What Is the caveman-commit Skill and How It Relates to Other Caveman Skills

The caveman-commit skill is a specialized sub-agent that generates ultra-compressed Conventional Commit messages, serving as the version-control counterpart to the general-purpose caveman compression engine.

The JuliusBrussee/caveman repository provides a suite of token-saving utilities for AI-assisted development workflows. Among these tools, the caveman-commit skill stands out as a dedicated agent for creating succinct git commit messages that preserve essential debugging context while strictly minimizing verbosity.

Core Purpose of the caveman-commit Skill

The caveman-commit skill operates as a focused LLM agent defined in skills/caveman-commit/SKILL.md and documented in skills/caveman-commit/README.md. Its primary responsibility is generating ultra-compressed commit messages adhering to the Conventional Commits specification.

Key constraints govern its output:

• Subject lines must not exceed 50 characters
• No AI attribution or filler text is permitted
• Essential "why" information must be preserved for future debugging

Unlike generalized text compression, this skill understands git semantics. It parses staged changes and produces messages like feat(api): add GET /users/:id/profile with optional body text explaining the rationale, ensuring version-control history remains machine-readable and human-scannable.

Relationship to Other Caveman Skills

The caveman ecosystem comprises multiple specialized skills that share a common goal of reducing token consumption. Understanding how caveman-commit differs from its siblings clarifies when to invoke each tool.

The General caveman Skill

The root caveman skill, defined in skills/caveman/SKILL.md, functions as a global output compression engine. It intercepts every LLM reply and rewrites it according to modes like lite, full, ultra, or wenyan. While this general-purpose tool reduces conversational verbosity, it lacks the structural awareness required for commit-message conventions.

caveman-review

caveman-review produces ultra-terse code review feedback formatted as single-line comments (e.g., L42: 🔴 bug: null pointer risk). Where caveman-commit summarizes changes, caveman-review critiques them. Both share the same compression philosophy but target different phases of the development lifecycle.

caveman-compress

caveman-compress handles static file optimization, rewriting long-term memory files such as CLAUDE.md into caveman-style versions. This permanently shrinks context loads for future sessions, whereas caveman-commit generates ephemeral messages for immediate git operations.

caveman-stats

caveman-stats provides runtime telemetry across all Caveman skills. It tracks token savings when you invoke /caveman-commit and can output share-ready summaries showing efficiency gains. This skill operates on the session log, measuring the delta between verbose and compressed outputs.

How to Use the caveman-commit Skill

Invoke the skill directly when you need a commit message, or let it trigger via natural language detection.

Direct Invocation

Run the command in your terminal or agent interface:

/caveman-commit

Example output:

feat(api): add GET /users/:id/profile

Mobile client needs profile data without the full user payload
to reduce LTE bandwidth on cold-launch screens.

Closes #128

The output follows the formal specification in skills/caveman-commit/SKILL.md, ensuring conformance to Conventional Commits.

Natural Language Trigger

The skill activates when the AI detects commit-related intent:

User: "Can you write a commit message for the new login endpoint?"
Assistant: /caveman-commit

Combining with Other Skills

Chain commands to generate commits and measure efficiency:

/caveman-commit
/caveman-stats

The stats command will report token reduction achieved by the compressed message compared to a verbose alternative.

Summary

• The caveman-commit skill generates Conventional Commit messages with subject lines strictly limited to 50 characters.
• It operates independently from the general caveman skill, focusing exclusively on version-control history rather than general conversation.
• Unlike broad compression algorithms, it preserves debugging context while eliminating filler text and AI attribution.
• It integrates with caveman-stats for token measurement and works alongside caveman-review and caveman-compress as part of a comprehensive token-reduction toolkit.

Frequently Asked Questions

Does caveman-commit automatically stage and commit files?

No. The skill only generates the commit message text. It does not execute git add or git commit commands. You must manually stage your changes and complete the commit using your preferred git workflow.

What is the maximum length for commit messages generated by caveman-commit?

Subject lines must be 50 characters or fewer. The skill may include an optional body paragraph for additional context, but the headline remains strictly constrained to fit standard git conventions and terminal displays.

How does caveman-commit differ from the general caveman command?

The general caveman skill, implemented in skills/caveman/SKILL.md, compresses any AI-generated reply using modes like lite or ultra. In contrast, caveman-commit specifically targets git commits with strict Conventional Commits formatting and semantic awareness of code changes.

Can I use caveman-commit with other Caveman skills in the same session?

Yes. You can chain /caveman-commit with /caveman-stats to measure token savings, or use it alongside /caveman-review when preparing a pull request that requires both a commit message and review comments.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}