trellis-extract-prd

Provide the requirement ID or Task ID plus the path to the source document, and the Skill locates the relevant sections, extracts them verbatim, and writes a structured prd.md into the task directory. If a version planning file exists, the scope is constrained to that task candidate's boundaries.

When a task/wave planning document exists, the PRD scope must stay within the corresponding task candidate's defined range, and batch creation follows the documented task creation order. Slugs include wave and task sequence keys like w01-t01 so tasks sort consistently.

Yes, batch generation is supported, but it must follow the task creation order defined in the version planning document, keeping tasks of the same wave consecutive. If no order is defined, the Skill asks you to confirm one rather than guessing.

The Skill falls back to extracting directly from the source requirements document while still enforcing task cohesion. It will not fabricate wave IDs or Task IDs, and wave fields are marked N/A in the PRD and task.json.

User-visible text such as button labels, toasts, and placeholders must be copied character-for-character so implementation matches the specification exactly. Rewriting or summarizing UI copy is explicitly listed as an anti-pattern.

The Skill cannot invent requirements: every statement must trace to the source document or a confirmed version plan. When requirements are vague or no document exists, a conversational discovery skill such as trellis-brainstorm is the recommended alternative.