FREE CLAUDE.MD LINTER

CLAUDE.md linter: check for bloat and vagueness

Paste or drop your CLAUDE.md. Get a score, flagged vague lines, bloat warnings, and a split suggestion, instantly.

This free CLAUDE.md linter scores a CLAUDE.md file for vague rules ("write clean code", "best practices"), filler that just restates what's already visible in package.json or the folder tree, and overall length bloat against the always-loaded context budget. It flags lines worth rewriting and suggests splitting large sections into per-folder CLAUDE.md files or dedicated subagents. Everything runs in your browser, with no signup and no upload.

CLAUDE.md
Drop your CLAUDE.mdClick or drop a file, no paste needed
Bloat & vagueness score

Paste, drop, or load the demo to score your CLAUDE.md.

FAQ

Why is "write clean code" flagged as vague?

A rule like "write clean code" or "follow best practices" gives an agent nothing to check against: there's no concrete test for whether it was followed. The linter flags these phrases and suggests swapping them for a rule with a verifiable condition, like "every new component gets a corresponding test file" or "never use `any` in TypeScript."

Why does it flag lines that describe my stack?

Claude Code can already see your package.json, lockfile, and folder structure when it reads the repo. A CLAUDE.md line that just says "this project uses Next.js and TypeScript" costs tokens on every single session without telling the agent anything it couldn't find itself. The linter flags this so you can keep only the constraints that aren't obvious from the code.

Why does file length matter for CLAUDE.md specifically?

CLAUDE.md loads in full at the start of every session, unlike a skill or subagent file which only loads when triggered. A long CLAUDE.md is a fixed tax on every conversation. The linter flags files over roughly 3,000 tokens and suggests moving folder-specific detail into nested CLAUDE.md files, or specialized workflows into subagents that load on demand.

What counts as a 'missing non-obvious constraint'?

The linter checks for hard-constraint language like "never", "must", or "do not" paired with a concrete action. A CLAUDE.md with no such language is usually all tone and no enforceable rule. The strongest files state the one thing a model would otherwise guess wrong, like a forbidden pattern or a required check before committing.

Is anything uploaded?

No. Parsing and scoring run entirely in your browser. Nothing is uploaded or stored, and no signup is required.

Where do I get a CLAUDE.md that's already structured well?

AgenticKit ships a CLAUDE.md template plus a 34-agent subagent catalog built for Claude Code, so folder-specific or specialized rules can live in a subagent instead of bloating the root file. See the kits page, or run the Skill Token-Cost Estimator on an individual skill once you've split one out.

This only scores my CLAUDE.md. What about MCP servers and skills?

Run the CLAUDE.md Context Budget Estimator next: it adds your CLAUDE.md's token cost to your installed MCP servers and loaded skills, so you see the full fixed tax a session carries before your first prompt.

By Rakshit Yadav · Published 2026-10-30 · Last updated 2026-10-30. Next: add MCP servers and skills to the estimate, check a skill's token cost, or check which of your rules are actually enforced by a hook.

[ RUNS IN YOUR BROWSER • DROP A FILE OR TAP GENERATE • NO SIGNUP ]

Go further

Liked the free tool? Get the kits.

Engineering, Marketing, or Both: agents, skills, and slash commands you drop into any Cursor project.