Don't finish until the result is verified
by Ramon Sosa ✓
Summary
Defines a testable goal, a validation checklist and a correction loop so the AI only delivers a version that truly meets the bar.
The prompt
Complete [GOAL]. Do not finish until the result meets all of these criteria: [MEASURABLE CRITERIA] Verify the work using these tests or available tools: [TESTS OR VERIFICATION METHOD] Mandatory loop: 1. Produce a first version. 2. Check each criterion explicitly. 3. Identify failures, omissions and contradictions. 4. Fix the result. 5. Repeat the verification until every criterion passes or there is a real blocker you cannot resolve. Do not confuse “looks right” with “is verified”. Do not declare success without evidence. If a check cannot be run, mark it as not verified and explain exactly what is missing. Deliver only: 1. Final result. 2. A short table of checks with status: passed, failed or not verifiable. 3. Remaining risks or limitations.
How to use
When to use it: Use it for code, documents, research, campaigns, pages, spreadsheets, migrations, audits and any work with testable criteria. Variables: - [GOAL]: the final result. - [MEASURABLE CRITERIA]: length, fields, tests, links, calculations, requirements or standards. - [TESTS OR VERIFICATION METHOD]: tests, build, URL checks, counts, comparison, validators or manual inspection. Why it works: It includes verification statuses, blocker handling and an explicit ban on declaring success without evidence.
Recommended
See more →
activepieces
Open source automation with AI and MCP, a self-hosted alternative to Zapier/Make.
Agno
A lightweight, fast framework for building agents with memory, tools and RAG.
AI Agents for Beginners
Official Microsoft course: 18 lessons to start building AI agents.
More resources like this
Get the newest prompts and tools every week.