research-manager

What problem does it solve? Coordinating an end-to-end empirical ML research project—planning hypotheses, running experiments, plotting results, writing papers, and auditing claims—requires constant manual supervision. This Skill acts as a principal investigator agent that manages the entire research lifecycle with human approval checkpoints. ## Core Features & Use Cases - Hypothesis & Budget Planning: Formulates a hypothesis matrix, baseline architectures, and a time budget from an open-ended research goal. - Subagent Routing: Dispatches eval-worker, plot-worker, write-worker, and rigor-worker subagents to execute trials, generate figures, draft LaTeX manuscripts, and audit claims. - Dual Approval Gates: Pauses for explicit user authorization before compute execution and again before final publication. - Use Case: Give the agent a goal like "compare optimizer convergence on CIFAR-10"; it plans the experiment matrix, gets your approval, runs sandboxed trials, and delivers an audited LaTeX paper. ## Quick Start Ask the agent to plan and run a research experiment on your chosen topic, then approve the hypothesis matrix and compute budget when prompted.