Cold Validation Launches Open-Source Agent Review

This is less a flashy agent framework than a governance primitive, and that is why it matters. It turns code review into a reproducible control with real separation of duties, instead of another AI second opinion with the builder's memory still attached.

• –Zero shared context is the right abstraction here: the reviewer sees only artifacts, so the audit is about evidence, not the builder's hidden rationale.
• –Fingerprints on findings are a practical antidote to review churn; resolved issues stay resolved instead of reappearing under new IDs.
• –The orchestrator overriding a naive pass verdict is important because it makes policy and blocking findings higher authority than model confidence.
• –The repo is intentionally lightweight and shell-heavy, which is a feature for a control layer. The 35 mechanical tests suggest they are trying to prove determinism, not just demo a prompt.