caveman-compress

PurposeThe caveman-compress skill compresses natural language memory files (CLAUDE.md, todos, preferences) into caveman format, saving ~46% input tokens on every subsequent session. It uses a Python script (`scripts/__main__.py`) for automated compression, preserving all technical content, code, URLs, and structure. The compressed version overwrites the original, with a human-readable backup saved as `FILE.original.md`. Key output: compress once, save tokens every session forever.

Domains

Coding

Forms

Workflow

Required Tools

python3

Languages

python

Package Manager

—N/A

Skill Composition

SKILL.md10.0%
scripts70.0%
other20.0%

Mechanism

When It Triggers · How It Works · What It Produces

Breaks down how the skill works and what it produces, so you can quickly judge whether it fits your scenario and value.

Trigger

User types /caveman-compress <filepath>.
User says "compress memory file" or similar natural language requests.

Mechanism

Agent locates scripts/__main__.py in the skill directory and runs python3 -m scripts <absolute_filepath>.
The Python CLI pipeline: detect file type → call Claude API for compression → validate output → retry up to 2 times → overwrite original.
Compression rules: remove articles/filler/pleasantries/hedging, use short synonyms; preserve code blocks, inline code, URLs, file paths, technical terms, table structures, YAML frontmatter entirely unchanged.
Original file backed up as <filename>.original.md; backup files are never compressed.
Only processes natural language files (.md/.txt/.typ/.typst/.tex and extensionless prose); never modifies .py/.js/.ts/.json etc.

Result

Original file overwritten with caveman-compressed version, preserving all structure and technical content.
Original backed up as FILE.original.md for human reading.
~46% average input token reduction, benefiting every subsequent session.

Use Cases

Two Typical Use Cases

Scenario 01

Compressing Claude Code project memory files

A Claude Code developer has a large `CLAUDE.md` project memory file that consumes significant input tokens on every session load. Running `/caveman-compress CLAUDE.md` compresses it into caveman format, removing filler and redundant narration while preserving all code examples, commands, and paths. A backup is saved as `CLAUDE.md.original.md`. Every subsequent session loads this file with ~46% fewer input tokens.

Scenario 02

Compressing todo lists and preference files

A developer maintains todo lists, coding preference configs, and other Markdown memory files. The natural language descriptions in these files consume many tokens with low information density. Using `/caveman-compress` batch-compresses these files — the agent identifies and compresses only prose sections, preserving all code blocks, YAML frontmatter, table structures, and other technical content unchanged, ensuring structural integrity and readability after compression.

Skill File

SKILL.md

Open in GitHub ↗

SKILL.mdLoading from GitHub…

Skill Composition

Open in GitHub ↗

SKILL.md10.0%
scripts70.0%
other20.0%

▸skills/caveman-compress/
▸caveman-compress
- ·README.md
- ·SECURITY.md
- ·SKILL.md
- ▸scripts
  - ·__init__.py
  - ·__main__.py
  - ·benchmark.py
  - ·cli.py
  - ·compress.py
  - ·detect.py
  - ·validate.py

Per the Agent Skills spec, SKILL.md is the only required file. The colored bars classify the rest into the three spec-aligned optional directories (scripts / references / assets) and an 'other' bucket. The full directory tree is below.