pr-review

This skill guides the review of pull requests against the standards of the MiniMax Skills repository. It solves a consistency-and-quality-control problem for a skills marketplace/monorepo: new skill submissions and edits need both hard structural checks and softer content judgment before merge, and this skill packages both into a repeatable two-phase process.

Phase one runs an automated validation script (validate_skills.py) that enforces hard rules: every skill directory has a SKILL.md, YAML frontmatter is parseable, required fields (name, description) exist, the name matches the directory name, and no high-confidence hardcoded secrets are present (OpenAI-style keys, AWS access key IDs, long Bearer/JWT tokens). ERROR-level checks are merge blockers; WARNING-level items like missing license/metadata are flagged but not blocking. Phase two applies soft quality guidelines through manual review: checking for scope overlap with existing skills, description quality with clear trigger conditions, reasonable reference-file sizes for context-window economy, safe credential handling via environment variables, script hygiene (shebang, requirements.txt, error handling), English-language content, and README/README_zh synchronization for new skills. A checklist summarizes blockers versus flagged items.

Target users are maintainers and reviewers of the MiniMax Skills repository and, more broadly, anyone running a curated skills catalogue who wants a defensible review rubric. Typical use cases are validating a new community skill submission, checking an existing skill for compliance, and enforcing conventional-commit PR titles and one-purpose-per-PR discipline. The skill is inherently defensive: it checks for secrets and insecure credential handling rather than performing any risky operation itself.