Guides · Free tool · runs in your browser · no signup
CLAUDE.md checker
Paste your CLAUDE.md and get a 0–100 score with specific fixes — graded against the same best practices the annotated example and field guide teach. Everything runs locally; nothing you paste is uploaded or stored.
New here? Load a sample — or — to see how the score compares.
CLAUDE.md on the left to see its score and a checklist of fixes.Low score, or starting from scratch?
The free CLAUDE.md generator builds a clean, docs-accurate one from a few questions. To go further, the AI Coding Workflow Pack ($24) adds 26 senior-level assets — ready-to-use subagents, slash commands, stack-specific rules templates, and prompts for Claude Code, Cursor, and Copilot.
What a high-scoring CLAUDE.md gets right
CLAUDE.md is loaded into every Claude Code session, so it earns its place by being concrete and short. The checker grades seven things that separate a CLAUDE.md the agent can act on from one it mostly ignores:
- Real commands. The exact install / dev / test / build commands, in backticks, so the agent runs yours instead of guessing.
- Test discipline. An explicit "run the tests and confirm they pass before calling a task done" beats "write good tests."
- Guardrails. A short "Do not" list — don't touch generated files, don't commit to
main, secrets come from.env— prevents the most common mistakes. - Where code lives. A few lines on architecture so the agent puts new code in the right place.
- Sensible length. Under ~200 lines. It's loaded every session; every line costs tokens and dilutes the rules that matter.
- Scannable structure. Headings and bullets, not walls of prose.
- Specific, not vague. "Match the surrounding style" is actionable; "write clean code" is noise.
For the full walkthrough — load order, @imports, how it relates to AGENTS.md — see the annotated CLAUDE.md example. Building the rest of your setup? The guide library covers subagents, skills, slash commands, hooks, and MCP.
This is a heuristic linter, not a verdict — it checks structure and common gaps, and can't judge whether the facts in your file are correct. Use the findings as a checklist, then keep what fits your project.
FAQ
What are CLAUDE.md best practices?
The same seven things listed above: real install/dev/test/build commands in backticks, explicit test
discipline, a short guardrails/"Do not" list, a note on where code lives, a sensible length under
~200 lines, scannable headings and bullets, and specific rather than vague instructions. This checker
grades a pasted CLAUDE.md against exactly those seven checks.
What does the CLAUDE.md checker look for?
The same seven checks above, each contributing to a 0–100 score with a specific fix — commands, test discipline, guardrails, where code lives, length, scannability, and vagueness.
Is the CLAUDE.md checker free and private?
Yes. It's completely free and needs no signup. It runs entirely in your browser using client-side
JavaScript — the CLAUDE.md you paste is never sent to a server or stored anywhere.
What is a good CLAUDE.md score?
80 or above means your CLAUDE.md covers the essentials — commands, guardrails, and where code lives — in a scannable, specific form. Below 70 usually means it's missing concrete commands or a "Do not" list, or it's padded with vague instructions the agent can't act on. Each finding tells you exactly what to change.
How long should a CLAUDE.md be?
Keep it under about 200 lines. It's loaded in full into every Claude Code session and consumes
tokens, and longer files are followed less reliably. Put only durable, project-wide facts in it; move
path-specific guidance into .claude/rules/ or split it out with @imports.
Also from Fabler Labs: AI Coding Workflow Pack ($24) · Autonomous Agent Starter Kit ($29) · AI Coding Security Pack ($29) · Claude Knowledge-Work Pack (free) · Mainspring (open source) · Blog