Agent Friendly Code

Agent Friendly Skill: score the repo you're in

Scores the repo you're in for 9 agents, locally and offline, and suggests a model class.

Install

The vercel-labs/skills CLI adds the skill to every agent you have set up.

npx skills add hsnice16/agent-friendly-skill#v0

Then run /agent-friendly (or however your agent runs skills) from the repo's top folder. You get all 9 scores (Claude Code, Cursor, Devin, Codex, Kimi Code CLI, Gemini CLI, Aider, OpenHands, and Pi) whichever agent runs it.

How it works

  1. The agent runs node <skill-dir>/dist/index.js ., one bundled file that needs only Node and never uses the network.
  2. It scores the current folder and warns if it has no project file (such as package.json, go.mod, a README or .git), so the wrong folder can't quietly score low.
  3. It runs the same 17 checks as this site and scores each agent with the same weights.
  4. It names the agent the repo is best set up for and suggests a model class from the table below. Switching models is up to you.

Which model for which score

A model class, not a provider. Pick the exact model and switch with /model (or your agent's equivalent).

Suggested model class for each score band
BandScoreSuggested model
High≥ 80Frontier: Opus / GPT-5 / Gemini Pro. The repo is set up well enough to use it fully.
Mid≥ 60, < 80Standard: Sonnet / GPT-5 mini / Gemini Flash. A good default; frontier is optional.
Low< 60Small: Haiku / GPT-5 nano / Gemini Flash-Lite. The repo isn't set up to get value from a frontier model.

Optional: score at the start of every session

Add a session-start hook to the matching settings file:

Claude Code · .claude/settings.json

{
  "hooks": {
    "SessionStart": [
      {
        "matcher": "startup",
        "hooks": [
          {
            "type": "command",
            "command": "node .claude/skills/agent-friendly/dist/index.js . --summary"
          }
        ]
      }
    ]
  }
}

Codex CLI · .codex/hooks.json

{
  "hooks": {
    "SessionStart": [
      {
        "matcher": "startup",
        "hooks": [
          {
            "type": "command",
            "command": "node .agents/skills/agent-friendly/dist/index.js . --summary"
          }
        ]
      }
    ]
  }
}

In other agents, run /agent-friendly whenever you want a fresh score.

FAQ

  • Does the skill contact this website?

    No. The scorer is bundled in, so it runs on your computer with no network request, even if this site goes offline.

  • Which agents does it score against?

    The same 9 this site scores: Claude Code, Cursor, Devin, Codex, Kimi Code CLI, Gemini CLI, Aider, OpenHands, and Pi. You get all 9 whichever agent runs it, and it installs in any agent that supports vercel-labs/skills (Cline, Copilot, Continue, Roo Code, Windsurf, Amp, and others).

  • Can it tell me which agent to use for my codebase?

    It tells you which agent your repo is best set up for and what to add for the others. It doesn't test how well each agent writes code: it is a readiness check, not a ranking of agents.

  • How does it pick a model to suggest?

    From the overall score's band (high, mid or low). A well set up repo gets real value from a frontier model; a poorly set up one doesn't, so a small, fast model does as well. The rules are in SKILL.md and favor no provider.

  • How do I pin a version?

    Add '#<tag>' to the install command: '#v0' follows the latest 0.x release, '#v0.1.0' stays on that version. It's '#', not '@', because '@' picks a skill by name.

Source

github.com/hsnice16/agent-friendly-skill — MIT-licensed. Its scorer is a copy of this site's, shared with agent-friendly-action .