iterative-self-review

What problem does it solve?

Run a clean-context blind verification loop when the user explicitly asks for verification, review, re-check, cross-check, proof, or another second-pass quality gate. The main agent drafts the current response, a clean-context sub-agent verifies only the applicable user requirements plus that current response, and the main agent revises and re-verifies until every blocker/major issue is gone and only minor residuals remain. Applies to all response types: code changes, document edits, design proposals, summaries, plans, recommendations, explanations, and file-change reports. If the user explicitly narrows verification to a specific part, scope both the user-input slot and the response slot to only that requested part. Do not auto-trigger on ordinary conversation or routine task completion.

Core Features & Use Cases

• Blind verification loop: a separate sub-agent checks the main draft against user requirements without editing the artifact.
• Applies to all response types: code changes, documents, designs, summaries, plans, and reports.
• Use Case: Before delivering a complex code change or design proposal, run this skill to verify requirements and surface only minor issues to the user.

Quick Start

Trigger the iterative self-review to run a blind verification pass on the current draft and iterate until all blockers are resolved.