iterate-until-verified

Applies an iterative execution and verification loop with independent quality gates to any task.

5.7k|685|Updated Feb 3, 2026
One-click install
npx skills add https://github.com/MengTo/Skills --skill iterate-until-verified
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: iterate-until-verified
Source: https://github.com/MengTo/Skills/tree/main/agent-skills/codex/iterate-until-verified
Command: npx skills add https://github.com/MengTo/Skills --skill iterate-until-verified

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Subjective goals like "make it perfect" or "production-ready" often lead to unverified, self-approved work. This Skill converts vague ambition into observable acceptance gates, separates making from judging, and loops until every gate passes with evidence.

Core Features & Use Cases

  • Acceptance Gate Matrix: Translates subjective quality language into binary pass/fail checks with defined verification methods and evidence requirements.
  • Independent Verification: Separates implementers from verifiers, supports blind comparison of anonymized candidates, and rejects unsupported self-assessments.
  • Compose Mode: Wraps the user's original task in a reusable prompt template that embeds the full verification workflow without altering the task itself.
  • Use Case: When asked to deliver a polished landing page, the agent defines gates for visual fidelity, accessibility, and responsiveness, assigns an independent reviewer, and iterates until all gates pass with screenshots as evidence.

Quick Start

Use the iterate-until-verified skill to complete this task with independent quality gates and evidence-backed verification before declaring it done.

Frequently Asked Questions about iterate-until-verified

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make an AI agent verify its own work before finishing?

Define observable acceptance gates with pass/fail conditions, assign verification to a role separate from the implementer, and require evidence such as test output or screenshots for each gate. The loop continues until every required gate passes or a concrete blocker is identified.

How to turn vague requirements like production-ready into testable criteria?

Translate subjective words into an acceptance matrix mapping each quality dimension to a verification method, binary pass condition, and evidence source. For example, production-ready becomes passing builds, regression tests, and security checks.

Can this workflow use subagents or independent reviewers?

Yes, independent workstreams can be fanned out to subagents when delegation is available and permitted. Each worker receives the task contract and returns artifacts, while one integrator maintains cross-workstream consistency.

What is blind comparison in AI output evaluation?

Blind comparison anonymizes and randomizes candidate outputs so the evaluator judges them against the same rubric without knowing authorship. It requires real comparable benchmarks and honestly labeled conditions to be valid.

When should the iteration loop stop instead of continuing?

Stop when all required gates pass with supporting evidence, or declare blocked when a gate depends on missing access, unavailable inputs, or infeasible constraints. Never weaken a gate just to claim success.