What problem does it solve? Running literature and web research manually means searching, scoring relevance, deduplicating, and filing findings by hand. This Skill coordinates that entire research cycle conversationally while a deterministic Python capability performs the actual searching, scoring, and writing behind a human approval gate. ## Core Features & Use Cases - Scope Negotiation: Reads search-seeds.json, domains.yaml, and governance.yaml to confirm full, domain-specific, or targeted research scope with the user before anything runs. - Gated Execution: Invokes construct research run to execute the search-score-dedup pipeline, which pauses at a review gate and writes nothing until approved. - Human Review Loop: Presents the pending gate queue with titles, sources, and relevance scores, then resumes the run via construct research review with approve-all, reject-all, or per-finding decisions. - Use Case: A researcher asks "what's new in machine learning this week" — the Skill confirms scope, runs the capability, presents 12 candidate papers for approval, and only the 8 approved items become ref files and seed cards in the workspace. ## Quick Start Ask the agent to run a research cycle for your domain, then approve or reject the pending findings it presents.
