autoresearch

Installation
SKILL.md

Autoresearch

Use this skill when the goal is to improve a specific, measurable outcome through repeated automated experiments — write a change, measure it, keep it or revert it, repeat. You define the metric; the agent runs the loop autonomously until interrupted or the budget is exhausted.

This is distinct from test-driven development (writing tests for behavior) and systematic debugging (diagnosing a known failure). Use this skill when optimization through iteration is the work.

Use this skill when

  • You want to iteratively improve a measurable outcome: execution time, bundle size, test pass rate, build time, latency, coverage, complexity, or a custom benchmark score.
  • You have a metric command that produces a numeric result and you want the agent to run experiments autonomously until interrupted or a budget is reached.
  • The right solution is not obvious upfront and you want to explore the space empirically.
  • You want to hill-climb: keep changes that help, discard changes that do not, and log everything.

Do not use this skill when

  • The task is writing tests for a new feature or behavior (use test-driven-development).
  • The goal is diagnosing why something is broken (use systematic-debugging).
  • There is no measurable metric — the outcome is purely qualitative or subjective.
  • The codebase does not have a git repository or the user has not confirmed a dedicated experiment branch is acceptable.
Installs
5
GitHub Stars
4
First Seen
Sep 1, 2026