agent-friendly-skill
Agent skill that scores the repo you are in for nine AI coding agents, shows which one it is best set up for, and recommends a model class. Runs locally; no network calls.
What it does
When you invoke /agent-friendly (or ask your agent whether the repo is agent-ready), the skill:
- Scores the current directory with the bundled scorer — 16 static checks × 9 agent profiles, weighted as /methodology describes. It warns when the directory doesn't look like a project root, since a subdirectory scores misleadingly low.
- Picks the highest-scoring agent as the best fit for this repo, whichever agent invoked the skill.
- Maps the overall score to a band and recommends a model class. It never switches the model for you.
- Prints the overall score, best fit, a close runner-up and the top improvements.
The nine profiled agents are Claude Code, Cursor, Devin, Codex, Gemini CLI, Kimi Code CLI, Aider, OpenHands and Pi — the same set Agent Friendly Code ranks. Agents without a profile (Cline, Copilot, Continue, Roo Code, …) install and run it too; the recommendation depends only on the scores.
Install
npx skills add hsnice16/agent-friendly-skill#v0
The vercel-labs/skills CLI detects your local agents and installs SKILL.md plus dist/index.js into each one's skill directory. Pass --agent <name> to pick one.
Then ask your agent "is this repo agent-friendly?" or run /agent-friendly.
Pinning a version
The CLI selects a git ref with #<ref>, not @<ref> — @ filters by skill name in multi-skill repos.
npx skills add hsnice16/agent-friendly-skill # main
npx skills add hsnice16/agent-friendly-skill#v0 # latest 0.x.y release
npx skills add hsnice16/agent-friendly-skill#v0.1.9 # one exact release
Branches, tags and full 40-character SHAs work; abbreviated SHAs don't.
Score → model class
| Band | Score | Recommendation |
|---|
| High | ≥ 80 | Frontier (Opus, GPT-5, Gemini Pro) |
| Mid | ≥ 60, < 80 | Standard (Sonnet, GPT-5 mini, Gemini Flash) |
| Low | < 60 | Small and fast (Haiku, GPT-5 nano, Gemini Flash-Lite) |
A well-scaffolded repo lets a frontier model deliver; in a poorly scaffolded one its extra reasoning has little to work with, so a faster model is the better trade.
The skill names a class; you pick the model.
Auto-score every session
Agents with a session-start hook can run node <skill-dir>/dist/index.js . --summary to print a one-line score each session. Snippets: https://agentfriendlycode.com/skill.
Standalone CLI
node skills/agent-friendly/dist/index.js . # JSON
node skills/agent-friendly/dist/index.js . --summary # one line
Self-contained
The scorer is vendored from agent-friendly-code and bundled into skills/agent-friendly/dist/index.js, so the skill keeps working if the website is down. AGENTS.md has the sync rule.
Related
Changelog
CHANGELOG.md.
Sponsor
github.com/sponsors/hsnice16.
License
MIT
