Detect stale paths, dead npm scripts, outdated framework patterns, and context rot before they make your coding agents expensive and wrong.
Two peer-reviewed studies in 2026 quantified what many teams had experienced: stale AGENTS.md files make your agents worse and more expensive.
"LLM-generated context files reduced task success by 2–3% while increasing cost by over 20%."
— ETH Zurich, ICSE 2026"None of the major coding agents expose lifecycle hooks for keeping context files accurate. That's a tooling gap waiting to be filled."
— Addy Osmani, Google Chrome, Feb 2026Each checker is independent. Run all of them in seconds, no dependencies needed.
Every path mentioned in your AGENTS.md is verified to exist on disk. Catches renamed directories, deleted packages, restructured monorepos.
Validates that every npm run <script> mentioned
exists in package.json. Supports workspaces and monorepos automatically.
Detects references to packages not in package.json, and flags
deprecated packages like moment, request,
tslint.
Catches outdated patterns: NgModules in Angular 14+, ReactDOM.render() in React 19, getInitialProps in Next.js App Router, CommonJS in ESM projects.
Checks for recommended sections (Setup, Testing, Build), context bloat, unresolved TODOs, and outdated year references.
When multiple context files exist (AGENTS.md + CLAUDE.md, etc.), flags conflicts: different test commands, package managers, or paths documented in one file but not the other.
Every run produces a 0–100 score with a letter grade. Errors cost 15 pts, warnings 7, infos 2. Set thresholds in CI.
Set a minimum score in CI. Fail the build before agents start producing costly wrong answers.
| Score | Grade | Meaning |
|---|---|---|
| 90–100 | A | Fresh and accurate — ready for agents |
| 80–89 | B | Minor issues, still reliable |
| 70–79 | C | Some stale references, review recommended |
| 50–69 | D | Significant context rot |
| 0–49 | F | Agents may produce incorrect or costly outputs |
No install required. Works with any repo that has a context file.
Detects your stack (React, Next.js, Vue…), test runner, and package manager. Pre-populates sections from package.json.
Lints each file independently, then cross-checks for conflicts between them. Pass a path to lint a specific file.
Walk through each issue and accept or skip the suggested fix. Changes are applied atomically — all at once or not at all.
The weekly schedule is intentional — context rot happens even when AGENTS.md hasn't changed.