deep-research

Investigates one research direction with a supportive or adversarial stance. Works in research/analysis passes — searches, analyzes what came back, names the remaining gaps, then searches only to close them. Writes a structured findings file. Spawned by the deep-research skill.

Scores a round's research findings, verifies quantitative claims against sources, folds findings into the deliverable documents, and curates the roadmap. Spawned by the deep-research skill once per round.

Reviews a finished deep-research working directory. Use when the user wants a debrief of a research run — what was found, what scored well, what failed, and the open questions. Reads the working directory; runs no new research.

Multi-agent iterative research. Use when the user has a research goal — a question to answer, a topic to investigate, a due-diligence subject — and wants a thorough, sourced, audience-targeted document. Runs an autonomous loop of parallel research workers and a judge.