review-agent-setup

What problem does it solve?

This Skill prevents an AI agent from taking review-surface actions without explicit human approval, reducing the risk of unreviewed comments, merges, release actions, or CI changes.

Core Features & Use Cases

• Human-in-the-loop gating: Requires a deliberate approval window before Claude Code can post reviews, comments, merge pull requests, or close issues.
• Auditable receipts: Produces signed approval and denial records so every attempted action can be verified later.
• Policy enforcement: Uses Cedar-based rules to block sensitive actions and support locked-down or compliance-focused workflows.
• Practical use case: A maintainer can approve a single PR review session, let the agent act, and then close the window immediately after the action completes.

Quick Start

Install the plugin, copy the default policy into your project, create the receipts and approval files, and then let the agent operate only during an explicitly opened approval window.