Open sourceGo · ≈116k stars · Apache-2.0
Caveman
A token-saving skill plus proxy that makes coding agents talk tersely.
Caveman makes coding agents answer tersely and read less, which it says cuts about 65% of tokens. One npx install sets up the skill and a local proxy, so plain claude, Codex, Gemini CLI, OpenCode and 30+ other agents stay caveman'd. Each of its six parts has its own switch, and the skill can be installed without the proxy.
Pros & cons
Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.
Pros
- DocsOne npx command installs it for Claude Code, Codex, Gemini CLI, OpenCode and 30+ other agents, using your existing keys or Pro/Max login.[1]
- DocsNever rewrites your prompt, and leaves code and tool calls alone; it compresses only the agent's prose.[1]
- User reportsA commenter notes it does more than compress: intensity modes, consistent output structure, hook-based persistence, and a safety escape on destructive operations.[1]
- MetadataApache-2.0 and actively released (v4.0.1 on 2026-10-13).[1]
Cons
- Independent testAn independent benchmark found that simply adding 'be brief.' held its own against caveman on compression.[1][2]
- DocsIts own numbers show /caveman cuts only 3% more than 'be concise' at the median (measured on older 3.1.0 rules), and the rules cost about 1,160 input tokens.[1]
- DocsCLI telemetry is on by default and sends content-free usage events stored with your IP address; turn it off with 'caveman telemetry off'.[1]
- User reportsNeeds Node 22.13+, and users reported an OpenCode install failing with ENOENT and having to reset the mode every session (both since closed).[1][2][3]
- Best for
- Heavy agent users who want terser replies across many tools and like the extra modes and safety rails.
- Watch out for
- A plain 'be brief' instruction captures most of the compression, and you have to opt out of CLI telemetry yourself.
Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction
Neighbours on the workbench
TypeScript · ≈393k stars · MIT
openclaw/openclaw
The viral open-source personal AI assistant that really does things.
Maintained by OpenClaw Foundation
MDX · ≈307k stars · MIT
obra/superpowers
An agentic skills framework and software methodology that works.
Maintained by Jesse Vincent (obra)
Shell · ≈306k stars · MIT
mattpocock/skills
Small, composable agent skills for real engineering work.
Maintained by Matt Pocock (AI Hero)
Connections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
Maintained by · 1
- JuliusBrusseeOrganisation
Built on · 2
- CodexProject
- Gemini CLIProject
Sources
Verified Oct 29, 2026