AutoResearchHarness-Local Automated Research.

A multi-agent research product built for automated deep learning research inside a single Harness workspace.

🧠

Research-Only Product Surface

The public product surface is focused on research orchestration only: ideation, design, implementation, training, analysis, writing, and review.

📂

Research state lives in `.omr/research/` with `project.json`, `pipeline.yaml`, `HANDOFF.md`, result artifacts, and review outputs.

🔄

Claude plans, Codex codes and judges, Gemini scouts. Work is intentionally reviewed by a different model family.

🎛️

Gates are now `human`, `auto`, or `default`. No legacy `auto-judge` protocol in the runtime path.

📡

Stage mirrors, handoff files, and trace IDs make it possible to stop, resume, and recover research sessions without losing context.

🖥️

The product runs only inside `~/Claude/Harness`, with a single source repo at `~/Claude/Harness/omr`.