What problem does it solve? When a research pipeline stalls—no performance gain, an exhausted candidate pool, or an unexplained failure signature—you need targeted literature evidence rather than keyword-matched paper lists. This Skill builds a bounded, failure-focused corpus and synthesizes it into an actionable literature map. ## Core Features & Use Cases - Failure-Driven Search: Starts from the measured baseline failure, method, metric, and failure signature instead of generic task keywords, expanding through citations, references, and official code. - Structured Contribution Extraction: For each deep-read paper, records stable IDs (DOI/arXiv), mechanisms, assumptions, ablation evidence, limitations, and plausible adaptation points in the current method. - Revisioned Synthesis: Writes research/LITERATURE.md grouping papers by mechanism, comparing assumptions against research/BASELINE.md, and separating author claims from transfer inference. - Use Case: After a screening experiment shows no gain, trigger this Skill to search for mechanisms addressing the observed failure, deep-read the four closest papers, and produce a synthesis that feeds the mapping phase with concrete hypotheses. ## Quick Start Search the literature for mechanisms that address the current baseline failure signature and write the synthesis to research/LITERATURE.md.
