autoresearch

Automate OpenClaw skill optimization by running iterative sub-agent cycles that test outputs, apply binary evaluation scoring, and mutate prompts. It autonomously experiments with prompt variations to improve and retain measurable quality enhancements.

Autonomous prompt mutation systematically alters prompt text during sub-agent runs to test different variations. It uses binary evaluation scoring to measure output quality, automatically retaining successful mutations and rolling back failures to ensure consistent improvements.

Start an autonomous evaluation loop by providing a SKILL.md path and a set of test prompts. The pipeline gathers context, runs baseline tests, mutates the skill instructions, and logs results for dashboarding to track performance.

Yes, the autoresearch process retains successful prompt changes automatically. It applies targeted prompt mutations during sub-agent runs and keeps the modifications only when binary evaluation scoring confirms a measurable quality improvement.

Track quality improvements during skill tuning using the built-in dashboarding and logging system. The pipeline records baselines, maintains changelogs of prompt mutations, and generates per-skill dashboards for full traceability of autonomous evaluation loops.

Avoid using autonomous prompt mutation when a skill lacks clear test prompts for binary evaluation or when baseline performance cannot be measured. The optimization pipeline relies on consistent scoring to evaluate sub-agent runs and retain successful prompt changes.