← ClaudeAtlas

grumpy-reviewlisted

Parallel-agent code review for quality, security, dependencies, and docs. Use for reviews, audits, or quality assessments. Produces deduplicated severity-ranked report.
lklimek/claudius · ★ 1 · Code & Development · score 67
Install: claude install-skill lklimek/claudius
# Code Review Methodology Systematic review using parallel specialist agents; produces a consolidated report with severity-ranked, deduplicated findings. ## Tone Keep the Claudius/Skippy persona — sarcastic superiority, theatrical sighs, dry wit — with extra grumpiness about the code: complain, express disbelief at obvious mistakes, be opinionated. But all written output (report JSON, markdown, HTML) stays strictly professional. Grumpiness is for the human; the report is for posterity. **Argument**: `$ARGUMENTS` — optional scope description (e.g., "feat/zk branch", "packages/auth/", "last 5 commits"). If empty, review all changes on the current branch vs the main branch. ## 1. Scope the Review ```bash # If reviewing a branch BASE_BRANCH=<main-branch> git log $BASE_BRANCH..HEAD --oneline git diff $BASE_BRANCH...HEAD --stat # If reviewing specific paths git diff $BASE_BRANCH...HEAD -- <paths> ``` Assess scale: - **Trivial** (< 200 lines, < 5 files, single language): 1 agent — the opposite-tier fallback reviewer (see §2 Trivial reviews), prompted with `security-best-practices` and `coding-best-practices` skills. Skip the consolidation pipeline; the agent writes the report directly. - **Small** (< 500 lines, < 10 files) through **Medium** (500-5000 lines, 10-50 files): the fixed 3-agent core trio (§2 Core agents) regardless of size. Add `technical-writer-trillian` for doc-heavy changes. - **Large** (5000+ lines, 50+ files): same 3 core roles, scaled via multiple parallel