autoresearch

Automate iterative skill improvement by mutating SKILL.md files, running test batteries, and scoring outputs against a binary rubric. This autonomous loop applies atomic mutations and Git-based rollback to safely enhance skill reliability and performance progressively.

Git-based rollback ensures safe experimentation during autonomous skill mutation by automatically reverting SKILL.md changes if test cases fail. Atomic mutations are scored against a binary rubric, allowing only objective improvements to persist while discarding failed iterations.

Binary scoring in automated AI testing evaluates skill outputs as strict pass or fail against a deterministic rubric. This evaluation method removes ambiguity from autonomous improvement loops, ensuring only mutations that yield objective enhancements are retained.

Yes, autonomous mutation loops run on existing SKILL.md files without external dependencies. The workflow coordinates a three-phase process of Setup, Loop, and Debrief, storing all test artifacts locally under the .autoresearch directory.

Atomic mutations for code quality improvement are limited to tiny, isolated changes in SKILL.md files to preserve the skill's original purpose. The system enforces binary-only scoring, which may not capture nuanced qualitative improvements in skill clarity.

Evaluate AI testing results from skill improvement loops by reviewing stored artifacts in the .autoresearch directory during the Debrief phase. This workflow records test case executions and binary scoring outcomes to audit progressive enhancements and analyze failures.