caveman vs impeccable
Two of the top projects in Skills, plugins and rules, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
caveman
Cuts token use in coding agents with terse replies and compressed inputs
impeccable
Design skill with 24 commands and a detector for AI-generated frontends
Verdict
caveman ranks higher, 94 to 93. Of the 4 score parts we measure for both, caveman leads on 2, impeccable on 1 and 1 is even.
- Adoptioncaveman leads by 6 points, 93 to 87.
- FreshnessEven at 100.
- Maintenancecaveman leads by 5 points, 98 to 93.
- Agent-readyimpeccable leads by 15 points, 85 to 70: a wide gap.
Pick caveman if you are
Developers paying for Claude Code, Codex or similar agents who want lower spend
- One npx command sets up 30+ agents; runs locally with your own keys
- README lists losing cases: HTML input 9.9% worse, grep-defs loses to bare grep
Pick impeccable if you are
Developers using AI coding agents who want better frontend design output
- Supports many harnesses: Claude Code, Cursor, Codex, Gemini CLI, Copilot, Grok Build, Hermes
- 59 deterministic detector rules run without an LLM or API key
| What we compare | caveman | impeccable |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 93, widely used | 87, widely used |
| Freshness | 100, active | 100, active |
| Maintenance | 98, healthy | 93, healthy |
| Easy to run | Not scored for this kind | Not scored for this kind |
| Agent-ready | 70, partly | 85, ready |
| Facts from GitHub and the README | ||
| Stars | 111.9k | 81.2k |
| License | Apache-2.0 (permissive) | Apache-2.0 (permissive) |
| Last commit | Oct 2026 | Oct 2026 |
| Last release | Oct 2026 | Oct 2026 |
| Language | Go | JavaScript |
| Docker | Yes | No |
| GPU | Not needed | Not needed |
| arm64 or Apple Silicon | Not stated | Not stated |
caveman
Caveman is a CLI and skill set that reduces token use in coding agents such as Claude Code, Codex, Gemini CLI and OpenCode. It shortens agent replies, compresses logs, JSON, CSV and test output before the model reads them, and can route each request to a model and effort level through a free account. Reported results include 33.2% fewer input tokens over 54 Claude Code runs and an interim 54% cost cut for Auto on 85 tasks.
Who it is for: Developers paying for Claude Code, Codex or similar agents who want lower spend
Strengths
- One npx command sets up 30+ agents; runs locally with your own keys
- README lists losing cases: HTML input 9.9% worse, grep-defs loses to bare grep
- Each part can be switched off separately; the skill installs alone
- Cited by Adobe Research paper, JetBrains A/B on 86 tasks, Elastic blog
Weaknesses
- Auto routing needs a Caveman account; its benchmark is interim and half complete
- Terse output saves nothing on per-request billing such as GitHub Copilot
- Rules cost about 1,160 input tokens; output benchmarks used older 3.1.0 rules
- No HTML compressor yet; Blocks is pre-release (v0.1.0-rc.5), Python only
- no GPU
- Docker + Compose
- Needs Node.js 22.13+, Caveman account (Auto only)
- Models: Claude Code, Codex, Gemini CLI, OpenCode, Aider
impeccable
Impeccable installs a single /impeccable skill into AI coding tools, with 24 commands such as audit, critique, polish, typeset and live browser variant iteration. It writes PRODUCT.md and DESIGN.md for project context, and ships 59 deterministic detector rules that run in a CLI and browser extension without an LLM or API key. Hooks flag common AI design tells on UI file edits in supported harnesses.
Who it is for: Developers using AI coding agents who want better frontend design output
Strengths
- Supports many harnesses: Claude Code, Cursor, Codex, Gemini CLI, Copilot, Grok Build, Hermes
- 59 deterministic detector rules run without an LLM or API key
- Edit hooks surface design findings; Cursor blocks bad writes before they land
- Several install paths: npx installer, Git submodule, plugin marketplace, manual copy
Weaknesses
- Native edit hooks are not available on every harness; Hermes and Veto get none
- Roll API at impeccable.style is called when a new design starts; telemetry is opt-out
- Downloads an engine binary into ~/.impeccable/bin/ on first run if not bundled
- Codex and Grok Build need manual hook or folder trust approval after install
- no GPU
- Needs Node.js (only for npx installer), impeccable.style roll API
- Models: Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot
More in Skills, plugins and rules
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.