Check a whole prompt library at once
A shared doc of prompts grows by copy and paste. Some are three versions of the same thing, some hold a customer's email, and the one everybody uses is the weakest. Paste or upload the library and see its health, the copies, the private details, the gaps and the prompts to fix first.
Read how to build a team prompt library people actually use.
New to this? What it does, in plain words
It reads every prompt in a library the way a careful team lead would, and tells you where to start.
For a team's shared prompts, a prompt pack you bought, or your own saved prompts. Each one is scored on the ten checks and matched to the task it does, so the library can be compared with the 669 expert prompts in the repository.
What you need
Nothing. No API key and no sign-up. When you press Audit, the prompts go to thinkhard.si, are read and are forgotten.
How to use it
- Paste the prompts with a line of three dashes between them, open a CSV, JSON or text file, or use the prompts saved in My Prompts in this browser.
- Press Audit the library.
- Start with Fix these first. Each one says what is wrong and which expert prompt to start from.
The prompts
Why libraries go stale
Nobody owns a shared prompt doc. People copy a prompt, change a line for their case and paste it back as a new one. Customer details from a real case stay in. And the first prompt anyone wrote for a task stays the one everybody uses, because nobody compares it with the others.
- You. Bring the libraryPaste, upload, or use My Prompts.
- thinkhard.si engine. Read every promptThe ten checks, the task it does, how ready it is for it.Instant
- thinkhard.si engine. Find the copiesNear-copies grouped, with the best one to keep.
- thinkhard.si engine. Find the risksEmails, phone numbers, card numbers and keys.
- You. Fix the right onesWeakest first on the tasks used most, with the expert prompt to start from.
What it reports
- Health. One number from 0 to 100: the average score on the ten checks, less for prompts with private details and for copies.
- How scores spread. How many prompts sit at each level from 1 to 10, and the checks the library misses most, which is usually one habit to fix across the team.
- Near-copies. Prompts that share most of their wording, grouped, with the strongest one to keep.
- Private details. Email addresses, phone numbers, card or account numbers, keys and passwords: what should never sit in a prompt others copy.
- Coverage and gaps. Which collections the library covers, and the expert prompts in those collections it does not have yet.
- Fix these first. The weakest prompts on the tasks the library uses most, each with its biggest gap and the expert prompt for the same task.
Questions
What file formats can I upload?
CSV with a column called prompt or text (and optionally name and system), JSON as a list of strings or of objects with text or prompt, and plain text or Markdown with a line of three dashes between prompts. My Prompts sends the latest version of every prompt saved in this browser.
Are my prompts stored?
No. When you press Audit the library, the prompts go to thinkhard.si, are read and are forgotten. Nothing is written to disk or logs and nothing is used for training. See the privacy page.
How is a near-copy found?
Two prompts that share most of their runs of five words are grouped, however they are named. The one with the higher score is suggested as the one to keep, and the others can be retired or merged into it.
Can I run it on a schedule or from code?
Yes. POST /api/v1/audit with the prompts returns the same result as JSON, and AI assistants that use the thinkhard.si MCP server get it as the audit_library tool.