dsh-deep-research

dsh-deep-research addresses the problem that many “deep research” flows are just fixed prompt pipelines with little self-correction. It builds the process on the official DSH workflow engine, turning planning, retrieval, gap-filling, and synthesis into an adaptive loop. A planner first defines the decision space, sub-questions, and acceptance criteria, while research agents search and fetch sources in parallel and track evidence as confirmed, uncertain, or missing.

The plugin focuses on information gain rather than endless searching. Each round checks whether new evidence actually reduces uncertainty, stops when marginal gain disappears, and automatically dispatches follow-up research for high-priority gaps. A synthesis agent then compresses the evidence into a decision-ready report, while optional adversarial review can spot-check citations, audit coverage, and flag contradictions or overconfidence.

It is a good fit for users who need multi-source research, competitive or solution comparisons, literature collection, or decision support inside DSH. It is also useful for developers who want to orchestrate research sub-agents with workflow-level control, model tiering, and concurrency limits.