菜单
Orchestrer une review parallele multi-angle d'une PR ou d'un composant, en deleguant a des subagents specialises (securite, tests, k8s).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Force a full-file rewrite (Write) instead of patch (Edit) when Edit keeps failing or diff is large. Context rot mitigation — use when Edit has failed ≥2× on same file, or when estimated diff >30% of file size.
STOA PR Guardian — advisory three-axis review (ADR compliance, security, AI code smell) with binary GO/NO-GO verdict + confidence. Never approves, never blocks merge.
Analyze cycle capacity gap and propose backlog items to fill the sprint to ~80% of proven velocity.
Visual roadmap snapshot — theme progress, milestones, velocity. Zero new infra (Linear MCP + velocity.json).
Cycle-aware bidirectional sync between plan.md and Linear. Discovers all cycle tickets, detects drift, updates markers.
Debug GitHub Actions CI pipeline failures. Analyse les logs de workflow, identifie la cause racine et propose un fix.
| name | parallel-review |
| description | Orchestrer une review parallele multi-angle d'une PR ou d'un composant, en deleguant a des subagents specialises (securite, tests, k8s). |
| disable-model-invocation | true |
| argument-hint | [PR-number ou component-path] |
Executer une review parallele complete sur: $ARGUMENTS
Si c'est un numero de PR:
gh pr view $0
gh pr diff $0 --name-only
gh pr checks $0
Si c'est un chemin de composant:
git diff main --name-only -- $0
| Fichiers modifies | Subagent a lancer |
|---|---|
*.py, *.tsx, *.ts, *.rs (code applicatif) | security-reviewer (toujours) |
*.test.*, test_*, *.spec.* | test-writer (evaluer couverture) |
| Tout code sans tests correspondants | test-writer (identifier les manques) |
Dockerfile, k8s/, *.yaml (k8s), nginx*, charts/ | k8s-ops |
docs/, *.md, CLAUDE.md, memory.md | docs-writer |
docs/*.md, blog/**, *.astro (contenu public) | content-reviewer |
Minimum: security-reviewer est TOUJOURS lance. Maximum: 4 subagents en parallele (budget tokens).
Lancer les subagents en parallele via le Task tool:
Combiner les rapports des subagents en un rapport unifie:
## Review Parallele: $ARGUMENTS
### Date: [date]
### Branche: [branch]
### Fichiers modifies: [count]
---
### Securite (security-reviewer)
**Verdict**: Go / Fix / Refaire
[Resume des findings]
### Tests (test-writer)
**Verdict**: Go / Fix / Refaire
[Resume: couverture actuelle, tests manquants]
### Deployment (k8s-ops)
**Verdict**: Go / Fix / Refaire / N/A
[Resume ou "Pas de changements infra detectes"]
### Conformite Contenu (content-reviewer)
**Verdict**: Go / Fix / Refaire / N/A
[Resume: concurrents, clients, prix, reglementations, disclaimers — ou "Pas de contenu public modifie"]
---
### Verdict global: **Go** / **Fix** / **Refaire**
### Actions requises
1. [Action specifique avec fichier et ligne]
2. ...
### Actions suggerees (non-bloquantes)
1. ...
Si l'argument est un numero de PR:
gh pr comment $0 --body "[rapport]"