autoresearch

name	autoresearch
description	Autonomous experiment loop that tries ideas, measures results, keeps what works, and discards what doesn't. Use when the user asks to optimize a metric, run an experiment loop, improve performance iteratively, or automate benchmarking.

Autoresearch

Run the /autoresearch workflow. Read the prompt template at ../prompts/autoresearch.md for the full procedure.

Tools used: init_experiment, run_experiment, log_experiment (from pi-autoresearch)

Session files: autoresearch.md, autoresearch.sh, autoresearch.jsonl

/goal Inheritance

This skill inherits the SeaBridgeAI /goal default protocol. Frame the work with a persistent goal, Definition of Done, validation plan, risks, dependencies, expected artifacts, and completion evidence. Do not claim completion until the DoD is validated or a hard blocker is documented.

Canonical protocol: C:\Users\adelm\SeaBridgeAI\everything-claude-code\protocols\GOAL_PROTOCOL.md