A coding skill for AI agents that enforces spec-first, human-readable code. Install it once and your agent gains three commands — /humane-spec to plan before touching any file, /humane-build to code the approved plan following strict human-readable rules, and /humanize-codebase to refactor for maintainability.
npx get-humane@latestYou will be prompted to pick your CLI:
Humane Skill Installer
Select your CLI:
1. Factory Droid → ~/.factory/skills/humane/
2. Claude Code → ~/.claude/skills/humane/
3. OpenAI Codex → ~/.codex/AGENTS.md
4. Gemini CLI → ~/.gemini/GEMINI.md
5. Antigravity → ~/.gemini/AGENTS.md
6. Cursor → ~/.cursor/rules/humane.mdc
7. Windsurf → ~/.windsurf/rules/humane.mdc
8. Cline → ~/.cline/rules/humane.mdc
9. All → Install for all tools above
Enter number [1-9]:
The installer creates the destination directory if it does not exist, then copies the skill files. No other changes are made to your system.
Three commands. One principle: no code without a plan, and no plan without a human who can read it.
/humane-spec — Plan before building.
The agent explores your codebase, then produces a structured spec:
- Goal (one sentence)
- Approach (2–5 bullets)
- Simplest version that fully solves the problem
- NOT doing (complexity explicitly rejected, with reasons)
- Files to create or modify
- Key function signatures or data shapes
No code is written until you approve the spec.
/humane-build — Code the approved spec.
The agent implements exactly what was specced, following hard rules:
- One function = one job. One file = one purpose.
- Names say what they do:
getUserById,formatCurrency,isEmailValid. - No abstraction until it appears twice in real code.
- Comments explain why, not what.
- Edit the minimum code needed. Ask before large refactors.
Runs lint and tests if the project has them, then reports what was built mapped to each spec point.
/humanize-codebase — Refactor for maintainability.
The agent analyzes your codebase against the "Humane" rules and proposes/applies refactors:
- Splits long functions (> 40 lines) and files (> 400 lines).
- Renames generic variables (
data,item,handler) to descriptive names. - Flattens deep nesting using guard clauses.
- Adds missing one-line summaries to exported functions.
You can also run the Humane validator manually on your codebase:
npx get-humane-validator@latest .(Or, if you installed it locally, use validate-humane .)
The validator checks for:
- File length (under 200 lines).
- Function length (under 20 lines).
- Explicit typing (e.g., no
anyin TypeScript).
| CLI | Install path | Format |
|---|---|---|
| Factory Droid | ~/.factory/skills/humane/ |
Full skill directory |
| Claude Code | ~/.claude/skills/humane/ |
Full skill directory |
| OpenAI Codex | ~/.codex/AGENTS.md |
Coding guidelines |
| Gemini CLI | ~/.gemini/GEMINI.md |
Coding guidelines |
| Antigravity | ~/.gemini/AGENTS.md |
Coding guidelines |
| Cursor | ~/.cursor/rules/humane.mdc |
Coding guidelines |
| Windsurf | ~/.windsurf/rules/humane.mdc |
Coding guidelines |
| Cline | ~/.cline/rules/humane.mdc |
Coding guidelines |
MIT