Tour

Suggestions for your agent's prompts

Prompt Analytics compares a prompt you wrote for an agent with prompts other people wrote. It suggests instructions that similar prompts and prompts of the same kind include, and shows the prompts each one comes from.

You give Prompt Analytics your prompt. You may also rate the suggestions and let your agent report how they worked.

Prompt Analytics keeps each prompt you send with its record. The site shows your records only to you. You can turn keeping off on the Keys page before you send a prompt, and delete a record from its page.

Connect your agent

Sign in with GitHub, make a key for your agent, and paste this message to it. The key goes where YOUR_KEY stands. Your agent knows Prompt Analytics as the MCP server "facet".

Read https://facet-prompt-analytics.fly.dev/skill.md and set yourself up for facet with this key: YOUR_KEY

Set up Claude Code yourself instead

claude mcp add --transport http facet https://facet-prompt-analytics.fly.dev/mcp --header "Authorization: Bearer YOUR_KEY"

Agents without MCP use the JSON API that skill.md, the setup file, describes.

Ask for suggestions

Ask your agent for suggestions on a prompt it uses, or paste a prompt on the Analyze Prompt page.

Use facet to suggest instructions for the system prompt in this repo, and show me which prompts each one comes from.

Or have your agent find every prompt in a repository first.

Find the prompts in this repository, the system prompts and prompt templates in code and files, and list them with their files. Ask me before sending any, and send no secrets. Then analyze the ones I choose with facet and show me the suggestions.

Analyzing a prompt here needs sign-in, since Prompt Analytics keeps each analysis as a record of yours.

Review and rate

Prompt Analytics saves each analysis as a record of yours, listed beside the Analyze Prompt form. You can rate the analysis as a whole and each suggestion in it. Prompt Analytics keeps the thumbs with the record. They never change what Prompt Analytics suggests.

A thumb on the run: Whether the run as a whole helped you.
A thumb on a suggestion: Whether that one instruction helped you.

Let your agent report back

After you apply suggestions and run your agent, your agent can tell Prompt Analytics what it adopted and how the run went. You choose on the Keys page when it may.

Ask me each time: Your agent asks you before each report. This is the default.
Always: Your agent reports without asking.
Never: Prompt Analytics keeps no report from your agent.

After you apply facet's suggestions and run the agent, ask me, then report to facet what you adopted and how it went.

This message fits "Ask me each time".

Choose on the Keys page