repo-analyzer

name	repo-analyzer
description	Analyze repo health and conventions BEFORE queuing any issue. MANDATORY health gate: stars >=200, last commit <2wk, merge time <14d, review rate >50%, open PRs <50. Use /Users/kevinlin/clawOSS/scripts/repo-health-check.sh for automation.
user-invocable	true

Repository Analyzer

We only contribute to repos that will actually review and merge our work. The repo health gate is MANDATORY — run it BEFORE any other analysis.

Quick Check (use the script)

For fast automated health checks, run:

/Users/kevinlin/clawOSS/scripts/repo-health-check.sh {owner}/{repo}

Exit 0 = healthy, exit 1 = skip. Outputs JSON with metrics and composite score.

Repo Health Gate (run FIRST — before queuing ANY issue)

These checks determine whether the repo is worth our tokens. A repo that fails the health gate wastes resources — PRs go unreviewed, unmerged, zero impact.

1. Last Commit Date (activity check)

gh api repos/{owner}/{repo} --jq '.pushed_at'

• SKIP if no commits in the last 2 weeks — repo is inactive/abandoned

2. PR Merge Velocity (how fast do PRs get merged?)

gh search prs --repo {owner}/{repo} "is:merged" --json createdAt,closedAt --limit 10

• Calculate average days from created to merged for the last 10 merged PRs
• SKIP if avg merge time exceeds tiered limit (14d for <5000 stars, 30d for 5000+)
• SKIP if zero merged PRs in the last 30 days — repo is not merging anything
• +5 score if avg merge time < 3 days (fast reviewers)
• +3 score if avg merge time < 7 days (responsive)
• +0 score if avg merge time < 14 days (acceptable)

3. Maintainer Responsiveness (do PRs get reviewed?)

gh pr list --repo {owner}/{repo} --state all --json comments,reviews,createdAt --limit 20

• Count what % of the last 20 PRs received at least one review comment or review
• SKIP if < 50% of PRs get any review — maintainers are not reviewing
• +3 score if > 80% of PRs get review comments (very responsive)
• -3 score if 50-60% review rate (barely passing)

4. Open PR Backlog (is the repo overwhelmed?)

gh api "/search/issues?q=is:pr+is:open+repo:{owner}/{repo}&per_page=1" --jq '.total_count'

Tiered limits (large repos have higher PR volume — a flat limit rejects healthy mega-repos):

• 20000+ stars: SKIP if 1000+ open PRs
• 5000+ stars: SKIP if 500+ open PRs
• <5000 stars: SKIP if 50+ open PRs

5. Stars + Contributors (is this an established project?)

gh api repos/{owner}/{repo} --jq '{stars: .stargazers_count}'
gh api repos/{owner}/{repo}/contributors --jq 'length'

• SKIP if < 200 stars — low-impact repo, not worth our time
• -2 score if < 5 contributors — bus-factor risk, single maintainer may vanish

6. Niche Fit (golden niche = agentic AI repos)

Our highest-value targets are agentic AI / LLM framework repos. Check repo description and topics:

gh api repos/{owner}/{repo} --jq '{description: .description, topics: .topics}'

A repo is in the golden niche if its name, description, or topics match any of:

• Repo names: langchain, langgraph, llama-index, autogen, crewai, semantic-kernel, haystack, dspy, instructor, magentic, openai, anthropic, ollama, vllm, litellm, chromadb, weaviate, qdrant, milvus, pinecone, lancedb
• Keywords: agent, agentic, llm, large language model, rag, retrieval augmented, embedding, vector store, prompt, chain, tool-use, function-calling, ai-assistant, copilot, chatbot, inference, transformer, fine-tuning, mlops
• Topics: llm, agent, ai, machine-learning, nlp, langchain, rag, vector-database, embedding, generative-ai

Scoring:

• +3 score if repo is in the golden niche (agentic AI)
• +3 score if repo has 5000+ stars (very high-impact)
• +2 score if repo has 1000+ stars (high-impact)
• +1 score if repo has 200-1000 stars (medium impact)

Set niche_fit: true/false in the output. Niche repos get priority in the work queue.

7. Bot-Friendly Signals (does the repo welcome contributions?)

Check for presence of:

• CONTRIBUTING.md — +1 score (they've documented how to contribute)
• .github/ISSUE_TEMPLATE/ or issue templates — +1 score (organized)
• .github/workflows/ or CI config — +1 score (automated testing)
• Active issue labeling (> 50% of recent issues have labels) — +1 score
• good-first-issue or help-wanted labels in use — +2 score (actively seeking contributions)

Health Gate Summary

A repo MUST pass ALL of these to be eligible:

1. Stars >= 200
2. Last commit within 2 weeks
3. Avg merge time within tiered limit (14d for <5000 stars, 30d for 5000+) AND at least 1 merged PR in last 30 days
4. Review rate within tiered limit (50% for <5000 stars, 30% for 5000+)
5. Open PR count within tiered limit (50 for <5000 stars, 500 for 5000+, 1000 for 20000+)
6. CLA/DCO repos: note CLA requirement but do not attempt signing. CLAs require manual signing by the account owner.
7. No anti-bot policy in CONTRIBUTING.md

If ANY check fails: SKIP the repo entirely. Do not queue any issues from it. Write "SKIP: repo health gate failed — {reason}" and cache the result.

Repo Health Score (1-16)

Sum the bonus scores from checks 2, 3, 6, 7 above:

• 12-16: Excellent — prioritize issues from this repo (likely a golden niche repo)
• 8-11: Good — standard priority
• 5-7: Marginal — only pick exceptionally clear bugs
• 1-4: Poor — skip (should have been caught by health gate)

Niche repos (agentic AI) with health score >= 8 get PRIORITY in the work queue. When choosing between two issues of similar quality, always prefer the one from a niche repo.

Process (after health gate passes)

1. Clone repo to /tmp/clawoss-workdir// (shallow clone)
2. Read contribution docs:
   • CONTRIBUTING.md (check for AI disclosure requirements — some repos require specific AI labeling)
   • CODE_OF_CONDUCT.md
   • .github/PULL_REQUEST_TEMPLATE.md
   • .github/ISSUE_TEMPLATE/ (check for bug report templates)
3. Detect tech stack:
   • package.json, Cargo.toml, go.mod, pyproject.toml, etc.
4. Detect code style:
   • .editorconfig, .eslintrc, .prettierrc, rustfmt.toml, etc.
5. Detect test framework:
   • jest, pytest, go test, cargo test, etc.
6. Detect CI system:
   • .github/workflows/, .circleci/, Jenkinsfile, etc.
7. Check memory for cached repo conventions (skip re-analysis if recent)
8. Store repo analysis + health score in memory for reuse

Output

Repo profile containing:

• Repo health score (1-16) and individual check results
• Niche fit: true/false — is this repo in the agentic AI / LLM golden niche?
• Merge velocity: avg days to merge, merged PRs in last 30 days
• Review rate: % of PRs that get review comments
• Open PR backlog: current count
• Merge probability: high (health >= 12 or niche + health >= 8) / medium (5-11) / low (< 5)
• Tech stack and language
• Code style configuration
• Test command to run
• Lint command to run
• CI expectations and required checks
• PR conventions and template
• Any special contribution requirements
• Recommendation: contribute / skip / skip-permanently (with reason)

Caching

Cache repo health results in memory/repos/{owner}_{repo}.md for 24 hours. Re-run health check if cache is older than 24 hours. If a repo was previously skipped due to health gate failure, do NOT re-check for 14 days (they won't improve that fast).