What problem does it solve?
Running a complete research project requires manually chaining literature surveys, idea validation, experiment implementation, GPU deployment, and iterative peer review, which is slow and error-prone. This Skill automates the entire lifecycle as a single pipeline.
Core Features & Use Cases
- Idea Discovery with Human Gate: Runs literature survey, idea generation, novelty checking, and pilot testing, then pauses at a checkpoint for user approval (or auto-selects the top-ranked idea when AUTO_PROCEED is enabled).
- Autonomous Experiment Deployment: Implements full-scale experiments from pilot code, syncs to remote GPU servers, and launches runs in screen sessions with monitoring.
- Auto Review Loop: Iterates up to 4 rounds of GPT-based adversarial review (medium/hard/nightmare difficulty) followed by automated fixes until the work reaches a target score.
- Use Case: A researcher provides a broad direction like "efficient long-context attention", goes to sleep, and wakes up to a reviewed, scored paper draft with full experiment history in AUTO_REVIEW.md.
Quick Start
Run the full research pipeline on the topic of sparse mixture-of-experts routing with AUTO_PROCEED enabled and medium reviewer difficulty.
