name: full-stop description: Use when told "full stop", or when asked to own a change end to end up to a merge-ready pull request. Verify, cross-lab review, open the PR ready, register the watcher, fix CI, escalate, then stop. Never merges. --- <!-- compact: SKILL-compact.md — if it exists in this directory, load it instead of following the full instructions below. --> # Full Stop "Full stop" means: you own this change from where it stands to a merge-ready pull request, then you stop. You do not merge, and you do not wait on a person except through a decision card. **Announce:** "I'm using the full-stop skill: owning this to a merge-ready PR." ## Hard limits - Never merge, enable auto-merge, enqueue a merge, or ask another agent to merge. The merge belongs to whoever the project's rules name (the vizier under the standing approvals, or the owner). - Never force-push, delete a branch, create or change a tag or release, or push the default branch. Push a work branch only. - Never bypass the pre-push guard or any hook. If a gate or permission denies an action, report the denial and stop. - Ask a human only through the project's decision-card channel. A question left in the thread is not asked. - A public repository's PR title, body, comments and footer carry no ruling IDs, card keys, private paths, hostnames or secrets. Determine visibility first (`gh repo view --json isPrivate`); if unknown, treat it as public. Check every payload you post, including copied review output and CI logs. ## Process ### 1. Verify Follow `clavain:landing-a-change` Steps 1–3 (tests, plan compliance, evidence checklist) on a work branch. Commit in logical commits. ### 2. Open the PR ready Push the work branch and open the PR **ready for review, not draft** (`gh pr create`, no `--draft`). Use `/describe-pr` for the body. Then `sonnerie register owner/repo#N`. The PR exists before review because the review receipt is keyed to the PR number and head SHA. ### 3. Cross-lab review loop Run the packaged reviewer against the exact head and base: ```bash dispatch.sh --role cross-lab-review -s read-only --producer-identity <model that authored the change, from its receipt> \ --repo OWNER/NAME --repo-id ID --pr N --head <head sha> --base <base sha> \ --prompt-file review.prompt -o review.out -C <worktree> ``` The prompt ends with "Verdict: PASS|PASS WITH NOTES|FAIL". A bare transcript is not a receipt. Receipts normalize FAIL to BLOCK. - Record every round: round number, head SHA, reviewer model, verdict. A BLOCK is never dropped from the record, even when a later round passes; say how each BLOCK was resolved. - Fix findings with new commits, then re-review at the new head under `clavain:code-review-discipline`: the brief carries the previous open findings verbatim, the reviewed head and base, and the fix diff, and the reviewer verdicts each finding ADDRESSED or NOT ADDRESSED. A fresh reviewer's PASS does not close an unaddressed earlier finding. - Escalation: a disproven premise escalates immediately. Two demonstrated failures request escalation (canon, `docs/canon/reasoning-routing.md`); do not start another autonomous round once a Critical or Important finding has survived a fix. The loop cap is three failed rounds, then escalate before a fourth. If this runs inside the final review of subagent-driven development or executing-plans, their stricter limit governs: one fix round and one scoped re-review, then a human blocker. One confirming review after a pass is allowed. Do not add a commit only to obtain a fresh receipt after a BLOCK; escalate at that head. - Operational failures (quota, auth, timeout) are not verdicts; retry or use the packaged fallback, and report an unavailable seat rather than skipping review. ### 4. CI Watch the PR's checks. Fix failures with new commits. After any commit, re-run verification and the review at the final head. Merge-ready means: checks green, a passing receipt at the current head and base, and nothing changed since. ### 5. Footer Before escalating, set the PR body's review footer with `/describe-pr` rules: rounds, reviewer model, per-round verdicts, final verdict, and the final head SHA. If a review returned BLOCK at any point, the footer says so. ### 6. Escalate and stop Send the PR URL, what merging it does, and the review evidence to the project's escalation channel. State whether a standing merge class covers it or it stays with the owner. If the change touches instruction files, hooks, settings, credentials or CI workflows, say so. Then stop.
Pliki w paczce (2)
- SKILL-compact.md 2 KB
- SKILL.md 4 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · mistakeknot/clavain | 100 | MIT |
| RIJUDEVx/agent-skills | 100 | MIT |
| cravingmaker/skills | 90 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 60b741ca5378 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90ds-ready
Przechodzi cały przepływ ds-ship, ale zatrzymuje pull request gotowy do scalenia przez człowieka i nigdy go nie scala. - 90st-ready
Przeprowadza pełny przepływ st-ship, ale zatrzymuje się na pull requeście gotowym do scalenia przez człowieka i nigdy go nie scala. - 100pr-ready
Doprowadza pull request do gotowości do scalenia: lokalny przegląd, poprawki, reakcje na komentarze i wyniki CI. - 100dev-issue
Wdraża zgłoszenie z GitHuba od początku do końca i przygotowuje pull request gotowy do scalenia. - 86ship
Prowadzi zmianę z gałęzi do scalenia i sprzątania: otwiera lub aktualizuje pull request, przegląda go i scala. - 80pull-request:create
Tworzy pull request, merge request lub change request, także zaraz po zatwierdzeniu zmian w repozytorium.