What problem does it solve? A model reviewing its own work shares every blind spot that produced the bugs, so self-review mostly re-confirms existing assumptions. This Skill gets a genuinely independent second opinion from a different AI provider's CLI agent and runs a structured back-and-forth until both agents converge. ## Core Features & Use Cases - Cross-provider verification: Detects the host agent and selects a verifier from a different model lab (codex, claude, gemini, or cursor-agent), with a clearly labelled same-provider fresh-context fallback when no independent provider is available. - Adversarial review loop: Sends a structured brief, collects severity-tagged findings, fixes or rebuts each one, and re-verifies across rounds using a findings ledger until explicit convergence. - Rigor and safety controls: Uses the best available model with maximum reasoning effort, read-only sandboxing, stdin-based briefs, and guardrails against leaking secrets to external providers. - Use Case: Before merging a security-sensitive PR, ask for a double-check; a Codex verifier at xhigh effort attacks the diff, you fix two findings and rebut one, and after three rounds both agents agree it is safe to merge. ## Quick Start Ask the agent to double check the finished work with a second opinion from another AI provider before merging.
