What problem does it solve? Running a complete research project requires juggling literature surveys, idea validation, implementation, GPU experiment deployment, and iterative paper review — a fragmented process that is hard to manage manually. This Skill chains the entire lifecycle into one autonomous pipeline with configurable human checkpoints. ## Core Features & Use Cases - Idea Discovery with Human Gate: Runs literature survey, idea generation, novelty checking, and pilot testing, then pauses at Gate 1 for user approval (or auto-selects the top idea when AUTO_PROCEED is enabled). - Autonomous Implementation and Deployment: Extends pilot code into full multi-seed experiments with proper baselines, then deploys to remote GPU servers via screen sessions. - Auto Review Loop: Iterates up to 4 rounds of GPT-5.4 review and Claude Code fixes until the work scores 6/10 or higher, producing AUTO_REVIEW.md and a final pipeline report. - Use Case: A researcher says "run the full pipeline on efficient video tokenization" in the evening; the skill discovers and validates ideas overnight, launches GPU experiments, and delivers a reviewed, submission-ready paper by morning. ## Quick Start Run the full research pipeline on the topic of machine-oriented video compression with AUTO_PROCEED disabled so I can pick the idea myself.
