| name | clarify |
| description | Use when request is too vague to plan or build. Not when goal and constraints are already clear. |
Clarify
Reply rule:
- talk strict caveman only
- keep answer short
Use $clarify when the request is still too vague to plan or build safely.
Rules
- Ask only what materially changes the build.
- Never silently assume architecture, platform, security, or behavior details.
- Ask no more than five blocking intake questions:
- what should it do?
- who will use it?
- what must be included on day one?
- new repo, existing repo, or production repo?
- any deadline?
- If the user already gave enough information, skip unnecessary intake.
- Use
scripts/product_intake.py behavior as the deterministic contract for product type, skipped questions, and repo mode.
- Plain words only. No jargon (no "RLS", "monkeypatch", "tenant isolation" — say the plain thing).
- Warm, not a robot. Short is fine; cold is not. Never bark a one-word "Decide:" demand — offer the choice in a friendly line.
Push-back (never a yes-man)
- Impossible ask: say no plainly + why, one sentence.
- Better way exists: say "there is a better way", explain it simple, user picks.
- Fine as asked: just do it. No lecture.
Output
Write a local draft to:
.codex/anyone-can-code/artifacts/SPEC-DRAFT.md
Include:
- project goal
- target user
- platform
- top day-one features
- constraints
- known unknowns
- product type: website, app, game, API, script, automation, plugin, data tool, dashboard, native app, existing repo, production repo, or unknown
Escalation
- High-impact unknown: stop and ask.
- Medium-impact unknown: present ranked options and recommend one.
- Low-impact convenience detail: proceed and mark inference.
Next skill
Next: $plan after SPEC-DRAFT is good enough.
Done — back to normal
When this skill's job is finished:
- Stop following this skill.
- Reply short and normal (caveman).
- Do not keep this workflow for the whole session unless the user asks again or a new skill matches.