Agent toolkit

Cavekit

One file, three commands, zero sub-agents

A Claude Code plugin for spec-driven development. It keeps a durable SPEC.md at your repo root that survives context resets, encodes it in caveman grammar (~75% fewer tokens than prose), and automatically backpropagates test failures into spec invariants — no sub-agents, dashboards or orchestration overhead.

npx skills add JuliusBrussee/cavekitGitHub950

The spec-driven loop

SPEC.mddurable · survives context reset

# invariants

✓ email is validated before any DB write

✓ tokens are JWTs; org id from claim only

/ck:spec2 invariants tracked

What it can do

Durable SPEC.md that persists across context resets
Caveman encoding — ~75% fewer tokens than prose
Auto-backpropagation of test failures into spec invariants
Three slash commands: /ck:spec, /ck:build, /ck:check
No sub-agents, no dashboards, no orchestration overhead

language: JavaScript
license: MIT

More from the cave

CavemanOutput compression for Claude Code & 30+ agents.CaveGemmaCaveman compression baked into Gemma's weights.Caveman CodeTerminal coding agent. Half the tokens.CavememPersistent memory your agents recall over MCP.Caveman ProxysoonThe byte-safe LLM gateway.