autoresearch

Autonomous iterative improvement loops cycle through Modify → Verify → Keep/Discard steps to refine artifacts. This process reduces manual intervention by running structured gates that enforce mechanical verification and explicit exit states per iteration.

Automating continuous verification uses debug and fix modes to translate goals into atomic changes with guard checks. Each single-change iteration is mechanically verified and rolled back if it fails, ensuring measurable progress without manual intervention.

The modify verify keep discard loop enforces runtime invariants by automatically loading foundational references and core principles. This guides decisions and safety checks, ensuring explicit holdout validation and atomic single-change progress per iteration.

Autonomous improvement workflows support security and ship modes to gate automation safely. These modes translate vague goals into measurable configurations, applying runtime invariants and explicit exit states to validate security checks and shipping readiness.

You do not need fully predefined configurations to start. The automation loop can take vague goals and translate them into measurable configurations and gates, using loaded core principles to guide the iterative refinement process.

You should avoid atomic single-change iterations when your task cannot be broken into mechanically verifiable steps. The loop requires explicit exit states and measurable progress, making it unsuitable for subjective artifacts lacking automated verification gates.