lite-prd

Turn a vague feature ask into a lightweight PRD by interviewing the user with AskUserQuestion, traversing the design tree (branch deeper when an answer opens new territory), persisting every round, and writing lite-prd.md.

What this does

1. Reads the vague ask and picks a slug + output folder.
2. Interviews the user in adaptive batches via AskUserQuestion (2–4 grouped questions per round), up to ~50 questions total.
3. Traverses the design tree — when an answer materially changes scope or leaves a real ambiguity, branch deeper into that thread before moving on.
4. Persists every question and answer to a Q&A log in the repo as it goes.
5. Synthesizes the answers into lite-prd.md in the pasted format (minus Monetization).

Flags

• --deep: exhaustive branching — pursue every plausible sub-thread until the tree is dry or the ~50-question budget is spent. Default (no flag) is model-judged, biased to fewer: only branch when an answer materially changes scope or leaves a real ambiguity, and prefer finishing over exhausting the budget.

Step 1 — Set up the workspace

• Derive a short kebab-case <slug> from the feature ask (e.g. "team invites" → team-invites).
• Create the folder docs/lite-prd/<slug>/ in the repo the skill was invoked in. If it already exists, reuse it and continue an in-progress run rather than clobbering.
• Create/open the Q&A log at docs/lite-prd/<slug>/qa-log.md. Seed it with the raw ask.

Step 2 — Interview by traversing the design tree

Read references/traversal-protocol.md and follow it exactly. It defines: how to seed the root questions, how to batch with AskUserQuestion, when to branch deeper vs. move on (and how --deep changes that), the ~50-question budget, and the persistence rule (write each round to qa-log.md immediately after the user answers, before asking the next batch).

Every round: ask → read answers → append the round to qa-log.md → decide whether to branch or advance.

Step 3 — Synthesize the PRD

When the tree is covered (or the budget is spent), read references/prd-template.md and fill every section from the Q&A log. Write the result to docs/lite-prd/<slug>/lite-prd.md.

Rules:

• Use ONLY what the user actually told you plus low-risk, clearly-labeled inferences. Never invent requirements the interview didn't support.
• Requirements go in as user stories with - [ ] checkboxes, exactly as the template shows.
• Omit the Monetization section entirely (per the format).
• Anything still genuinely unresolved goes under Critical Questions or Clarifications — do not paper over gaps.

Step 4 — Report

Tell the user the two paths written (qa-log.md and lite-prd.md), the question count, and list anything parked under Critical Questions so they know what's still open.

Honesty rule

If the user's answers leave a section unsupported, write "TBD — see Critical Questions" rather than fabricating plausible-sounding content. A confidently-wrong PRD is worse than an honestly-incomplete one.