Promptocop — AI skill for Claude Code

A prompt linter for Claude Code and Cursor. Analyzes your prompts for anti-patterns and surfaces issues as context for the model — like ESLint, but for the things you type.

---

Setup

There are two integration modes. Pick one — or use both.

Option 1: Agent Skill (recommended)

Installs a `SKILL.md` that Claude Code and Cursor load automatically. The editor applies the rules using its own reasoning — no shell execution, no `npx` on every prompt:

# Claude Code (default)
npx promptocop skill install

# Cursor
npx promptocop skill install --target cursor

This writes a `SKILL.md` to your editor's skills directory. Your editor picks it up immediately. To install at project scope instead:

npx promptocop skill install --project
npx promptocop skill install --target cursor --project

To remove:

npx promptocop skill uninstall
npx promptocop skill uninstall --target cursor

**Other editors:** If your editor supports agent skills but isn't listed above, print the `SKILL.md` to stdout and place it wherever your editor expects:

npx promptocop skill generate > /path/to/your/editor/skills/promptocop/SKILL.md

Option 2: Hook (shell-based)

Adds an editor hook that runs `npx promptocop lint --hook -` before every prompt. Regex-based, fast, zero Claude tokens:

# Claude Code (default)
npx promptocop hook install

# Cursor
npx promptocop hook install --target cursor

For Claude Code this adds a `UserPromptSubmit` hook to `~/.claude/settings.json`. For Cursor it adds a `beforeSubmitPrompt` hook to `~/.cursor/hooks.json`. When a prompt has issues, the lint results are injected as context so the model can factor them into its response — nothing is blocked by default.

To remove:

npx promptocop hook uninstall