Adversarially verify code review findings

Adversarial review gate for a backpressured loop

Finding subtle bugs or policy violations across large codebases that require consistent multi-agent consensus.

• · args object (workflow parameters)
• · source document or file

• · structured result object
• · formatted report or summary
• · output files or modifications

• · Claude agent runtime (Workflow)
• · agentic subagents (multi-agent orchestration)

• · structured output schema defined
• · Claude agent runtime available
• · file system access for read/write

• · schema validation failure (invalid JSON or required fields missing)
• · agent timeout or inference failure
• · reviewer consensus not reached (oscillation or deadlock)
• · early termination signal (kill signal or max cycles hit)

• · strict structured output schema enforces correctness
• · adversarial review ensures rigor
• · kill signals or approval gates prevent bad state progression
• · independent verification phase double-checks results
• · convergence logic (loop termination on stable state)