anivar/contributor-codebase-analyzer

Overview

This skill performs deep-dive code analysis with incremental saving to enable long-running, resumable reviews. It offers two focused modes: Contributor mode for commit-level annual reviews and promotion readiness, and Codebase mode for repository structure, cross-repo relationships, and enterprise governance mapping. It integrates with GitHub (gh) and GitLab (glab) and stores checkpoints under $PROJECT/.cca/ to resume work across sessions.

How this skill works

Contributor mode reads every commit diff from the local git clone, combines platform CLI metadata (PRs/MRs, reviews) and computes metrics such as accuracy rate, bug-introduction signals, and a seven-phase annual review report. Codebase mode scans single-repo structure, enumerates multi-repo dependencies, and produces governance artifacts like a portfolio and debt registry. Each run saves incremental files and .last_analyzed markers so subsequent invocations resume from the gap instead of reprocessing already-saved data.

When to use it

• Run annual performance and promotion readiness reviews for individual contributors
• Compare two or more engineers with evidence-based strength/weakness mapping
• Map repository architecture or generate a repo structure.json for documentation
• Inventory cross-repo dependencies across an org or group for governance
• Resume interrupted analysis without repeating work using checkpoints

Best practices

• Onboard first: run the provided onboarding to detect platform and create .cca config
• Use local git for diffs and blame; rely on gh/glab only for PR/MR metadata to avoid rate limits
• Follow batch sizing guidance: split large commit sets into smaller agents to avoid failures
• Always run the checkpoint check command before a full analysis to determine FRESH/INCREMENTAL
• Keep the $PROJECT/.cca/ directory under versioned backup or secure storage for audit trails

Example use cases

• Perform a 2025 annual review: read every commit diff, compute accuracy, and produce a structured review with development plan
• Compare Alice and Bob for promotion by generating side-by-side evidence from diffs and metric summaries
• Generate codebase/structure.json for a repo to feed architecture diagrams and onboarding docs
• Scan an org to create dependencies.json and a governance portfolio for executive reporting
• Resume a mid-scan run after hitting API rate limits; analysis picks up from last checkpoint