What problem does it solve? Running a full research project requires coordinating literature ingestion, idea generation, experiment design, execution, evaluation, and paper writing across many sessions, with progress easily lost between steps. This Skill orchestrates that entire pipeline with human gates, state persistence, and cross-session recovery. ## Core Features & Use Cases - End-to-End Pipeline Orchestration: Chains sub-skills (/ideate, /exp-design, /exp-run, /exp-eval, /paper-plan, /paper-draft, /paper-compile) into a five-stage workflow with two human approval gates. - Cold-Start Bootstrap: Automatically searches DeepXiv, Semantic Scholar, and arXiv and ingests the top 5 papers when the wiki is empty, so no manual initialization is needed. - Non-Blocking Experiment Execution: Deploys experiments asynchronously, saves progress to pipeline-progress.md, and resumes later via --start-from flags for collection and verdict stages. - Use Case: A researcher provides a direction like "sparse LoRA for edge devices"; the pipeline bootstraps a literature wiki, proposes ideas, designs and deploys experiments, evaluates claims, and produces a compiled paper draft for a target venue such as ICLR. ## Quick Start Run the research pipeline on the direction "sparse LoRA for edge devices" in auto mode targeting ICLR, resuming from saved progress if a previous pipeline exists.
