About this Skill
Review Loop is a reusable Agent Skill for alternating implementation and critique. The worker completes a draft, gives the work product and task-specific criteria to a reviewer, then addresses actionable feedback until the quality gate or maximum loop count is reached. The upstream defaults are two to four loops and a reviewer score of 8 out of 10; users can request different limits.
Why we picked it
A structured review prompt rather than an independent guarantee of quality. Scores depend on the reviewer and criteria; extra review rounds consume time and model usage.
What it helps with
- Get structured feedback and revisions on a code change or technical specification.
- Review an existing document using explicit criteria and a defined loop limit.
How the review loop works
Define the work and criteria
Provide the work product, the expected result and criteria suited to the task.
Review and revise
Ask a separate reviewer for specific findings, then revise the work against relevant feedback.
Stop and report
Stop when the requested gate is met or the maximum number of loops is reached, and report outstanding issues.
Before you use it
- A compatible Agent with subagent support gives a separate reviewer context. Without it, the upstream suggests a fresh chat and manual transfer of the work product.
- Each review round consumes model time and may incur usage costs. A model-assigned score is subjective and does not replace tests or human review for critical work.
Sources & references
Original sources are linked below. Practical starting points are editorial suggestions from SkillHey.