Moradin Define

moradin:define

Turn "I have an idea" into a plan someone can build from — in 10–15 minutes at light level. Roughly 43% of failed products die because nobody needed them, so this stage forces who has this problem before any feature talk.

When invoked

/moradin:define (dispatcher routes here) or directly with the idea: /moradin:define "an app that ..."

Requires .forge/profile.md (run /moradin:forge first if missing). Read profile.md and, if it exists, the current plan.md (amendment mode — supersede only the affected sections).

The one interaction rule

You draft, the operator corrects. Never ask them to produce an artifact from a blank page ("write your user stories" is banned). For every step: propose a concrete draft from what they've told you, then ask what's wrong with it. Timebox the whole conversation — say up front "this takes about 15 minutes" and honor it. Good-enough v0 answers are explicitly fine; say so when they stall.

The interview — light level (all profiles do these 6)

Work through in order. One step at a time, conversationally.

1. Who + problem. Draft: "[specific person] struggles to [specific thing]." Reject "everyone" and "save time" phrasings — push for a person type they could actually name and a struggle they've actually seen. This is the step most failed products skipped.
2. Job story. Draft: "When [situation], I want to [motivation], so I can [outcome]." One sentence. They edit.
3. Riskiest assumption. Ask: "What has to be true for anyone to use this? What's the ONE thing v1 must prove?" Also ask: "How do you know this problem exists?" If the honest answer is "I don't" — note it; milestone 0 becomes a validation test (the measure stage will set it up).
4. Walking skeleton. Ask: "Walk me through one successful use, step by step, as the user." Each step becomes a story. Cap at 7; if they list more, the extras go to the backlog.
5. Appetite + the cut. Ask: "How much time before something is usable — a weekend? Two weeks?" Then go step by step: in, or not-now, against that appetite. Binary only — no maybe-tiers. Everything cut is parked in the backlog with its number, not deleted; say that out loud, it makes cutting painless.
6. Premortem. Ask: "It's three weeks from now and you abandoned this. Why?" Draft one mitigation for the top answer. Milestone 1 = the skeleton works end-to-end.

Full level adds (medium picks what's useful)

• 2–3 named user types; pick the primary.
• Directional outcome + 2–3 opportunities before discussing solutions (catches "I built my idea, then looked for a problem").
• Acceptance criteria per skeleton step ("how will you know this step works?").
• A non-goals section.
• Full premortem: 3–5 failure reasons, each mapped to a mitigation or a milestone.
• 2–3 milestones, each with its own appetite and a demoable outcome.

Never, at any level: RICE scores, opportunity-tree upkeep, Should/Could tiers, effort estimates.

Write the outputs

1. .forge/plan.md from templates/forge/plan.md — hard cap: 2 pages light / 5 pages full. If the draft runs over, cut prose, not decisions. Every sentence should encode something the operator chose.
2. Seed .forge/backlog.md with every parked item (status deferred, one-line reason).
3. Update .forge/journal.md: stage done, one-line summary, next action — "Run /moradin:measure to set the success numbers before we research tools."

Then read the plan back in three sentences, plain language, and ask: "Is this the thing you meant?"