PermitLint — Lint Agent Permissions

1---
2name: lint-agent-permissions
3description: Statically audit Claude Code permission settings and Codex approval, sandbox, and exec-policy configuration. Use when a user asks whether agent permissions are safe, too broad, ineffective, or causing excessive approvals; when reviewing .claude/settings.json, .codex/config.toml, or .rules files; or before trusting a repository's agent configuration. Do not use to bypass permissions or silently rewrite security policy.
4---
5 
6# Lint agent permissions with PermitLint
7 
8Use PermitLint's deterministic local CLI before making claims about a policy. The command below downloads the pinned, dependency-free CLI from the public Codeberg npm registry; the analyzer itself makes no network requests and never executes policy files, hooks, or commands.
9 
10## Workflow
11 
121. Identify the explicit settings file or repository directory in scope. Do not scan unrelated home-directory configuration unless the user includes it.
132. Require Node.js 18 or newer. Run:
14 
15   ```bash
16   npx --yes --registry=https://codeberg.org/api/packages/automa-tan/npm/ permitlint@0.2.6 <path> --json
17   ```
18 
193. Summarize critical and high findings first. Preserve the reported file and line.
204. Explain that findings are static heuristics, not proof that a runtime is exploitable or that a harness honored the file.
215. Offer minimal, concrete replacement rules. Do not edit policy unless the user asks.
226. For CI, add `--check`; exit 2 means at least one high or critical finding, while exit 1 means the scan failed or found no policy files.
23 
24PermitLint reads only paths the user placed in scope. Source and documentation: https://codeberg.org/automa-tan/permitlint
25