soleri-compresslisted

# Compress Compress natural language files (CLAUDE.md, instructions, vault notes) into terse format. Reduces input tokens ~40-50% while preserving all technical substance. Compressed version overwrites original. Human-readable backup saved as `<filename>.original.md`. ## When to Use - CLAUDE.md files loaded every session (multiplied savings) - Instruction files that are mostly prose - Vault entry descriptions that are verbose - Any markdown file where prose can be tightened ## When NOT to Use - Code files (.ts, .js, .py, .json, .yaml, etc.) - Config files (.env, .toml, .ini) - Files that are already terse - Files with `.original.md` suffix (backups) ## Process ### Step 1: Read and Classify Read the target file. Classify it: | Type | Action | | ---------------------------- | ----------------------------------------------------------------- | | Natural language (.md, .txt) | Proceed to compress | | Code / config | Stop. Tell user: "Skipping — file is code/config, not prose." | | Mixed (prose + code blocks) | Compress ONLY prose sections. Code blocks pass through unchanged. | ### Step 2: Backup Before any modification, copy the original to `<filename>.original.md`. **HARD-GATE: If `.original.md` backup already exists, STOP.** Tell user: "Backup already exists at {path}. Remove or rename it before re-compressing." This pr