AGENT SKILL
review-pr
Evidence-based two-phase PR review and cross-model audit in clean context.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- review-pr
- description
- Evidence-based two-phase PR review and cross-model audit in clean context.
02 资源文件 1
- 参考文档SKILL.pl.md1.2 KB
03 SKILL.md 正文
review-pr
Two-phase unbiased PR review and cross-model audit.
Protocol
- Context Isolation: Reset conversation state (
/clearin interactive sessions, or isolated--bare/--ephemeralsession). Do not review code with authoring bias ("grading own homework" antipattern). - Diff Extraction: Fetch target PR/branch diff against base ref using
gh pr difforgit diff base...HEAD. - Cross-Model Adversarial Audit: Invoke an independent reviewer model (Codex, Claude, or independent-reviewer agent) to probe for:
- Correctness, regressions, and silent breakage
- Security, secrets, and OWASP vulnerabilities
- Edge cases, concurrency races, and resource leaks
- Test coverage gaps
- Structured Findings: Every finding must specify: Severity (CRITICAL, HIGH, MEDIUM, LOW), Evidence (file:line), Impact, Minimal fix.
- Self-Healing: Extract recurrent failure patterns and propose updates to project rules (
CLAUDE.md, skills, linters).
04 同仓库更多 Skills
secure-coding
Application security, input validation, and OWASP defense.
★ 0
python-quality
Python implementation and review quality.
★ 0
postgres
Safe PostgreSQL changes.
★ 0
ocr-pipeline
OCR pipeline and evaluation.
★ 0
genealogy-etl
Genealogical ETL and provenance.
★ 0
docker-quality
Container packaging, reproducibility, and Docker security.
★ 0