Caveman
Agents like to explain: a friendly opening, a recap, “I’d recommend…”. This plugin gives a connected agent caveman’s reply rules. The filler goes; code, commands, file paths and error messages stay exactly as they are.
Built on caveman, open source by Julius Brussee; its rules are MIT-licensed:github.com/JuliusBrussee/caveman
−29%
fewer words in the answer, in our one test with Claude Code
≈1,440
tokens the rules add once per session (then ≈62 per turn)
0
restarts to switch it on or off
Same diagnosis, smaller mouth
caveman’s own example: one question, answered without caveman and at each level the card offers.
“Why does my React component re-render?”
The reason your React component is re-rendering is likely because you're creating a new object reference on each render cycle. When you pass an inline object as a prop, React's shallow comparison sees it as a different object every time, which triggers a re-render. I'd recommend using useMemo to memoize the object.316 characters
“Why does my React component re-render?”
New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo.Drops articles, fragments are fine, short synonyms. The author’s default.
88 characters
Sample answers quoted from caveman’s README and rules, in the author’s English.
Fewer words, not fewer facts
caveman is a set of reply-style rules. The plugin hands them to the agent with each turn — nothing is installed into your CLI.
Terse answers
No pleasantries, no hedging, no recap of what it just did. Short sentences, one idea each.
Exact where it matters
Code, commands, file paths and error messages are never shortened. Security warnings and irreversible steps come back in full sentences.
Normal where others read it
Code, commit messages, docs and pull-request text stay in normal prose. The agent keeps answering in your language.
One real session, arrow on and off
Claude Code with Haiku 4.5 on Windows, level Full, three similar questions in one session, no restart.
- 1No arrow225Characters: 1,506Output tokens: 413
- 2Arrow drawn159Characters: 1,164Output tokens: 408
- 3Arrow removed254Characters: 1,706Output tokens: 419
What the rules cost: ≈1,440 input tokens once per session, then ≈62 per turn (estimated at 4 characters a token).
A demonstration, not a benchmark: different questions, one run each, on 2026-09-29. Output tokens are the provider’s count and include the model’s thinking, which caveman doesn’t shorten — so here the visible answer got shorter while billed output barely moved.
A card, a level, an arrow
Nothing to download, nothing to configure.
Add the card
Add-card menu → “Plugin…” → Plugins tab → Caveman → “Add to canvas”.
Pick a level
Lite, Full or Ultra — or classical Chinese. Change it any time; the next turn uses the new one.
Draw an arrow
From an agent to the card. Its next answer already follows the rules. No restart.
Remove it to stop
Take the arrow away and the next turn carries a short note to answer normally again.
Works with three CLIs today
The ones whose prompt hook runs before every turn and can add text for the model.
Other agents can be connected, but the card marks them “Not supported”: their CLIs don’t offer a way to add text to a turn from outside yet.
Credit
caveman is made by Julius Brussee
The rules are caveman’s own — its caveman skill text and its per-turn reminder — copied into NeuroSquad unchanged from a pinned version under the MIT license. caveman’s separate proxy and engine are not part of this plugin.
Questions about Caveman
Shorter answers from the next turn
Download NeuroSquad, add Caveman, and draw an arrow.