Cook with AI Enrollment closed
AI & LLMs Runs locally

System Prompt Linter

Lint system prompts in your browser: contradictions, vague wording, missing format spec, wasted tokens. Rule-based, not AI.

ELI5

It checks your AI instructions for common mistakes.

A little more detail

What this tool does

A system prompt can contain conflicting rules, vague phrases, repeated text, or an unclear output format. This linter catches mechanical problems with fixed rules. No model reads your prompt.

Use it to
  • Find instructions that say both "always" and "never"
  • Spot vague wording, duplicate sentences, and missing output formats
  • Review sensitive prompts without sending them to another AI service
Interactive workspace Results update as you type
System prompt

Not stored in the URL or sent anywhere — linting runs locally in your browser.

Findings
~ tokens (estimate)
Paste a prompt above to see findings.
No issues flagged — still read it yourself; heuristics miss nuance.

Rule-based checks only — no LLM reads your prompt. Good for catching obvious mistakes, not for judging whether instructions will work in production.

System prompts are code you never typecheck. This linter runs simple heuristics: length bounds, keyword contradictions, vague phrases, missing output format, politeness filler, ALL CAPS ratio, duplicate sentences, unclosed ``` fences, and multiple "you are…" personas.

It is not an AI review and it cannot know your product. A clean lint pass does not mean the prompt works — it means you avoided a few common footguns. Always test on real inputs.

Token estimate uses ~4 characters per token — close enough for English prose, wrong for code-heavy prompts. Treat it as a ballpark.