CLAUDE.md bloat checker
Paste your CLAUDE.md (or similar) for a quick health score using line-count heuristics aligned with TokenCheat's audit rules: long files and dense rules increase context cost and contradiction risk.
For anyone maintaining agent config files that ship with every request. The scoring compares your file's length and content patterns against what we found analyzing 100 public configs: the median file was 964 tokens across 92 lines, and 24% of files exceeded 200 lines.
Heuristics current as of .
Optional: paste .claude/settings.json to check auto-fixable settings
Context health score
- high-100 pts · length
Empty content.
How to read the results
The score is a heuristic, not a semantic review: it flags length, log-like content, and rule density. That's still useful because waste is concentrated — in the 100-config corpus, the top 10 of 100 files held 97% of the detectable waste. If you score poorly, a handful of deletions usually recovers most of the cost.
FAQ
- How is the score computed?
- Line-count and pattern heuristics: total length (tokens estimated at ~4 characters each), log-shaped content, and density of imperative rules. It costs nothing to run and gives a directional answer, not a verdict on whether your rules are good.
- What actually counts as bloat?
- The #1 anti-pattern in the corpus was session logs pasted inside CLAUDE.md — historical narration the model re-reads on every request. Duplicated boilerplate, stale rules for removed tooling, and whole style guides inlined instead of referenced are the other repeat offenders.
- Is my file too long?
- There's no hard limit, but the corpus median of 964 tokens / 92 lines is a useful anchor: files several multiples above it pay that overhead on every single request, and 24% of the files we analyzed exceeded 200 lines. Length is only defensible when each rule earns its context cost.
- What are the limitations?
- Heuristics can't tell a dense, high-value rule from filler of the same length, and token counts use the chars/4 estimate — verify real usage against your provider's reported token counts. Imports and referenced files aren't followed.
- When should I use this vs the full audit?
- This checks one file. The full stack audit covers the whole context stack — CLAUDE.md plus MCP servers, skills, and caching — and ranks fixes by estimated token impact.
For your full setup, run the free stack audit — or see the 100-configs report.