autoresearch

name

autoresearch

description

Run the knowledge-driven OpenUI research loop that coordinates the training pipeline with curated knowledge. Read and update the repo + personal brains (OKF / Obsidian) and OpenWiki, run the prior-work discovery loop to find related research, drive the autotrain hypothesis loop to generate falsifiable hypotheses, and file new ideas and experiments as Linear issues / milestones / projects. Use to ORCHESTRATE research; to RUN a pipeline phase use autotrain; to RUN an experiment campaign (init/research/validate/execute) use openui-autoresearch; to change harness code use improve-openui-harnesses.

Autoresearch — the OpenUI research loop

Facade for orchestrating research with progressive disclosure: this file routes; each stage's full instructions live in references/ and are read only when that stage runs. This skill coordinates the machinery it does not own — it drives autotrain (pipeline + hypothesis loop), openui-autoresearch (campaign execution), the openwiki CLI, the brains under docs/brains/, and the Linear MCP. It never builds a parallel trainer, evaluator, or knowledge store.

What this loop does

Turns curated knowledge into falsifiable, tracked experiments and folds results back into knowledge:

brains + OpenWiki  ──▶  discovery (prior work)  ──▶  hypotheses  ──▶  Linear issues
      ▲                                                                     │
      └──────────────  update brains from measured results  ◀──────────────┘

Stage routing

Orient once per session with references/loop.md + references/contracts.md — loop.md owns the full ordered sequence. The stages below run in table order; read a reference only when its stage runs.