dsh-rich-questions

When an agent needs to scope a build across a dozen interacting dimensions, one or two flat questions simply do not cut it. Choosing A versus B changes which follow-up dimensions even matter, and a one-line label is not enough to make a confident choice. dsh-rich-questions lets the agent compose a directed-graph questionnaire straight from the conversation context: every option declares its own branch, carries a three-part insight (Promise, Price, Present), source citations, and a compact Mermaid diagram, all rendered in the same chat composer you are already looking at. No forms product, no account, no copy-paste.

Beyond branching paths, the plugin ships one-click Quick decision templates, per-step Bank and continue for long surveys (answers lock on commit, survive browser loss, persist across reloads), Reroll-Push-Discuss pre-flight steering, and a research-first Question Builder that studies a dozen comparable systems before locking structure, avoiding the failure mode of one giant payload that truncates on small local models. Drafts live as git-diffable files in your workspace with a revision history you can inspect. The whole plugin has zero runtime dependencies, is MIT-licensed, and survey language tracks your chat language automatically.

It is built for developers using DeepSeek Harness who need deep expectation gathering, acceptance-criteria alignment, or multi-dimension scoping. You want the agent to ask questions with paths, depth, and speed, without leaving the current conversation context or blocking the session on a pending form. Simple one-to-three question flows still go through the native tool, and this plugin leaves that path untouched.