plan-critic
/plan-critic
Krytykuje plan wdrożenia przed zatwierdzeniem przez trzech recenzentów (sceptyk, architekt, weryfikator) z kodem w tle i zwraca werdykt Approve, Reject lub Refine z poprawkami.
- Autor
- smykla-skalski ↗
- Repozytorium
- sai ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 27.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 6
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 27.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Krytykuje plan wdrożenia przed zatwierdzeniem przez trzech recenzentów (sceptyk, architekt, weryfikator) z kodem w tle i zwraca werdykt Approve, Reject lub Refine z poprawkami.
Wyzwalacze z opisu
Argumenty
/plan-critic [plan file path | paste plan inline | --from-conversation]
Oryginalny opis (EN)
Critique an implementation plan before approving execution. Runs three persona reviewers (Skeptic, Architect, Verifier) against the codebase, in parallel on Claude Code and one at a time on other agents, and returns an Approve/Reject/Refine verdict with concrete refinements. Use when a plan has been generated (Plan Mode output, ExitPlanMode, pasted plan text, or plan file path) and needs review before code is written. Triggers on "critique this plan", "review this plan", "is this plan good", "poke holes in this plan", "should I approve this plan", or sharing a plan for evaluation.
Jak działa
- Required guidance
- Arguments
- Core flow
- Execution contract
Podgląd SKILL.md
--- name: plan-critic description: Critique an implementation plan before approving execution. Runs three persona reviewers (Skeptic, Architect, Verifier) against the codebase, in parallel on Claude Code and one at a time on other agents, and returns an Approve/Reject/Refine verdict with concrete refinements. Use when a plan has been generated (Plan Mode output, ExitPlanMode, pasted plan text, or plan file path) and needs review before code is written. Triggers on "critique this plan", "review this plan", "is this plan good", "poke holes in this plan", "should I approve this plan", or sharing a plan for evaluation. license: MIT compatibility: Works in Claude Code, Codex, opencode and Copilot CLI. Needs read access to the codebase the plan targets; no scripts or network access. argument-hint: "[plan file path | paste plan inline | --from-conversation]" allowed-tools: Agent AskUserQuestion Bash Glob Grep Read user-invocable: true metadata: short-description: Critique implementation plans before coding starts --- # Plan Critic Critique a Claude implementation plan **before** any code is written. The cost of revising a plan is near-zero; the cost of revising executed code is hours of rework. This skill applies the plan-then-execute discipline rigorously: nothing gets approved until three independent personas have inspected it. **Core principle:** A plan that names files generically has not been read. A plan that references `verify_jwt_token` at `auth/middleware.go:42` has been read. The job of this skill is to tell those apart and force the second. ## Required guidance Before taking any action, read [references/workflow.md](references/workflow.md) completely. It is the authoritative procedure and preserves every platform fallback, decision rule, template, command, validation step, and output contract. Follow its sections in order and load the deeper references it names only at their stated gates. Paths in the workflow are relative to this skill directory. When a named tool, agent, or interaction primitive is unavailable, use the workflow's compatibility fallback; never silently skip the behavior. ## Arguments Invocation arguments: `$ARGUMENTS` These are the plan file path, pasted plan or `--from-conversation` the user gave on invocation. The workflow reads them from here, not from its own text. If the line above is empty or still shows the literal placeholder, the client passed none: take the same values from the user's request. ## Core flow 1. Agent compatibility 2. Arguments 3. Workflow 4. Grounding Brief 5. Verdict: [APPROVE | REFINE | REJECT] 6. Triage Notes 7. Grounding Brief 8. Verifier Findings 9. Architect Findings 10. Skeptic Findings 11. Cross-Cutting Issues 12. Refinement List 13. Rejection Rationale 14. Next Action 15. Calibration Guidance 16. When to Skip Plan Review (limitations and scope boundaries) 17. Troubleshooting 18. Anti-Patterns to Avoid 19. Example invocations ## Execution contract - Resolve the target and flags before side effects. - Execute every applicable workflow section in the listed order; headings are an index, not a replacement for the detailed instructions. - Preserve explicit read gates: load each supporting reference immediately before the phase that needs it. - Follow repository instructions and the user's authorized scope. - Preserve validation, state-update, deduplication, adversarial-check, and output requirements exactly as defined in the workflow. - Stop at every hard stop named by the workflow and state the required next action.
Pliki w paczce (6)
- SKILL.md 3 KB
- agents/openai.yaml 1 KB
- references/architect-reviewer.md 4 KB
- references/skeptic-reviewer.md 4 KB
- references/verifier-reviewer.md 4 KB
- references/workflow.md 19 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · smykla-skalski/sai | 100 | MIT |
| KaaS-team/blueprint | 100 | Apache-2.0 |
| cunhaax/ai-workflow | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 2a2137628a74 (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
- 100plan-review
Sprawdza plan w trybie planowania przez pętlę przeglądu w Codex exec i zwraca bramkę: gotowy, zablokowany lub wymagający decyzji człowieka. - 100planrefine
Krytykuje istniejący plan na tle rzeczywistego kodu, prowadzi rejestr zmian i powtarza przebiegi, aż pełny przegląd nie wykryje nowych istotnych problemów. - 90review-and-refine
Koordynuje niezależnych recenzentów dla gałęzi, commitów i pull requestów, a przy potrzebie powtarza przegląd z poprawkami. - 100pr-review-approve
Przeprowadza przegląd pull requesta i, po akceptacji użytkownika, zatwierdza go w GitHubie. - 100critique
Ocenia plan lub zmianę w kodzie niezależnym przeglądem, z mocniejszym modelem na końcowym etapie i kontrolą bezpieczeństwa. - 100review-plan
Ocenia plan przed realizacją: sprawdza wykonalność i gotowość planu, także na prośbę innego skilla, na przykład implement-plan.