Claude Code Context and Token Budget Audit
Find what's eating your Claude Code context and usage limits, then get ranked fixes with estimated token savings per turn.
Before & after
Same prompt, same agent — the only difference is the skill.
Token usage can definitely vary depending on how you're using Claude Code. There are a lot of factors that influence token consumption - like how many files you're reading, the size of your codebase, and how complex your requests are. You might want to think about optimizing your workflow. Consider doing things in smaller chunks or breaking down larger tasks. You could also try to be more specific with your prompts to get better results. There are probably ways to improve efficiency, but it really depends on your specific situation.
TOKEN OPTIMIZER ANALYSIS Session context waste: 34% reduction possible VACUUM STEPS (execute in order): 1. Remove old conversation artifacts from session memory (freed: 2847 tokens) 2. Drop unused tool schemas from scope (freed: 1204 tokens) 3. Collapse nested function context into summaries (freed: 892 tokens) 4. Archive files read but not modified (freed: 4103 tokens) BEFORE THIS SESSION: 87,400 tokens (session context: 45%) AFTER OPTIMIZATION: 57,300 tokens (session context: 22%) ONGOING RULES: Don't re-read files already in memory. Grep for symbols instead of full reads. Request Read-only for static assets. Use Bash for commands before manually repeating them. RECOMMENDED SESSION LIFECYCLE: Close and restart every 50k tokens consumed to reset implicit context overhead.
About this skill
Your Claude Code sessions auto-compact every 20 minutes, Claude forgets decisions you made an hour ago, or you hit your plan limit before lunch. This skill finds the specific overhead you pay on every turn (bloated CLAUDE.md, unused MCP servers, huge file reads, noisy test output) and gives you a ranked fix list with estimated savings.
What it does
- Reads your pasted
/contextbreakdown and names the top offenders, bucket by bucket - Separates three problems that need different fixes: context window pressure, API cost, and plan usage limits
- Rewrites a bloated CLAUDE.md into a lean version and tells you where the cut material should live instead
- Audits MCP servers and suggests project scoping or CLI replacements for tools you rarely call
- Adds read guards and quieter test and log commands so single tool calls stop dumping 30k tokens
- Gives in-session triage: when to
/compactwith focus instructions, when to/clear, and how to hand off to a fresh session
What you get
A ranked list of fixes, each with the exact change (config snippet, command, or rewritten file) and an estimated token saving per turn and per session, labeled as an estimate.
Best for
Developers using Claude Code daily on real codebases, especially monorepos or setups with several MCP servers, who keep hitting context limits or plan usage caps.
SKILL.md preview
---
name: claude-code-context-token-audit
description: Audits where a Claude Code session or project spends tokens (CLAUDE.md and memory files, MCP tool definitions, file reads, tool output, conversation history) and returns ranked fixes with estimated savings, prioritizing fixed per-turn overhead over one-off waste. Use when the user says "my Claude Code sessions keep hitting the context limit", "it auto-compacts every 20 minutes", "I burn through my Max limit by lunch", "what's eating my context", "trim my CLAUDE.md", "do I have too many MCP servers", "this session feels heavy", or "make Claude Code cheaper".
---
# Claude Code Context and Token Budget Audit
… (sign up to view the full skill)More development skills
View all Development skills →Performance Bottleneck Profiling: Node, Python, Go
Classifies a slowdown, gives exact profiler commands, reads the output, and proves the fix with same-conditions p95/p99 numbers.
Feature Scaffolding with Working Tests
Turn a short feature spec into code, runnable tests, and integration wiring that match your existing project.
Feature Flag Rollouts and Kill Switches
Design staged flag rollouts, verify rollback paths, and audit flag debt with explicit defaults, gates, and retirement plans.
Microcontroller Firmware and RTOS Engineering
Build and debug MCU firmware with bounded ISR/RTOS timing, safe DMA ownership, and measured power budgets.