SuperAPI AI Directory

generar-prd

Writing & research

View repository

Updated 10-15 10:55

About this skill

PRD Generator (Light)

Goal

Generate a first-draft PRD that is intentionally fast and light — a starting point to review and refine, not a finished requirements process. This skill does not replace deep product discovery; it exists to unblock the next phase (design, architecture) with a working draft.

The primary outcome is a file written to disk: PRD.md at the project root, unless the user specifies another path.

Required Input

Before writing, the user must have provided (or be asked for, in a single concise question if missing):

  • The product idea, in a sentence or two.
  • Who it's for (target user).
  • The core problem it solves.

If any of these three is missing or too vague to write a meaningful PRD, ask once, concisely, before generating. Do not invent the product idea itself — secondary details (success criteria, edge cases) may be filled as best-effort suggestions the user is expected to revise.

Workflow

  1. Confirm the three required inputs (idea, target user, problem). Ask once if missing.
  2. Generate PRD.md using the template in assets/prd-template.md, filling every section. Leave no placeholder unresolved — if information is missing, write a best-effort draft and flag it inline with <!-- REVISAR: ... -->.
  3. Write the file to disk. Do not just paste the PRD into chat.
  4. Close with a short reminder: this is a fast draft (minutes) — the real work is the review that follows (ambiguities, success criteria, scope, edge cases) before using it as input for design or architecture.

PRD.md Sections

See assets/prd-template.md for the full template. Sections, in order:

  • Problema
  • Usuario objetivo
  • Objetivo / resultado esperado
  • Alcance (qué sí incluye esta versión)
  • No alcance (qué explícitamente no incluye esta versión)
  • Criterios de éxito
  • Casos borde a contemplar
  • Supuestos y riesgos abiertos

Quality Gate

Before returning, silently check:

  • No section is empty — a thin bullet beats a missing section.
  • "No alcance" contains at least one explicit exclusion, not just "todo lo no mencionado".
  • Success criteria are verifiable, not vague adjectives ("rápido", "intuitivo") without a measurable anchor.
  • Any assumption the model made stands out with <!-- REVISAR: ... --> so the human catches it during review.