page:docs-reference-repos-clawhub-mco0l-caveman-claw-research
mCo0L/caveman-claw reference
Token cost reduction plugin for OpenClaw with two mechanisms: (1) Output mode -- instructs the AI to respond in terse, fragment-based style (same technical accuracy, fewer words), and (2) Auto-compression -- algorithmically trims older messages in conversation history before each request (strips articles, filler, synonyms shortened).
mCo0L/caveman-claw
- **Archetype**: clawhub-plugin
- **Stars**: 3
- **Last pushed**: 2026-04-12
- **License**: MIT
- **Discovered**: 2026-04-12
- **Language**: TypeScript
- **Fork**: No
Summary
Token cost reduction plugin for OpenClaw with two mechanisms: (1) Output mode -- instructs the AI to respond in terse, fragment-based style (same technical accuracy, fewer words), and (2) Auto-compression -- algorithmically trims older messages in conversation history before each request (strips articles, filler, synonyms shortened).
Activates globally by default on every new session. Tracks token savings via /cc stats. Simple, minimal approach -- no LLM calls, no vector databases, no external services.
Assessment
LOW extractable value. The concept is simple and already partially covered by babysitter's compression layer. The "caveman mode" (terse AI output) is a prompt engineering technique, not a plugin architecture pattern. The auto-compression of older messages (article stripping, filler removal) is a basic NLP technique. Tracked for completeness but low priority.
**Extraction priority**: LOW
Extractable Value: mCo0L/caveman-claw
Processes
None.
Plugin Ideas
1. Terse Mode Plugin
- **Category**: Context & Memory
- **install.md**: Installs a session-level toggle that instructs the AI to respond in terse, fragment-based style for token savings. Also algorithmically compresses older conversation history by stripping articles, filler words, and shortening synonyms. Tracks cumulative token savings.
Library Mapping
| Extractable Process | Library Status | Action | Existing Path | Target Placement |
|---|---|---|---|---|
| N/A | N/A | No multi-step processes identified - terse mode is prompt engineering technique | - | N/A |
Plugin Marketplace Mapping
| Plugin Idea | Marketplace Status | Action | Existing Plugin | Target Placement |
|---|---|---|---|---|
| Terse Mode Toggle | POTENTIAL OVERLAP | Session-level terse response mode with conversation compression | babysitter compression system | plugins/a5c/marketplace/blueprints/terse-mode-toggle/ |
Implicit Procedural Knowledge
- **Dual-layer token reduction**: Combining output mode (prompt engineering for terse responses) with input compression (algorithmic trimming of history) addresses both sides of the token cost equation. Simple but effective for cost-sensitive environments.
- **Global-by-default activation**: The
globalByDefaultsetting pattern -- plugin activates automatically on every new session unless explicitly disabled. Relevant for babysitter plugins that should be always-on.