What problem does it solve? Turning a vague research idea into a well-grounded, paper-ready proposal requires literature search, novelty assessment, and narrative structuring that is slow and error-prone when done manually. This Skill automates that upstream stage, producing a validated story.json, a Markdown proposal, and a verified citation seed for downstream paper writing. ## Core Features & Use Cases - Idea packaging and retrieval: Normalizes a raw idea into a structured brief, then recalls candidate research patterns from a knowledge graph via kg_recall.py (semantic + lexical + graph-boosted ranking). - Agentic literature search: Runs intent-tagged web searches (core method, contrast, evaluation) and records every real paper into retrieved_papers.json as a citation seed, never fabricating abstracts. - Critique-refine-fusion loop: Iteratively critiques the story against blind exemplars, routes defects to reserved patterns, and fuses conceptually with a calibrated pass bar capped at 3 rounds. - Integrity gates: novelty_check.py embeds the story and flags collision risk (0.82/0.88 bands), while story_lint.py enforces the 8-field schema and rejects fabricated results. - Use Case: A researcher with a one-line idea about reframing dense prediction as retrieval runs this Skill to get a validated story_proposal.md plus a deduplicated set of real supporting papers, ready to hand to the ts-paper pipeline. ## Quick Start Turn my research idea about using retrieval-augmented memory for long-horizon agents into a grounded research story proposal with supporting papers.
