skillzer

When asked to review a PR or branch ("review this PR", "PR review", "argus", before merging), fan the diff out to parallel read-only reviewers and return a severity-ranked verdict; optionally post inline via gh.

When the user has reviewer feedback on a GitHub PR (human or bot like gemini-code-assist) and wants the pertinent items fixed — "j'ai des retours sur la PR", "résous/corrige ce qui est pertinent", "lis les commentaires de la PR via gh" — triage every comment thread via gh, apply only the pertinent fixes (drop nits), then reply to and resolve each triaged thread. Not for publishing our own review findings — that is argus-review --post.

Audit Claude Code agents, skills, and commands for quality and production readiness. Use when evaluating skill quality, checking production readiness scores, or comparing agents against best-practice templates.

Enforce project coding conventions: reduce duplication, keep naming/file placement consistent, and minimize regression risk. Includes React component and layout guidelines. Use when writing or reviewing code.

Generate a Product Requirements Document from user instructions using the PRD template. Outputs a human-readable self-contained HTML file plus a markdown source for implementation tooling. Includes a macro-level task checklist at the end.

After modifying code, map all consumers of changed symbols and flag unhandled mismatches. Not for debugging (use systematic-debugging).

Review changed code (session, current branch, or colleague's branch) for principle violations, reinvented wheels, and dead code. Trigger after coding, before merge, or for PR review.

Review the last 24h of conversations and suggest improvements to CLAUDE.md, skills, and gotchas sections.

Summarizes commits and changes made on the current branch compared to its parent branch. Outputs a concise bullet-point list in English. This skill should be used when the user wants a quick overview of what was done on a branch (e.g. "what did I do on this branch?", "summarize this branch", "what changed since main?").

Review a markdown plan (from EnterPlanMode or /plan) before implementation. Triggers on "review this plan", "check my plan", "is this plan good".

Progressive code exploration — structure first, drill second. Reduces code reading from 10-15k tokens per file to 200-500 tokens.

Before writing any new function, hook, type, or component, search the codebase for existing implementations to reuse.