Version updated for https://github.com/WouterDeBot/AgentRepoCoach to version v0.2.0.

• This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

AgentRepoCoach is a GitHub Action and CLI tool that evaluates the readiness of a codebase for autonomous AI agents by calculating a composite Codebase Agent Health (CAH) score. It automates the analysis of key aspects like navigability, error quality, decision-making support, testing practices, and module organization, offering actionable insights to improve code maintainability and AI compatibility. This tool helps teams identify and address structural and documentation gaps in their repositories while providing automated thresholds for CI/CD workflows.

What’s Changed

v0.2.0 — Full Language Coverage + Coaching Recommendations

AgentRepoCoach now scores repos in 5 languages and tells you exactly what to fix first.

What’s new

Full language adapters for TypeScript, Go, and Rust

All three previously-stubbed adapters are now fully implemented:

• TypeScript — tsconfig.json/package.json detection, throw-site scanning with multi-line context, JSDoc detection, Jest/Vitest test method extraction
• Go — go.mod detection, errors.New/fmt.Errorf/custom error mapping, Go doc comment detection, Test* function extraction
• Rust — Cargo.toml detection, panic!/Err(Custom) mapping, /// doc comment detection, #[test] attribute detection

All adapters use regex-only analysis (no AST parser dependencies) and implement the full 9-method LanguageAdapter interface.

Coaching recommendations engine

AgentRepoCoach no longer just scores your repo — it coaches you through the fixes. The new coaching engine:

• Analyzes sub-component score gaps across all five components
• Surfaces the top-3 actionable fix tips ranked by weighted impact
• Works in every output format: terminal summary, verbose mode, markdown PR comments, and JSON reports (new coaching array)

Dogfood improvements

AgentRepoCoach now scores 100/100 on its own repo:

• AGENTS.md for agent-friendly codebase navigation
• codebase-map.md for repo structure overview
• cli-manifest.json for CLI discoverability
• docs/architecture.md documenting the system design
• 5 Architecture Decision Records (ADRs)
• Fix hints on all raise sites; docstrings on all public declarations

Bug fixes

• Python adapter _TEST_METHOD_PATTERN was missing re.MULTILINE flag, causing zero test methods to be detected in Python repositories

What’s supported

Highlights

• Still zero runtime dependencies (Python 3.11+ stdlib only, including tomllib)
• Composite Action (no Docker, no slow cold start)
• TOML config (.agentrepocoach.toml) with zero-config defaults
• JSON + Markdown output formats
• fail-threshold input for PR gating
• Output is safe to publish as a CI artifact (no source snippets)

Upgrade

GitHub Action

- uses: WouterDeBot/agentrepocoach@v0.2.0

CLI

pip install --upgrade agentrepocoach

Feedback

Feedback welcome via GitHub Issues and Discussions.

License

Apache 2.0

• WouterDeBot
• GitHub Actions