Antigravity Skills - Build and Discover AI Agent SkillsAntigravity Skills

Caveman Skill · High-signal communication

Caveman Skill: Say Less. Keep the Signal.

Compress the prose. Keep every fact that matters.

The optimized Caveman Skill co-locates intensity, protected-literal, language, safety, and completion rules while retaining all six modes and the wenyan alias.

Source diff verifiedImmutable package verified
Terse repliesHigh-signal communicationCoding agents
View canonical source
0
0

About This Skill

Caveman is a communication-mode Skill for terse, high-signal replies that compress prose without changing code, commands, paths, API names, numbers, versions, or quoted errors.

Source and Release

Provenance

Upstream author
Julius Brussee
Revision
0d95a81d35a9f2d123a5e9430d1cfc43d55f1bb0
Upstream version
v1.9.1
License
MIT

Antigravity release

Version
0.300.1-c641e0a63f43
Package hash
sha256:3ac3331c9bfaa5d159e6d01b442734a856ec2b33387301a2cb87e9a7d4a50688
Package status
approved
Files
2
ZIP size
3.5 KB

Changelog

  • Deterministic ZIP built from the optimized package with a versioned SHA-256 artifact.

What We Improved

One lifecycle condition for persistence

The candidate states that the selected mode persists until the user stops it, returns to normal mode, or the session ends.

Protected literals and clarity rules stay together

Code blocks, commands, paths, API names, functions, technical terms, numbers, versions, and quoted errors remain exact while prose scaffolding is compressed.

Six modes and the wenyan alias remain explicit

Lite, full, ultra, wenyan-lite, wenyan-full, and wenyan-ultra stay discoverable, with /caveman wenyan mapped to wenyan-full.

A four-check delivery gate

The final answer checks requested facts and actions, order and scope, protected literals, and the selected intensity before sending.

Quick Start

  1. 1Invoke /caveman or state the desired lite, full, ultra, or wenyan mode.
  2. 2Keep code, commands, paths, API names, numbers, versions, and quoted errors in their exact form.
  3. 3Compress prose scaffolding while expanding risky or ambiguous spans when clarity would otherwise suffer.
  4. 4Run the delivery gate before sending and stop the mode with “stop caveman” or “normal mode”.
Example Command
/caveman full

Core Capabilities

Mode persistence

The selected intensity remains active through the task or until the documented stop phrase.

Literal protection

Technical content stays exact while prose wrappers become shorter and more direct.

Dominant-language replies

The candidate preserves the user's language and keeps embedded technical literals in their original form.

Auto-clarity boundary

Security warnings, irreversible actions, ordered sequences, and ambiguity expand when compression would obscure meaning.

Use Cases

Coding-agent handoffs

Return concise implementation results without dropping paths, symbols, or next steps.

Debugging summaries

State the root cause, exact fix, and validation in a compact technical form.

Persistent response mode

Keep a chosen intensity across a session while retaining an explicit escape hatch.

Source diff and package proof verified

Evidence: source diff

Verified source-diff evidence

This release proves canonical source lineage, package inventory, the optimized true diff, and an immutable artifact. It does not claim runtime outcomes or visual comparisons.

  • .scratch/kan-300/production-release/source-evidence/candidates/caveman/original-inventory.json#/entries?sha256=76aa634fdf0a27e285702933c585f1a36a5398f7500409bb354c9c7ec4cd7d60
  • .scratch/kan-300/production-release/source-evidence/candidates/caveman/optimized-inventory.json#/entries?sha256=a15e6d7de00509357d0c9f42112038ec6c1f1714c0ae61b367fe76dab9745254
  • .scratch/kan-300/production-release/source-evidence/candidates/caveman/change-set.json#/evidenceUnits/0?sha256=10c62a83a2d3287c5d94c1e9e89bdb77350c09d074225e3dc5206526f850ac53
  • .scratch/kan-300/production-release/source-evidence/candidates/caveman/change-set.json#/evidenceUnits/1?sha256=10c62a83a2d3287c5d94c1e9e89bdb77350c09d074225e3dc5206526f850ac53
  • .scratch/kan-300/production-release/source-evidence/candidates/caveman/change-set.json#/evidenceUnits/2?sha256=10c62a83a2d3287c5d94c1e9e89bdb77350c09d074225e3dc5206526f850ac53
  • .scratch/kan-300/production-release/source-evidence/candidates/caveman/change-set.json#/evidenceUnits/3?sha256=10c62a83a2d3287c5d94c1e9e89bdb77350c09d074225e3dc5206526f850ac53
Verification receipt
.scratch/kan-300/production-release/source-evidence/candidates/caveman/change-set.json?sha256=10c62a83a2d3287c5d94c1e9e89bdb77350c09d074225e3dc5206526f850ac53
Source-diff receipt
.scratch/kan-300/production-release/source-evidence/candidates/caveman/change-set.json?sha256=10c62a83a2d3287c5d94c1e9e89bdb77350c09d074225e3dc5206526f850ac53

SKILL.md

# Caveman

This edition derives from Julius Brussee's Caveman Skill v1.9.1 at 0d95a81d35a9f2d123a5e9430d1cfc43d55f1bb0.

## Communication contract

Lead with the result. Compress prose scaffolding, but preserve code, commands, paths, API names, technical terms, numbers, versions, and quoted errors exactly. Keep the selected mode until stop or session end, and expand risky or ambiguous spans when clarity needs it.

The natural individual package remains README.md plus SKILL.md; the complete MIT notice must travel with any approved release.

Related Skills

Ponytail SkillMinimalist coding discipline for AI agents.Stop Slop SkillRed-pen editing for clearer AI prose.

Frequently Asked Questions

FAQ

What is Caveman Skill?
Caveman is Julius Brussee's communication mode for terse, high-signal replies that preserve technical substance while removing prose scaffolding.
Which modes are available?
Lite, full, ultra, wenyan-lite, wenyan-full, and wenyan-ultra remain available; /caveman wenyan means wenyan-full.
What does Caveman never compress?
The candidate protects code blocks, commands, paths, API and function names, technical terms, numbers, versions, and quoted error text.
Does it always use fragments?
No. Security warnings, irreversible actions, ordered sequences, ambiguity, and any span where compression obscures causality expand into normal prose.
Can I download it now?
Yes. Download the approved immutable Caveman Skill: Terse, High-Signal Replies for AI Agents package from this page. Its source revision, file count, byte size, SHA-256, and release receipt are shown here; this evidence does not claim runtime outcomes.

Discussion

Discussion

0 comments
U

Quick Info

Author
Julius Brussee
Published
Aug 1, 2026
Category
Communication
Version
0.300.1-c641e0a63f43

Tags

Terse repliesHigh-signal communicationCoding agents

On This Page

Trigger Phrases

Use these phrases to activate this skill in your AI coding assistant:

/cavemantalk like cavemanuse fewer output tokenscaveman litecaveman ultra