Intelligence

CLAUDE.md Slim-Down

Find bloat taxing every Claude Code turn

npx promptreports-cli audit claude-md

Deep token-budget analysis of your root CLAUDE.md. Detects duplicated phrases, oversized examples, dead file references, sections duplicating other docs, and filler words. Shows projected savings per session.

CLAUDE.md gets loaded on every turn and every compaction. A 5,000-token CLAUDE.md costs real money across a month of sessions. This subcommand runs five analyzers: duplicate-phrase detection (phrases appearing 3+ times), verbose-example detection (code blocks over 30 lines), dead-reference detection (cited files that don't exist on disk), redundant-section detection (content that duplicates .claude/LESSONS.md, TECH_STACK.md, etc.), and filler-word detection (please, basically, actually — words Claude reads past). Output is sorted by token impact so you fix the biggest wins first.

Prerequisites

A CLAUDE.md file at the project root

Flags & Options

Flag	Description	Default
--json	Machine-readable findings for dashboards or scripts	—

Examples

Analyze CLAUDE.md

npx promptreports-cli audit claude-md

Full slim-down report with findings by category.

JSON for CI

npx promptreports-cli audit claude-md --json

Pipe into scripts or fail CI if tokens exceed threshold.

Output

Top box shows current vs. projected token count and total savings. Findings are grouped by kind (duplicates, verbose examples, dead refs, redundant sections, filler) and sorted by token impact.

┌─ CLAUDE.md Slim-Down Analysis ──────────────────┐
│  Current:   4,280 tokens  (612 lines, 3,104 words)  │
│  Projected: 2,640 tokens  after applying 11 fixes    │
│  Savings:   1,640 tokens (38%) per session           │
└──────────────────────────────────────────────────────┘

═══ Duplicated content — 3 findings, ~420 tokens ═══
  HIGH   L142  every development decision cascades...
         Phrase appears 5× (first at line 142). Keep one.  (~240 tokens)

═══ Dead file references — 2 findings, ~30 tokens ═══
  HIGH   L87   → .claude/plans/archived-auth-v2.md
         File doesn't exist. Update or remove the reference.

═══ Sections duplicating other docs — 3 findings, ~890 tokens ═══
  MED    L234  ## Lessons Learned
         Duplicates .claude/LESSONS.md. Replace with reference.  (~610 tokens)

What it reads and writes

Reads

CLAUDE.md
.claude/LESSONS.md
.claude/TECH_STACK.md
.claude/DESIGN_SYSTEM.md
.claude/IMPLEMENTATION_PLAN.md
.claude/BACKEND_STRUCTURE.md
filesystem (for dead-ref checks)

Writes

Nothing (read-only)

Free vs Pro usage

Free tier

Slim CLAUDE.md once and save tokens on every future session
Catch dead references after deleting or renaming files
Verify .claude/ docs actually get referenced from the root

Pro tier

Upgrade

CI gate on CLAUDE.md token budget with --json
Track CLAUDE.md size over time on a team dashboard
Auto-apply safe slim-down fixes via platform integrations

Pro tips

Run after any major docs reorganization — that's when dead refs accumulate
If savings estimate is high, do the refactor: it compounds across every session this week
Filler-word findings feel pedantic but add up on a 600-line CLAUDE.md

Free to try

Install the whole toolkit.

One command unlocks all 23 tools.

Start Free Back to all tools