AgentConfigScore CLI
Deterministic AGENTS.md and CLAUDE.md checks: compares a change with its baseline and blocks regressions in CI with no model.
- 62 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The tool reads instruction files and writes configuration and a CI workflow after `init`
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/agentconfigscoreDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The tool reads instruction files and writes configuration and a CI workflow after `init`
This is third-party code. Review the repository files before installing.
What it does
AgentConfigScore checks agent instruction files locally: `AGENTS.md`, `CLAUDE.md`, and Cursor, Copilot and Gemini rules. Checks follow defined rules with no model call and no runtime dependencies. The tool compares a change with its baseline, reports regressions with a rule ID, an explanation and a file line, and makes CI decisions from reviewed baseline policies. An example from the README: an added remote-shell instruction lowers the score from 100 to 82, points at `AGENTS.md:5` and fails the new-error gate. The `init` command enables new-error blocking and a no-score-decrease rule, so the check can be adopted in a repository whose score is below 100.
Who it is for. Tech leads and repository maintainers whose agent instruction files change through pull requests and need quality control.
Good fit when
- Someone added an unsafe command to `AGENTS.md` in a pull request and it must be caught before merging
- An instruction file was deleted or a directive reversed while the numeric score stayed the same
- A known exception must be accepted with a reason, an expiry and an audit record
- An existing repository scores below 100 and only new errors should block
Not a fit when
- You must judge whether instructions are useful and semantically correct, since the check covers only defined rules
- A score of 100 should serve as a safety guarantee, while it only means no active rule matched
- The repository has no instruction files
Example request
Run agent-config-score diff for this branch and explain which rules regressed and on which line of `AGENTS.md`.Limitations
A score of 100 means no active rule matched, not that instructions are useful, semantically correct or safe. The check relies on defined rules and does not understand the meaning of text. Python with the `agent-config-score` package is required. The `init` command creates `.agentconfigscore.json` and the `.github/workflows/agent-config-score.yml` workflow, which you review and commit by hand. By default it does not overwrite agent instructions or conflicting setup files. The project is small, one author maintains it and releases come often.
How to disable. Delete `.agentconfigscore.json` and the `.github/workflows/agent-config-score.yml` workflow, then run `python -m pip uninstall agent-config-score`.
FAQ
Which files are checked?
`AGENTS.md`, `CLAUDE.md` and Cursor, Copilot and Gemini instruction files. The check runs locally with no model and no network calls.
What does `init` do?
It enables new-error blocking and a no-score-decrease rule and creates the configuration and CI workflow. It does not write agent instructions.
How do I accept a known exception?
A suppression is added with a reason and expiry and stays in a visible audit record. A candidate cannot use its own exceptions, which are taken from the baseline.
Is a score of 100 enough?
No. It means no rule fired and does not confirm semantic correctness. The limits are described in `docs/limitations.md`.
Can I compare against another branch?
Yes. By default the comparison runs without fetching, and `agent-config-score diff origin/main` takes an explicit ref.
Discussion
No questions or reviews yet.
Related
CLI toolsA linter for the agent harness checks a repository's AGENTS.md, CLAUDE.md, CI and hooks and scores six dimensions.
A CLI that analyses a repository in six model phases and writes AGENTS.md or CLAUDE.md, with ExecPlan planning on top.
A 2.4 KB charter for CLAUDE.md and three on-demand skills: the agent stops guessing, verifies results and gates irreversible steps
AGENTS.md instructions and skills with a test-first process that scales to the task, for Cursor, Codex CLI, Gemini CLI.
A skill that analyzes a repository and generates files for AI agents: AGENTS.md, pointers, reviewers, CI and templates
Agentic OS adds a five-step workflow and checks in git hooks and CI so that a step without evidence does not count as done
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| AgentConfigScore (this card) | no | 62 | |
| AgentLint | no | 156 | checks one set of rules; AgentConfigScore compares a change with its baseline and blocks regressions by policy |
| agnix | yes | 479 | an agent configuration linter; AgentConfigScore adds scoring, comparison and an exception audit |
README in short
The check answers whether things got worse. A score drop, a new error and a deleted instruction file give separate findings even when the numbers match.