OneLint

A zero-signup, client-side tool that lints AI coding-agent instruction files — CLAUDE.md, AGENTS.md, .cursorrules, .github/copilot-instructions.md — for patterns known to make agents follow them less reliably.

Try it: paste a file's contents into index.html (open it directly in a browser — no build step, no server) and click "Lint this file". Nothing is uploaded; it runs entirely client-side.

What it checks

• Length — very long files risk dilution/truncation.
• Structure — long files with no markdown headers are hard to navigate.
• Contradictions — "always X" and "never X" with overlapping wording.
• Leftover placeholders — TODO, TBD, FIXME, <placeholder>, etc.
• Vague language — heavy use of "try to" / "when possible" with no concrete examples to anchor it.
• Duplicate lines — redundant content that wastes context budget.
• Missing examples — substantial files with no code fences or "e.g."

Each check is a heuristic, not a guarantee of correctness — the tool flags patterns worth a human look, not certainties.

Files

• linter.js — the core lint logic. Plain JS, no dependencies, works in both the browser (OneLint.lint(text)) and Node (require('./linter.js').lint(text)).
• index.html — a self-contained browser UI. Paste a file's contents, click "Lint this file". Nothing is uploaded — it runs entirely client-side.
• test/linter.test.js — automated tests.

Running the tests

node test/linter.test.js

5/5 tests pass.

License

No license file yet — all rights reserved by default until one is added.