caveman-compress

name	caveman-compress
description	Compress natural language files into a shorter caveman form to save prompt tokens while preserving code, URLs, commands, paths, headings, and structure.

Caveman Compress

When To Use

• User asks to compress a file or long markdown instruction to cavemean speak.
• Goal is to reduce input-token cost without losing technical meaning.

Scope

• Good targets: .md, .txt, and extensionless prose-heavy files.
• Mixed prose plus code is okay, but only compress the prose.
• Do not use on source code, structured config, lockfiles, env files, or data files.

Rules

Remove

• articles
• filler
• pleasantries
• hedging
• redundant connective prose
• repeated examples that teach the same thing

Preserve exactly

• fenced code blocks
• inline code
• URLs and markdown links
• file paths
• commands
• API names and library names
• proper nouns
• dates, versions, and numbers
• environment variables
• markdown headings
• bullet and numbered list structure
• tables and frontmatter structure

Workflow

1. Confirm the file is a prose-friendly target.
2. Read the file.
3. Save a backup as <filename>.original.md before overwriting.
4. Compress natural-language sections only.
5. Leave uncertain code-like regions unchanged.
6. Write the compressed file back.
7. Re-read or diff if the file is important.

Preferred Tools

• as_file_read
• as_file_write
• as_file_copy_move
• as_file_patch

Boundaries

• Never compress source code files.
• Never alter content inside backticks or code fences.
• If unsure whether text is code or prose, leave it alone.