autoresearch

Yes, you can benchmark and improve an existing repo skill by running repeated evaluations against it. The autonomous evaluation loop tracks scores iteratively to guide optimization and select the best-performing prompt mutations for your skill.

Autonomous evaluation is a framework that benchmarks and improves existing repo skills through repeated evals, prompt mutations, and score tracking. It runs test cycles across multiple prompts and configurations to identify superior variants and guide optimization.

Yes, the autonomous evaluation framework is designed for Codex and Codex-like environments. It requires the canonical autoresearch workflow and compatible evaluation tooling to track and compare performance across iterations within these platforms.

Yes, compatible evaluation tooling is required to track and compare performance across iterations. The framework relies on these tools to collect metrics over iterations, enabling score tracking and comparison to guide prompt optimization.

The best way is to start an autoresearch cycle against an existing skill, then review the score history and mutate prompts for improvement. This allows teams to iteratively enhance skills by continuously evaluating variants and selecting the best-performing prompts.