What problem does it solve? Research teams struggle to systematically generate novel, feasible ideas without repeating past failures or duplicating existing work. This Skill automates a structured ideation pipeline that scans the research landscape, brainstorms with dual models, filters candidates, validates novelty, and records results in a persistent wiki. ## Core Features & Use Cases - Five-Phase Pipeline: Landscape scan (wiki + WebSearch + Semantic Scholar + DeepXiv), dual-model brainstorm (Claude + Review LLM), first-pass feasibility/novelty filter, deep validation via /novelty and /review, and wiki write-back. - Anti-Repetition Memory: Eliminated ideas are written to the wiki with failure reasons, forming a banlist that prevents future runs from revisiting dead ends. - Maturity-Adaptive Behavior: Adjusts search breadth based on wiki maturity (cold/warm/hot), expanding external search when the knowledge base is empty. - Use Case: A researcher runs the pipeline on the direction "sparse LoRA fine-tuning"; the Skill scans recent arXiv papers, generates 10 candidates, filters to 4, validates the top 3, and writes ranked idea pages with graph edges into the wiki. ## Quick Start Ask the assistant to run the ideate pipeline on a research topic such as "efficient fine-tuning methods" and write the top ranked ideas into the wiki.
