discover-intent

Discover Intent

Collaboratively refine a rough idea into a clear direction through structured dialogue. This skill asks questions one at a time, explores alternatives, surfaces constraints, and arrives at a shared understanding of what to build and why.

Do NOT begin any design work, write any spec, create any architecture document, or invoke the shape-design skill until the user has explicitly approved the direction. "Sounds good" or "let's go with that" count as approval. Silence or ambiguity does not.

Process Flow

digraph discover_intent {
    "Read project context" [shape=box];
    "Assess scope" [shape=diamond];
    "Flag decomposition needed" [shape=box];
    "Ask clarifying question" [shape=box];
    "More questions needed?" [shape=diamond];
    "Summarize direction" [shape=box];
    "User approves?" [shape=diamond];
    "Invoke shape-design" [shape=doublecircle];

    "Read project context" -> "Assess scope";
    "Assess scope" -> "Flag decomposition needed" [label="too large"];
    "Assess scope" -> "Ask clarifying question" [label="right-sized"];
    "Flag decomposition needed" -> "Ask clarifying question";
    "Ask clarifying question" -> "More questions needed?";
    "More questions needed?" -> "Ask clarifying question" [label="yes"];
    "More questions needed?" -> "Summarize direction" [label="no"];
    "Summarize direction" -> "User approves?";
    "User approves?" -> "Ask clarifying question" [label="no, refine"];
    "User approves?" -> "Invoke shape-design" [label="yes"];
}

Checklist

Complete these steps in order:

1. Read project context -- check existing files, docs, recent git history to understand the current state
   • If the work involves visual or spatial decisions (UI layout, component structure, data flow diagrams), offer to generate a visual companion to support the conversation.
2. Assess scope -- if the request describes multiple independent systems, flag immediately. Help decompose before refining details. Each sub-project gets its own discovery > design > plan > execute cycle.
3. Ask clarifying questions -- one question per message. Prefer multiple choice when possible. Focus on:
   • What problem does this solve? Who benefits?
   • What are the boundaries? What is explicitly NOT in scope?
   • What does success look like? How will you know it works?
   • Are there existing patterns in the codebase to follow or avoid?
   • What constraints exist (performance, compatibility, timeline)?
4. Summarize the direction -- present a concise summary of:
   • The problem being solved
   • The approach being taken
   • What is in scope and what is not
   • Key constraints and success criteria
5. Get explicit approval -- ask the user to confirm the direction before proceeding

Anti-Patterns

"Let me just start coding, it's obvious" Nothing is obvious. The gap between what the user said and what they meant is where wasted work lives. Even a 30-second discovery conversation catches misalignment.

"I'll ask all my questions at once" Batched questions get batched (shallow) answers. One question at a time lets each answer inform the next question. The conversation is the discovery mechanism.

"The user seems impatient, I'll skip ahead" Impatience is not approval. A rushed discovery phase creates a rushed design phase creates rework in execution. Hold the line.

"Text-only discovery for visual work" Text-only discovery for UI features misses layout, spacing, and flow concerns. When the work is visual, offer a diagram or mockup to surface spatial issues that text alone misses.

Evidence Requirements

• The user has explicitly approved the direction (their approval message is the evidence)
• A clear problem statement, approach, scope boundary, and success criteria exist in the conversation

Transition

When the user approves the direction, invoke the shape-design skill to create a spec document.