What problem does it solve? Research tasks often stop prematurely when a model declares itself done without verifiable evidence. This Skill replaces the deprecated omx autoresearch CLI with a stateful, native-hook workflow that keeps iterating until an explicit validator confirms the research deliverable meets its mission criteria. ## Core Features & Use Cases - Validator-Gated Completion: The loop only stops when a completion artifact records a passing result from either a mission validator script or a prompt-plus-architect review. - Persistent Mode State: Validation mode, artifact paths, and validator configuration are stored in .omx/state/.../autoresearch-state.json so the loop survives across turns. - Structured Intake: Pairs with $deep-interview --autoresearch to clarify the mission and evaluator, then materializes mission.md, sandbox.md, and result.json specs. - Use Case: You need a bounded research report that must demonstrably improve a metric beyond baseline. Configure a mission validator script, start the loop, and let auto-nudges continue until the result artifact records a passing verdict. ## Quick Start Run $deep-interview --autoresearch to define the mission and evaluator, then start $autoresearch with your chosen validation mode and let it iterate until the completion artifact passes.
