# JuliusBrussee/caveman

> A skill and local proxy for coding agents that cut token usage: the skill makes the agent answer in terse caveman style, the proxy compresses logs, diffs and test output before they reach the model.

- Magnitude: 3.3 out of 10 — Steady
- Stars: 116,945 total · +3 stars measured 2026-11-03, 00:32–01:17 UTC, ≈ 166 by evening
- Star trust: star growth looks organic
- Category: AI agents · Language: Go · License: Apache-2.0 · Created: 2026-04-04 · Last push: 2026-11-02
- GitHub: https://github.com/JuliusBrussee/caveman · Homepage: https://docs.caveman.so/caveman · Page: https://gitnova.dev/en/r/JuliusBrussee/caveman

## Useful for

- Install the skill via npx skills add so the agent answers shorter and cheaper
- Run caveman learn to rank the biggest token sinks in your agent history
- Run caveman claude or caveman codex to compress logs and diffs before sending them to the model

## Why it’s here

- Star-counter measurements on 2026-11-03 (UTC), 00:32–01:17: 116942 → 116945 stars (+3). This is the change over that interval.
- Estimated end-of-day forecast: about +166 stars, using observed gains and the previous day.
- About 25 forks a day — people are taking the code.

## Star trust

Star growth looks organic.

Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer.

## Numbers

- Forks: 6,795
- Issues and pull requests: 1,373
- Watchers: 255
- Average over the last week: 231 per day
- Usual pace: 271 per day
- Stars in the last hour (measured): 8
- Latest release: bin-v4.0.2 (2026-10-23)

## Stars per day, last 30 days (oldest → newest, today is partial)

2026-10-05 … 2026-11-03: 201, 226, 213, 198, 204, 155, 174, 213, 300, 436, 833, 327, 241, 218, 229, 217, 300, 282, 227, 206, 266, 288, 238, 299, 312, 221, 206, 231, 182, 3

## Magnitude by day, last 14 days

10-21 3.8, 10-22 3.8, 10-23 3.6, 10-24 3.4, 10-25 3.5, 10-26 3.6, 10-27 3.7, 10-28 3.8, 10-29 4.2, 10-30 3.9, 10-31 3.6, 11-01 3.5, 11-02 3.6, 11-03 3.3

## More in this category

1. **storytold/craft-anything** — 9.2 · Breakout · AI agents · Rust · +145 stars measured 2026-11-03, 00:30–01:15 UTC, ≈ 2,481 by evening
   A pure-Rust reverse-engineering engine with 284 tools that inspects websites, apps and games, then has a coding agent rebuild a similar version and checks the result against the original.
   Full card: https://gitnova.dev/en/r/storytold/craft-anything.md
2. **vercel-labs/agent-desktop** — 9.0 · Breakout · AI agents · Rust · +106 stars measured 2026-11-03, 00:30–01:15 UTC, ≈ 2,236 by evening
   Rust CLI that lets AI agents control desktop apps via macOS accessibility trees, with stable element refs and a snapshot-to-action loop. Combines a native binary with a C-ABI cdylib for Python/Swift/Go/Node bindings.
   Full card: https://gitnova.dev/en/r/vercel-labs/agent-desktop.md
3. **pingdotgg/t3code** — 9.0 · Breakout · AI agents · TypeScript · +80 stars measured 2026-11-03, 00:30–01:15 UTC, ≈ 2,857 by evening
   T3 Code is an agent harness control surface for agents on your machine, with mobile (iOS, Android), web and Electron desktop apps. It works with your Claude Code, Codex, Cursor, Grok Build, OpenCode and Google Antigravity subscriptions if…
   Full card: https://gitnova.dev/en/r/pingdotgg/t3code.md
4. **deepseek-ai/dsh-superagent-prototype** — 8.4 · Peaking · AI agents · TypeScript · +118 stars measured 2026-11-03, 00:30–01:15 UTC, ≈ 3,017 by evening
   Research prototype of a proactive coding agent for a single GitHub repository, packaged as a bundle for DeepSeek Harness: it listens to GitHub and Feishu, opens work items and replies in the original conversation, for example with a pull…
   Full card: https://gitnova.dev/en/r/deepseek-ai/dsh-superagent-prototype.md
5. **Donchitos/Claude-Code-Game-Studios** — 7.1 · Breakout · AI agents · Shell · +26 stars measured 2026-11-03, 00:30–01:15 UTC, ≈ 774 by evening
   A Claude Code configuration that turns a single session into a structured game studio: 49 specialized subagents, 74 slash-command skills, hooks, rules and templates mirroring a real studio hierarchy. It adds design reviews, QA passes and…
   Full card: https://gitnova.dev/en/r/Donchitos/Claude-Code-Game-Studios.md

---

Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score.
Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast.
Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself.
Data as of 2026-11-03 01:27 UTC, updated every 30 minutes.