I open-sourced a 3-agent blind eval team. Any agent runtime can call it for pre-commitment review of its own plans.

An open-source, 3-agent blind evaluation workflow, released this weekend, allows any AI agent runtime to pre-commit review its plans via an HTTP endpoint. This system addresses the issue of models reliably self-evaluating by providing an external, blind primitive for honest assessment.