Design Discovery Ideation

Style: Mode-B-lite. Divergent product/UI exploration before decisions lock; no production code.

Purpose

Help the Product Owner turn "show options" / "try approaches" requests into clear, bounded alternatives before implementation. The skill separates fixed constraints from open design dimensions, presents 2-4 meaningfully different directions, and feeds the chosen direction into spec.md or decisions.md.

When to invoke

• Operator asks for options, variations, alternatives, ideas, directions, or other possible shapes.
• A UI, workflow, prompt, dashboard, or product behavior needs divergent exploration before lock.
• requirements-specification finds unclear product framing or user experience scope.
• implementation-planning needs real alternatives for a decision instead of a single recommendation.
• Architect escalation is active and the design space spans multiple files, flows, or constraints.

Do not invoke when

• The task is a deterministic bug fix with known expected behavior.
• A relevant decision is already LOCKED; AI Developer must implement or stop and surface a conflict.
• Operator asks for direct implementation, not option discovery.
• The only requested change is a narrow styling tweak with no product/design ambiguity.
• Required context is unavailable and cannot be inspected read-only.

Required inputs

Procedure

1. Confirm role boundary.
   • Product Owner owns ideation and option recommendation.
   • Architect escalation may deepen read-only investigation.
   • AI Developer may only flag underspecified or conflicting design and return to PO.
2. Understand the current surface:
   • User goal and workflow position.
   • Information collected or shown.
   • Interaction states and failure states.
   • Existing structure and neighboring screens/prompts.
   • Constraints from spec, constitution, policies, accessibility, platform, and locked decisions.
   • For Fusebase Apps UI/product work, supporting CLI guidance from app-ui-design, app-dev-practices, or another relevant CLI skill named in docs/fusebase-cli-edition.md.
3. Separate design space:
   • Fixed: cannot change without violating scope, constraints, or locked decisions.
   • Assumed: current shape that may be changed if operator chooses.
   • Open: dimensions worth exploring.
4. Pick 2-3 exploration dimensions that matter for this ticket. Examples:
   • Structure / flow order.
   • Content hierarchy / semantic framing.
   • Interaction behavior / state progression.
   • Visual tone / density when UI-facing.
   • Technical approach only when it changes product trade-offs.
5. Produce 2-4 options. Each option must have:
   • A short name.
   • Design hypothesis.
   • What changes.
   • What stays fixed.
   • Trade-off.
   • Verification or smoke implication.
6. For frontend/UI tickets, also draft an implementation-ready design brief:
   • Product identity and target user.
   • Selected option or option set.
   • Affected routes, pages, workflows, prompts, or components.
   • Data types, fields, states, and real interactions the UI must support.
   • API/helper surfaces by name/signature when known.
   • Applicable stack conventions or project-local frontend rules by path when known.
   • Stable test selector strategy for interactive and meaningful dynamic elements.
   • Trust-critical flows that must work with real behavior, not placeholders.
   • Brand/source inputs, only when provided or gathered from approved read-only sources.
   • Explicit non-goals: routes, entities, workflows, or visual constraints the implementer must not invent.
7. Recommend one option in chat text with (Recommended) marked. Do not use popup / clickable menu tools.
8. Wait for operator direction:
   • If approved, write the selected direction into spec.md, clarify-conversation.md, or decisions.md as appropriate.
   • If redirected, revise the options without preserving stale discarded content in authoritative artifacts.
   • If unresolved, leave status DRAFT / PENDING and name the blocking question.
9. If delegated investigation is useful, invoke task-delegation with a read-only/doc-only brief. The brief must include the selected exploration dimensions and forbid code edits unless the session is AI Developer and the task is already locked.

Output artifacts

Failure cases

Escalation path

• Cross-cutting product/workflow change -> workflows/architect-escalation.md.
• Need implementation sequencing -> flow-skills/implementation-planning/SKILL.md.
• Need smoke criteria for a UI/operator-facing outcome -> flow-skills/smoke-testing/SKILL.md.
• Need parallel read-only analysis -> flow-skills/task-delegation/SKILL.md.

Prototype before build

Use when a UI / screen / app is about to be implemented and the operator would benefit from seeing it before code is written — cheap visual agreement avoids expensive rebuilds.

Pre-build prototyping is a generic technique (no project artifact needed). Keep prototypes disposable; the real build follows the locked direction. ASCII mockups belong in chat (Mode A), never embedded in Mode-B artifact files (FR-08).

Anti-patterns

• Do not write production code from this skill.
• Do not treat an HTML prototype as the real implementation — it is throwaway.
• Do not present only one option unless the operator explicitly narrowed the choice.
• Do not produce alternatives that differ only by color, wording, or spacing unless the operator asked for that narrow axis.
• Do not prescribe exact UI structure, colors, typography, spacing, or component patterns unless the operator explicitly made them constraints.
• Do not let UI implementation briefs omit real data, states, and primary interactions.
• Do not bury the recommendation in prose; mark it clearly.
• Do not lock a direction without operator approval.
• Do not use popup / clickable menu tools; options must be written in chat text per FR-19.
• Do not let AI Developer use ideation to bypass locked decisions.

Clean-room note

Original Fusebase Flow content. Derived from operator-provided capability requirements and generalized into repo-local workflow discipline; no third-party code, prompts, skill files, or hook scripts are copied. See docs/source-map.md.