What problem does it solve? Conducting a rigorous literature survey for an ML theory research direction is slow and error-prone: papers must be found across direct, foundation, and empirical scopes, deduplicated, classified by their actual theorem claims, and distilled into gap statements that downstream idea generation can trust. This Skill automates that survey and produces a single auditable LITERATURE_SURVEY.md artifact. ## Core Features & Use Cases - Structured theory search: Searches Direct Theory first per sub-target, then recursively falls back to Component Theory or Testbed Theory only on zero-hit scopes, with explicit paper-count budgets. - Analysis-target taxonomy: Organizes search around Optimization, Generalization, Expressive Power, and Implicit Bias families, with explicit Other buckets for nonstandard theorem properties. - Gap-oriented output: Writes LITERATURE_SURVEY.md with direct theory, foundation theory, empirical practice, and gap sections, where every gap states evidence, downstream relevance, and supported/uncertain status. - Use Case: Given a research direction like "generalization analysis of component A in algorithmic setting B", the Skill detects the two primary objects, searches convergence and generalization sub-targets across local PDFs and web sources, and produces a survey with per-sub-target gaps ready for perspective selection. ## Quick Start Ask the agent to run the literature survey for your research direction, for example: "Run research-lit on theoretical scaling laws of kernel regression and write LITERATURE_SURVEY.md."
