Instruction checker
Find instructions that may be repetitive, overbroad, or pulling in different directions. Every suggestion explains the rule that triggered it.
Your input stays in this tab. No model calls or automatic saving. Export anything you want to keep before leaving.
Check your instructions
Up to 20,000 characters. Text stays in this browser tab.
Review your output
Your output will appear here.
How to use the instruction checker
- Paste your custom instructions, system prompt, CLAUDE.md, or AGENTS.md text.
- Run the check and inspect the flagged lines. A possible conflict may be intentional when it has a clear scope.
- Select duplicate lines to remove. Rewrite other suggestions manually in the output, then test the changes on real tasks.
A concrete example
“Always be concise” plus “Explain every detail” produces a possible conflict suggestion. A scoped rule such as “Be brief for routine questions; explain complex decisions” may fit better.
Questions and limits
What does the instruction checker detect?
It checks repeated non-heading lines, always/never wording, some safeguard-override requests, requests to hide uncertainty, paragraphs over 400 characters, and brevity alongside exhaustive-detail wording. It does not understand every semantic conflict.
Is a clean result proof that a prompt is safe or effective?
No. These are transparent text rules, not a security audit, model evaluation, or prompt quality score. Review the full text and compare actual responses.
Go further
Test personality prompts against plain language explains the method and practical limits.
- Communication preferences builder — Create instructions without taking a personality test.
- Muse Soul.md editor — Safely add communication preferences to Muse Soul.md.
- Instruction test bench — Compare real answers before and after an instruction change.
For developers and agents: tool catalog (JSON), local API documentation, and reusable JavaScript functions. These process text; reading a file does not authorize installing instructions.