cogpros Audit Time
Token optimization audit of bootstrap files. Everything that loads into context every session has a cost. This skill measures that cost and finds ways to reduce it.
When to Use
- When sessions feel slow or context is being consumed too fast
- After adding new skills, hooks, or expanding MEMORY.md
- Periodic maintenance (monthly recommended)
- When the operator says "audit time"
The Protocol
Measure these four sources of per-session context weight:
- MEMORY.md — Size in bytes. Check for entries that should be in topic files instead. Known issue: double-load (appears twice in context).
- CLAUDE.md — Size in bytes. Check for content that could move to skills (triggers, protocols).
- Skills registry — Count entries, estimate ~200 chars per entry. Flag skills that could be suppressed (disable-model-invocation: true).
- Hooks + plugins — Check SessionStart hook outputs. Flag verbose outputs that could be slimmed.
Optimization Tiers (from Jeremy's claude-context-audit)
- Archive — Completely unused skills. Reversible. Move to _archived/.
- Suppress — Rarely-needed skills. Set disable-model-invocation: true (~200 char savings per skill).
- Slim — Verbose hook outputs. Reduce output size while keeping function.
- Restructure — MEMORY.md as lean index with on-demand topic files.
Output
Report total fixed context weight per session. Compare to previous audit if available. Recommend specific changes with estimated savings.
Portability
Works in any Claude Code environment. The measurement targets (CLAUDE.md, MEMORY.md, skills, hooks) are standard Claude Code infrastructure. The optimization tiers apply universally.
Origin
Adapted from jeremyknows/claude-context-audit. Extended with measurement methodology.