Caveman (token-compressed replies) plugin thumbnail

Caveman (token-compressed replies)

Cuts output tokens by roughly 65% while keeping full technical accuracy by having the agent answer in compressed caveman-speak, across six intensity levels including Classical Chinese modes. Toggleable per chat from the WebUI, with live response validation and cumulative token-savings stats.

Author Olszalsik 0 stars Version 0.4.0 Updated
PromptingLlmTools
GitHub Repository Discussion How to install

README

Caveman plugin for Agent Zero

Ultra-compressed communication mode for your Agent Zero agent. Same answers, ~65% fewer output tokens (measured upstream), full technical accuracy.

Derived from Julius Brussee's caveman plugin and adapted to Agent Zero's plugin conventions. MIT licensed.

What it does

Injects a system-prompt fragment that tells the model to answer in tight caveman-speak: drop articles, filler, pleasantries, hedging. Code, commands, errors, and technical terms stay byte-exact.

Six intensity levels are supported:

Level Same sentence, shrunk
lite Wrap object in useMemo. New ref created every render.
full (default) New ref each render. Wrap object in useMemo.
ultra New ref/render. useMemo it.
wenyan-lite 組件頻重繪，以每繪新生對象參照故。以 useMemo 包之。
wenyan-full 每繪新生對象參照，故重繪；以 useMemo 包之則免。
wenyan-ultra 新參照則重繪。useMemo 包之。

Honest numbers: caveman only shrinks output tokens. The skill itself adds a small amount of input tokens per turn. The real wins are readability and speed. Cost savings are the bonus. See upstream's HONEST-NUMBERS.md for the full breakdown.

Install (local plugin)

The plugin is already placed at /a0/usr/plugins/caveman/. To enable it:

  1. Open Settings → Developer in the WebUI.
  2. Find the Caveman subsection.
  3. Toggle Enabled ON.
  4. Pick a Level (default: full).
  5. Save.

Alternatively, edit default_config.yaml and set:

enabled: true
level: full
auto_clarity: true

Slash commands (per-chat)

Command Effect
/caveman Turn ON at the default level (full).
/caveman lite Switch this chat to lite.
/caveman full Switch this chat to full.
/caveman ultra Switch this chat to ultra.
/caveman wenyan-lite Switch this chat to wenyan-lite.
/caveman wenyan-full Switch this chat to wenyan-full.
/caveman wenyan-ultra Switch this chat to wenyan-ultra.
/caveman off / stop caveman / normal mode Turn OFF for this chat.
/caveman on / talk like caveman / use caveman Turn ON for this chat.

Per-chat state is the killer feature: one chat can be on ultra while another is on lite. State is stored in <workdir>/.caveman/state.json.

Sub-skills (loadable)

Loadable via skills_tool: load or by saying "use caveman-X":

Skill Purpose
caveman-stats Show estimated tokens saved this session.
caveman-commit Terse Conventional Commits. Subject <=50 chars.
caveman-review One-line PR review comments.
caveman-compress Compress .md memory files into caveman-speak.
caveman-help This help card.

Cavecrew subagents (Commit 3)

Profile Role
cavecrew-investigator Read-only code locator. Returns path:line table.
cavecrew-builder Surgical 1-2 file editor. Refuses 3+ file scope.
cavecrew-reviewer PR/diff reviewer. One-line findings.

Use with call_subordinate(profile="cavecrew-investigator", ...).

HTTP API (read/write per-chat state)

POST /api/plugins/caveman/caveman_state

// Read
{ "action": "get", "chat_id": "<id>" }
// Response: { "ok": true, "level": "full", "enabled": true, ... }

// Set level
{ "action": "set_level", "chat_id": "<id>", "level": "ultra" }

// Toggle enabled
{ "action": "set_enabled", "chat_id": "<id>", "enabled": true }

// List all chats (diagnostics)
{ "action": "list" }

What ships in v0.1.0 (Plan A - thin prompt port)

  • plugin.yaml, default_config.yaml, hooks.py, execute.py
  • One system_prompt extension that injects the caveman style into every agent turn when the plugin is enabled.
  • Six intensity fragments + the base style prompt + an auto-clarity fragment.
  • A welcome-screen discovery banner (dismissible, low priority).

What ships in v0.2.0 (Plan B - per-chat state + slash command)

  • Per-chat level state (<workdir>/.caveman/state.json).
  • /caveman <level> slash command detection (monologue_start extension).
  • HTTP API for read/write per-chat state.
  • Five sub-skills (caveman-stats, caveman-commit, caveman-review, caveman-compress, caveman-help) loadable via skills_tool.

What ships in v0.3.0 (Plan B complete)

  • Three cavecrew subagent profiles (investigator, builder, reviewer) usable via call_subordinate(profile="cavecrew-investigator", ...).
  • Soft response validator (response_stream_end extension) that flags banned filler phrases in the model's output as a soft warning in chat_extras (never a hard block).

What's planned for v0.4.0 (Plan C - deep port)

  • caveman-shrink equivalent (compress tool descriptions via the tool_data extension point).
  • Live token-savings API + dashboard.
  • Response sanitizer (response_before post-processor).
  • install.py for community distribution.
  • Reproducible benchmarks against the Agent Zero port.

License & attribution

MIT. Original caveman plugin by Julius Brussee (https://github.com/juliusbrussee/caveman). Agent Zero port by Agent Zero contributors. See LICENSE for the full text.

MIT License

Copyright (c) 2025 Julius Brussee
Copyright (c) 2026 Agent Zero contributors (port to Agent Zero)