swe-review-reviewlisted

# swe-review-review > Wraps `swe_review.ReviewSkill` (and the underlying `ReviewerSubAgent` + `ExplorerSubAgent`). ## When to use Use this skill when the user gives you: - a real software issue (text or issue link) — as **change context / intent** - a candidate PR (diff / patch / title / body) … and you need to judge whether the change is worth merging and produce structured feedback. Do **not** use this skill if the user only wants a chat reply, a quick sanity check, or has not provided an issue+diff. ## Prompt styles (`prompt_style`) | style | focus | |---|---| | `engineering` (default) | **高级代码审查**：审查对象是 Change 而不是 Bug。8 个维度（Design 20 / Maintainability 15 / Consistency 15 / Simplicity 10 / Readability 10 / Testability 10 / Risk 10 / Change Scope 10）100 分制评分；P0–P4 证据绑定 findings；Hard Gate；4 档决策。 | | `concise` | legacy bug-fix-centric：判断 patch 是否修复 issue 根因。 | | `detailed` | legacy bug-fix-centric：Step 1→6 workflow + symptom-fix detection rules。 | Engineering mode core rules (enforced by the system prompt): - 先理解再评价：Diff → Change Surface → Context → Call/Data Flow → Architecture → Intent → Judgment。 - Finding 必须有 Location / Observation / Why It Matters / Evidence / Recommendation / Severity / Confidence；禁止低价值评论。 - 反事实验证 + Project Baseline：Project Convention > Generic Best Practice；个人偏好不产生 Finding。 - 同时识别 Over-engineering 与 Under-engineering；复杂度与业务复杂度匹配。 - Hard Gate（严重架构/安全/资源/维护性问题、无理由大重构）→ 无论总分一律 BLOCK。 - 没有问题就明确说没有问题，不为凑数制造 Finding。 ## Capabilities - Repository-gr