What problem does it solve? Going from a vague research direction to a refined, venue-ready proposal requires many disconnected steps: surveying literature, brainstorming ideas, checking novelty, simulating peer review, and iterating on the proposal. This Skill chains all of these phases into a single autonomous workflow with checkpoint logging and resumable state. ## Core Features & Use Cases - Four-Phase Pipeline: Runs literature survey, idea generation (8-12 ideas filtered to 4-6), multi-dimensional screening (novelty, venue reviewer simulation, strategic fit), and deep refinement of the top 2 ideas. - Autonomous Operation with State Persistence: Logs every checkpoint decision to outputs/PIPELINE_LOG.md and writes PIPELINE_STATE.json after each phase so an interrupted run can resume from the last completed phase. - Venue-Targeted Screening: Passes a target venue (e.g., ICML, NeurIPS, VLDB) through to reviewer simulation and composite ranking. - Use Case: A PhD student types a direction like "efficient long-context LLM inference" and receives a final IDEA_DISCOVERY_REPORT.md containing the landscape map, ranked ideas with composite scores, two refined proposals, and documented reasons for every eliminated idea. ## Quick Start Run the full idea discovery pipeline on the research direction "efficient long-context LLM inference" targeting ICML and produce the final report.
