Most developers burn tokens on patterns they don't realize are wasteful: asking the same clarifying questions every session because their CLAUDE.md is empty, sequential file reads when parallel would cut turns in half, no verification step so Claude claims success on broken code, retry loops on the same failed approach instead of pivoting.

I review your actual usage and deliver a concrete report.

1. Session review — You share 3-5 recent Claude Code sessions (transcripts or screenshots). I analyze where tokens went, what worked, and what burned context for nothing.

2. Pattern diagnosis — Your top 3-5 token-wasting habits. Not generic advice. Specific patterns from your actual sessions: "You asked Claude to read 14 files sequentially across turns 3-18. Parallel reads would have cut that to 2 turns."

3. Config recommendations — What to add to your CLAUDE.md based on mistakes Claude keeps making in your sessions. Specific rules like "Always check for existing utils before creating new ones" because I caught Claude duplicating a helper three times.

4. Skill recommendations — Which skills would eliminate your repeated pain points. If nothing exists, I recommend what a custom skill should do.

5. Before/after estimate — Rough projection of tokens saved per month and quality improvement from implementing the changes.

Deliverable: A written audit report (2-4 pages) with specific, actionable fixes ranked by impact.

Timeline: 2-3 business days from receiving your session transcripts.

- You use Claude Code regularly but suspect you're not getting full value

- Your sessions feel like they take more turns than they should

- Claude keeps making the same mistakes in your project

- You want a quick, low-commitment way to level up