Repo Agent Kit
Full benchmark
Pinned snapshot · 2026-08-31

openai/codex AGENTS.md evidence report

A deterministic look at the repository evidence and root instruction file available to a coding agent at commit 13bc770eaf0a.

83/100
Repository
54/100
AGENTS.md
Interpretation note. The root file contains extensive, repository-specific Rust and Codex guidance. It exceeds this benchmark’s focused-root guideline, and the phrase-based checks do not treat every domain-specific rule as a match for a generic category.

Detected

What the root file clearly covers

4 checks

Repository purpose

Stack and runtime

Runnable commands

Architecture map

Not detected

What did not match the rubric

4 checks

Validation loop

Change boundaries

Safety guardrails

Definition of done

Root instruction file

3,141 words across 323 lines

The focused-length check expects 35–900 words and no more than 220 lines. This repository does not pass that non-scoring readability check.

Open pinned AGENTS.md

Repository changed since the snapshot?

Re-run the live public audit.

The live report reads the current public repository directly from GitHub in your browser.

Audit current branch

Other AGENTS.md evidence reports