# Agent Readiness Benchmark

The Repo Agent Kit benchmark is a deterministic, commit-pinned snapshot of 10 public development repositories. It measures repository evidence a coding agent can discover before editing: agent instructions, project context, setup reproducibility, tests, automation, and maintenance guardrails.

## Snapshot

- Date: 2026-08-31
- Repositories: 10
- Average score: 62/100
- Root `AGENTS.md`: 4/10
- Agent-ready under this rubric: 2/10

## Method

The audit reads root files plus direct children of `.github` and `.cursor` at an exact Git commit. It assigns 30 points for a root `AGENTS.md` or partial credit for recognized tool-specific guidance, 20 points for instruction quality, 10 each for a root README and build manifest, 8 each for a lockfile and visible root test surface, and 7 each for CI and maintenance guidance.

The static audit does not execute code, inspect nested package quality, measure overall code quality, or predict whether an agent will complete a task successfully. Missing evidence may exist outside the deliberately narrow scan surface.

## Resources

- [Interactive benchmark](https://repoagentkit.com/agent-readiness-benchmark)
- [Downloadable JSON dataset](https://repoagentkit.com/api/benchmark)
- [Run a live public repository audit](https://repoagentkit.com/github-repo-audit)

## Pinned AGENTS.md evidence reports

- [microsoft/vscode](https://repoagentkit.com/agent-readiness-benchmark/microsoft-vscode)
- [openai/codex](https://repoagentkit.com/agent-readiness-benchmark/openai-codex)
- [rust-lang/rust](https://repoagentkit.com/agent-readiness-benchmark/rust-lang-rust)
- [vercel/next.js](https://repoagentkit.com/agent-readiness-benchmark/vercel-nextjs)