Skip to main content
Back to Vibe Coder Toolkit
Intelligence

Context Optimizer

Stop burning tokens on bloat

npx promptreports-cli context

Measures context window consumption: CLAUDE.md, skills, memory, system prompt. Finds skills that load but never get invoked.

Every Claude Code turn loads your CLAUDE.md, all skills, memory files, and system prompt — even if you never use them. context measures that startup tax. Shows biggest consumers, skills that never get invoked (dead weight), and files you read every session (candidates for context injection). Calculates estimated daily savings if you trim the bloat.

Prerequisites

  • CLAUDE.md and skills optional
  • Session files for usage patterns

Flags & Options

FlagDescriptionDefault
--days NLookback for usage patterns7
--jsonJSON output

Examples

See your context cost

npx promptreports-cli context

Default — what are you paying to load every turn?

Export for trim plan

npx promptreports-cli context --json

Feed into automated CLAUDE.md optimizer.

Output

Context budget bar + biggest consumers table + unused skills list + top file reads + estimated savings.

  CONTEXT BUDGET: 200K tokens
  USED AT START: 18,450 tokens (9.2%)
  ████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░

  BIGGEST CONSUMERS:
  CLAUDE.md                       8,200 tokens (44%)
  Skills (7 loaded)               5,400 tokens (29%)
  System prompt                   4,500 tokens (24%)
  Memory files (12)               1,900 tokens (10%)

  SKILLS NEVER INVOKED (last 7 days):
  - old-deploy-script            (~450 tokens)
  - experimental-ai-feature      (~380 tokens)
  - legacy-image-tool            (~290 tokens)
  Removing saves ~1,120 tokens/session

  MOST-READ FILES (candidates for context injection):
  - lib/prisma.ts                read 23 times
  - app/api/auth/route.ts        read 18 times

  Estimated savings: $0.14/day if you apply optimizations

What it reads and writes

Reads

  • CLAUDE.md
  • .claude/skills/*/SKILL.md
  • .claude/memory/
  • ~/.claude/memory/
  • ~/.claude/projects/**/*.jsonl

Writes

Nothing (read-only)

Free vs Pro usage

Free tier

  • Know your startup token cost
  • Identify skills to delete
  • Optimize CLAUDE.md to fit under 2K tokens

Pro tier

Upgrade
  • Auto-apply trim suggestions (PR with before/after)
  • Track context size over time — alert on bloat
  • A/B test CLAUDE.md variants for quality impact
  • Skill auto-pruning — unused for 30d? auto-archive

Pro tips

  • Target: 5K tokens startup or less
  • CLAUDE.md > 2K words is usually bloat — trim
  • If a skill has 0 invokes in 30 days, delete it
Free to try

Install the whole toolkit.

One command unlocks all 23 tools.