SuperAPI AI Directory

to-prd

General & other

View repository

Updated 10-16 09:47

About this skill

Where this sits

PORTFOLIO — many features        ideate  or  chat-to-approach
                                        ↓
                            ./docs/approach.md  →  ./ideas/roadmap.html
                                                        ↓  pick ONE card
FEATURE — one feature per run    to-prd → to-scaffold → to-tickets → ticket-tracker
                                 ^^^^^^ you are here
                                 non-code work: to-prd → to-tickets

Portfolio → feature is a fan-out, not a step. A roadmap card is a feature. Six cards means six PRDs, never one covering a whole roadmap or approach doc.

Never send the user backwards. A missing upstream artifact is never a reason to stop.

What a PRD is

Every PRD has a core: a one-line problem, journeys, and delivery slices. A small change is the core and nothing else. Journeys and slices are defined in references/project-doc.md; read it before writing.

Build order, tracer bullets, module boundaries and per-file design stay out: they are to-scaffold's and to-tickets' job.

Priors

Before asking a question, skip it if one of these already answers it:

  1. The conversation.
  2. On-disk priors — ./docs/approach.md and ./ideas/reports/*-ideate.md (use artifact-scan to locate them if installed, else ls). Confirm and grill deltas only.

Priors are accelerants, not prerequisites: with none, run the full interview. They are portfolio level; if one holds several features and the target isn't obvious, ask which one this PRD is for.

Process

Skip steps that aren't necessary.

1. Get the problem in the user's words

Ask for a long, detailed description of the problem and any ideas for solutions.

2. Explore the repo

Verify their assertions and learn the current state. Note what already half-exists, so the journeys build on it rather than reading as all-net-new.

3. Grill for scope and behaviour

Run the grilling skill; it owns the loop. Focus on who does what and what they see, not build order.

4. Write the journeys

For each way a user gets value, draft a journey in the reference's format. Check the set with the user: does it cover the problem, and does every journey end somewhere you could point at? If a prototype would help, the prototype skill builds one.

5. Cut delivery slices

Ask for the smallest release that delivers the core outcome, then what follows. Sequence by user value, not technical dependency.

6. Offer the optional layer

For a bigger feature, offer to add why, goal, concepts, out of scope and open questions. Skip it for a small change.

The format of this layer is the user's call and not fixed by this skill. If they want it, ask: "How should the optional layer look: prose sections, a table, bullets, or something else?" Use their answer; do not choose for them.

7. Save it

  1. Create docs/prds/ if missing. Write the PRD to docs/prds/<feature>.md, kebab-cased.
  2. Show it to the user and confirm.
  3. Print the file path. to-scaffold and to-tickets read it from there.

Filing a GitHub issue is opt-in, only when the user asks, and from the file so the copies don't drift:

gh issue create --title "[PRD] <Feature Name>" --body-file docs/prds/<feature>.md

Then hand off: /to-scaffold for code work; /to-tickets straight from the PRD for non-code work.

When finished, ask: 'Would you like to log feedback? (yes/no)'. If yes, invoke skill-feedback-collector passing this skill's name and path.