refine_requirement

refine_requirement is the step before get_stack when the user's prompt is vague. It returns a structured sub_needs array that you pass unchanged into the stack builder.

Input

prompt················································································stringrequired

The user's raw use-case.

classification················································································"tool_discovery" | "stack_building" | "tool_comparison" | "tool_configuration"optional

Hint from classify_prompt, if already run.

project_context················································································objectoptional

Optional constraints from the current project (language, framework, existing tools).

Example

{
  "prompt": "I'm building a SaaS with auth and billing",
  "classification": "stack_building",
  "project_context": { "language": "TypeScript" }
}

{
  "decomposition_prompt": "A full-stack SaaS with user auth and subscription billing.",
  "sub_needs": [
    { "sub_need_type": "framework",
      "keyword_sentence": "react, full-stack, ssr, typescript" },
    { "sub_need_type": "database",
      "keyword_sentence": "sql, postgres, type-safe, migrations" },
    { "sub_need_type": "auth",
      "keyword_sentence": "oauth, session, email-password, passkeys" },
    { "sub_need_type": "payment",
      "keyword_sentence": "subscriptions, webhooks, checkout" }
  ]
}

Pipeline

Agent pattern: classify_prompt → refine_requirement → get_stack. The first step is optional; the latter two are the core loop.

Related

• get_stack — the consumer of the sub_needs returned here.
• Keyword-Sentence Retrieval — why the output is structured this way.