review-agent-setup

Cedar policy checks enforce human-in-the-loop approval gates by blocking sensitive AI review actions like posting comments or merging pull requests. Cedar rules evaluate whether an approval window is explicitly opened before allowing Claude Code to execute tool calls.

To set up an approval gate for Claude Code reviews, install the plugin, copy the default Cedar policy into your project, and create the receipts and approval files. The agent can then only operate during an explicitly opened approval window.

Yes, you can use Cedar policy to block AI agents from editing CI and protected branch workflows. Cedar-based rules enforce compliance-focused workflows by preventing sensitive actions unless explicit human approval is granted.

Auditable receipts for AI review actions are generated automatically as signed approval and denial records. Every attempted tool call by the agent produces a signed receipt so the action can be verified later for compliance.

Approval windows require active human management to open and close, meaning the agent cannot act autonomously outside that window. Maintainers must deliberately close the approval window immediately after the review action completes to maintain security.

Human-in-the-loop gating works for triaging issues and publishing releases by requiring a deliberate approval window before Claude Code can perform these actions. Signed receipts and Cedar policy checks ensure every attempted action is auditable.