Mission Control · Cheat sheet

The Claude optimization cheat sheet

Practical reference for cutting Claude spend without losing capability. Three surfaces — Claude.ai, the API, Claude Code — and the cross-cutting fundamentals that apply to all three. Each section deep-links to the Mission Control module that puts it into practice.

Fundamentals

§1Token economics in one minuteThe four numbers that decide what a Claude call costs.READ →

Cross-cutting

§6.4The seven cardinal sins of token wasteThe catalogue Mission Control's Auditor grades you against.READ →

Chat (claude.ai)

§3.2claude.ai Project Knowledge — the silent budget killerEvery doc you attach to a Project is sent on every message. Curate ruthlessly.READ →
§3.3Project Instructions — compress without losing intentA 1,500-token instruction block can usually do the work of a 5,000-token one.READ →

API

§4.1Prompt cachingHow to cache the prefix so repeat calls cost a fraction.READ →
§4.3Cache hit rate as a north-star metricWhy this number, and how to drive it up.READ →
§4.6Output length, max_tokens, and schema designOutput costs 5× input. Treat every output token like rent.READ →

Claude Code

§5.1/clear vs /compact — pick the right resetTwo commands, two different jobs. Knowing when to use which.READ →
§5.5CLAUDE.md — the file every Claude Code project pays for foreverThis file gets read on every session. Keep it short, focused, and pruned.READ →
§5.6opusplan, subagents, and budget disciplineWhen to reach for Opus, and how to budget a swarm of subagents.READ →