caveman-compresslisted

# Caveman Compress Compress memory and context files to reduce Claude Code input token usage by ~46%. ## When to Use - `CLAUDE.md` exceeds 4000 estimated tokens (chars/4) - Context audit score drops below 70 - Sessions end early due to context limits - Before distributing vault content (lovecraft `memory distribute`) ## How It Works Caveman Compress rewrites verbose prose into dense, information-preserving shorthand: | Before | After | |--------|-------| | "When the user asks you to perform X, you should always..." | "X: always..." | | "Important: never do Y under any circumstances" | "NEVER Y" | | "The following commands are available for use" | "Commands:" | | Redundant section headers | Merged or removed | | Explanatory preambles | Stripped | Rules preserved: all semantic content, all instructions, all code blocks, all lists. ## Usage ``` /caveman-compress CLAUDE.md /caveman-compress docs/spec.md /caveman-compress agents/_shared/base-rules.md ``` ## Process 1. Read the target file 2. Identify verbose patterns: preambles, redundant headers, explanatory bridges 3. Compress: terse imperative style, merge related bullets, remove filler 4. Write compressed version in-place 5. Report: original size → compressed size → % reduction ## Output Format ``` Compressed: CLAUDE.md Before : 8317 chars (~2079 tokens) After : 4500 chars (~1125 tokens) Saved : 3817 chars (~954 tokens, -46%) ``` ## Integration with ulk Context Audit (55) will suggest this skill when `CL