What problem does it solve? Running an open-ended ML research project requires constant hypothesis management, experiment tracking, literature review, and synthesis — work that stalls when a human cannot supervise continuously. This Skill turns an AI agent into an autonomous research project manager that keeps experiments running, records results, and steers direction without waiting for human input. ## Core Features & Use Cases - Two-Loop Research Engine: An inner loop runs rapid experiments against locked protocols and measurable metrics, while an outer loop periodically synthesizes results into findings.md and decides whether to deepen, broaden, pivot, or conclude. - Structured Research Workspace: Initializes research-state.yaml, research-log.md, findings.md, literature/, experiments/, and to_human/ directories so state survives across sessions and agent restarts. - Continuous Operation: Sets up a mandatory 20-minute continuity loop via Claude Code /loop or an OpenClaw cron job so research never idles between cycles. - Domain Skill Routing: Delegates execution to specialized skills for data processing, training, evaluation, interpretability, and paper writing (NeurIPS/ICML/ICLR LaTeX templates). - Use Case: Give the agent a question like "does cosine warmup improve small-model convergence", and it will survey literature, form hypotheses, run experiments, plot the optimization trajectory, generate progress reports, and draft a paper when the evidence supports a contribution. ## Quick Start Ask the agent to start an autonomous research project on your research question and let it bootstrap the workspace, set up the continuity loop, and begin running experiments.
