What the data says

372 tools, one checklist, live numbers from GitHub. Some of this will surprise you. Every line links to the proof.

No release in 18 months, but not in Hold 5

Quiet does not mean dead, but by our own rule these should move. Pull requests welcome.

Avoid for now: tools in Hold 27

Archived, deprecated or abandoned. Some of them you may still have in a lockfile.

Best of both: easy to use and made for big systems 11

Easy wins: very easy, small job

Most starred on GitHub

  1. OpenCode ★ 222621 · Adopt
  2. Ollama ★ 185098 · Trial
  3. Claude Code ★ 153687 · Adopt
  4. LangChain ★ 148611 · Adopt
  5. Turbopack ★ 143180 · Adopt
  6. Next.js ★ 143180 · Adopt
  7. Go modules ★ 139485 · Adopt
  8. llama.cpp ★ 132621 · Trial
  9. Codex ★ 131399 · Trial
  10. Kubernetes ★ 128580 · Adopt
  11. Deno ★ 108789 · Adopt
  12. Gemini CLI ★ 108590 · Trial

New this year: in Assess 2

Agent skills a small model uses best

Tested with Haiku on 400 skills. See Skills for how.

  1. a11y-audit alirezarezvani/claude-skills · 100/100
  2. adversarial-reviewer alirezarezvani/claude-skills · 100/100
  3. agent-browser vercel-labs/agent-browser · 100/100
  4. agentic-actions-auditor trailofbits/skills · 100/100
  5. agents-md getsentry/skills · 100/100
  6. api-test-suite-builder alirezarezvani/claude-skills · 100/100
  7. appinsights-instrumentation microsoft/azure-skills · 100/100
  8. autoplan garrytan/gstack · 100/100
  9. aws-ami-builder hashicorp/agent-skills · 100/100
  10. azure-deploy microsoft/azure-skills · 100/100
  11. benchmark garrytan/gstack · 100/100
  12. benchmark-models garrytan/gstack · 100/100

Agent skills a small model gets wrong most

GitHub numbers fetched 2026-10-28. They refresh every week.