Skip to content

Instantly share code, notes, and snippets.

@septimlabs-code

septimlabs-code/septim-gist-claude-md-analyzer.md

Created April 20, 2026 07:26
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save septimlabs-code/4da2e760b208db0abf81b1ed59c40790 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save septimlabs-code/4da2e760b208db0abf81b1ed59c40790 to your computer and use it in GitHub Desktop.
Download ZIP
Free tool: audit your CLAUDE.md for length, corporate-speak, and recurring instructions that should be skills
Raw
septim-gist-claude-md-analyzer.md

I built a free tool that audits your CLAUDE.md file

If you've been using Claude Code for more than a month, your CLAUDE.md has probably grown into a 2000-line wall of instructions, aspirations, and repeated copy-pasted sections. Mine did.

I wrote a free client-side tool that audits a CLAUDE.md file and tells you what's wrong:

septimlabs.vercel.app/tools/claude-md-analyzer

What it checks

Paste your CLAUDE.md. It runs 4 checks entirely in your browser (no upload, no account, verify in DevTools):

  1. Length. >1500 lines = too heavy, drill-ify it. Every session reads the whole thing.
  2. Instruction ratio. How many lines are actual imperatives ("review X", "check Y") vs aspirational fluff ("we value excellence"). Aspirational lines burn tokens without changing behavior.
  3. Corporate-speak. Terms like leverage, synergy, revolutionary, game-changing, seamless, best-in-class get flagged. Claude can't act on them, and they dilute the lines that do matter.
  4. Recurring-instruction patterns. Detects about 25 common instruction patterns ("review my PR", "check this migration", "find test gaps", "write a changelog", etc.) and flags which should be moved out of CLAUDE.md into reusable skills at ~/.claude/skills/.

Why "move them to skills" matters

A skill in Claude Code enforces scope, output shape, and sequence. A freeform instruction in CLAUDE.md doesn't. That means: review this PR with a 40-line CLAUDE.md block produces different output each time, because Claude interprets the block freshly every session. use the pr-review-comprehensive drill produces the same structured output every invocation because the drill file encodes the constraints.

The analyzer is literally a live demo of this: the 4 checks it runs are the kinds of things a skill does. Enforced scope (only these 4 checks), structured output (tiles with scores), sequence (always in the same order). Same format every time.

The product

This is a free tool. If you find it useful and want 25 pre-built skills like this already written and tuned for Claude Code, Septim Drills is $29 lifetime, private GitHub repo, covers PR review / test gaps / migration safety / security triage / changelog / launch copy and 20 more.

3 free samples to try first: github.com/septimlabs-code/septim-drills-samples.

Tonight only: Drills + Vault (encrypted dev-secret vault) for $39 — saves $19 vs $58 separate. septimlabs.vercel.app/tonight. Expires midnight ET.

Try it

Paste in your biggest CLAUDE.md. If it gets flagged "heavy," you probably have 5+ drill candidates living in it.

→ Run it on your CLAUDE.md

Septim Labs — free tools + premium packs for Claude Code users. April 20, 2026.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment