autoresearch-create

What problem does it solve?

This Skill sets up and runs an autonomous experiment loop that automates iterative optimization of a measurable target by proposing changes, executing benchmarks, and keeping only the interventions that improve the primary metric.

Core Features & Use Cases

• Automated experiment lifecycle: initialize sessions, run baselines, loop experiments, and log structured results so optimization proceeds without manual oversight.
• Deterministic benchmarking and checks: supports a fast benchmark script that emits structured METRIC lines, optional correctness checks that gate keeps, and confidence scoring across runs.
• Safe git workflow and annotations: create a feature branch for the session, auto-revert failed or discarded changes, and annotate each run with ASI to preserve reasoning for resumption.
• Use Case: speed up a build or benchmarked workload by continuously proposing small code changes, evaluating the effect on a primary metric, and retaining only validated improvements.

Quick Start

Create autoresearch.md and a fast autoresearch.sh, run init_experiment, record a baseline, and start the autonomous loop to iteratively optimize the chosen metric.