maga2010org/hackathon-run/idea-clarify

idea-clarify

When a team shows up with an empty repo and a sentence, the problem is not "we have no features". It's "we have no shared picture of what we're building." This skill forces that picture into the open.

Input contract

Required:

• user_brief: the raw one-paragraph description from the user
• audience: who the demo is for (judges, customers, internal team)

Optional:

• time_remaining_minutes: integer >= 0 (default 36 * 60)
• prior_artifacts: list of any pre-existing notes, sketches, links

Execution

1. Surface the problem

Ask the user the four canonical questions:

Refuse to proceed until each answer is at least one concrete sentence.

2. Detect contradictions

If two answers contradict (e.g. "we want ML" + "no ML allowed"), surface the contradiction. Do not silently pick one. The agent must resolve it before scope-knife runs.

3. Emit a one-page brief

Write .hackathon/artifacts/idea-brief.md with:

• the four answers verbatim
• any contradictions + how the agent resolved them
• a one-line demo_goal candidate
• a one-line mvp_axis (the smallest axis along which the MVP can grow)

Output contract

Files written:

• .hackathon/artifacts/idea-brief.md (human-readable, one page)

This skill does NOT write to .hackathon/state/*.json directly. It is a pre-scope-knife step; scope-knife reads the brief and writes plan.json from it.

Acceptance criteria

• [ ] The four canonical questions are asked exactly once.
• [ ] Contradictions are surfaced, not silently resolved.
• [ ] demo_goal is a single sentence (max 20 words).
• [ ] mvp_axis is a single noun phrase.
• [ ] .hackathon/artifacts/idea-brief.md exists and is <= 1 page.

Failure modes

Trigger phrases (for agent intent matching)

• "we have an empty repo"
• "what should we build"
• "the brief is one paragraph"
• "we have an idea but no plan"
• "idea is too vague"