Agent code review before human review
Two agents read the diff before you do: one checks it against the acceptance criteria and calls out risk, the other runs a security-specific pass on anything with real blast radius.
How it works
ReviewerAgent reads the diff against the spec's acceptance criteria the moment a proposal is created, and calls out risks alongside a summary.
SecurityReviewAgent runs a focused pass — secrets, injection, authz, data exposure — only on proposals flagged high or critical.
Malformed or failed output shows as “pending” or is skipped silently — never a fake pass, so a human trusts what's missing as much as what's there.
Both agents annotate the same proposal gate; neither can approve, block, or reject on its own.
The agents behind it
Human control
Proposal gate — both reviews are advisory annotations; the human weighs them alongside test evidence before deciding.