three-brain

What problem does it solve?

This Skill prevents low-quality self-review by routing Claude’s own outputs into specialized reviewer/rescue models, and it ensures multimodal evidence is extracted with the right preprocessing when media or long documents are involved.

Core Features & Use Cases

• Forced expert review for Claude’s output: Detects when users ask to “check/review/proof/verify” Claude’s just-produced work and routes the review to Codex instead of performing a blind self-audit.
• Adversarial risk review on high-blast-radius edits: Triggers Codex adversarial review when an active edit touches sensitive code paths such as auth, billing, migrations, secrets, policy/ACLs, and deployment/IaC concerns.
• Multimodal handling with preprocessing contracts: Routes video/audio/PDF-heavy requests to Gemini with explicit pipelines (clip/cap duration, extract audio, cap pages, and require timestamped or file/line citations).
• Failure-counter rescue: Monitors repeated failure patterns and escalates to Codex rescue after defined retries on the same failing operation.
• High-stakes consensus mode (when explicitly requested): Runs Claude/Codex/Gemini in parallel and enforces a structured adjudication template.

Quick Start

Ask for a second opinion by saying: "Review what you just wrote for bugs and missing tests."