Discovering Intent

Discovering Intent Skill

Prevent building the wrong thing. Discover user's intent (WHY), validate assumptions, and agree on approach (WHAT) before taking action.

What This Skill Does

• Discovers INTENT behind surface requests (WHY they want it)
• Surfaces and validates AI's assumptions before acting
• Explores solution options informed by intent
• Reaches mutual agreement on both problem and solution
• Works for any context: software, documents, brainstorming, automation

What This Skill Does NOT Do

• Follow rigid scripts
• Skip to implementation without understanding
• Accept surface requests without exploring intent
• Make assumptions without validating them

Core Problem This Skill Solves

AI builds the wrong thing because it:

1. Takes surface requests literally without understanding intent
2. Makes hidden assumptions it never validates
3. Proceeds without confirming alignment

This skill ensures:

1. Intent (WHY) is discovered, not just request (WHAT)
2. Assumptions are surfaced and validated
3. Both problem and solution are agreed before proceeding

The WHY + WHAT Model

Surface WHAT → Discover WHY → Surface Assumptions →
Informed WHAT → Agree on Both → Proceed

When to Trigger

Don't over-trigger: Simple, clear requests don't need full discovery.

Discovery Flow

Before Starting

Gather available context before asking questions:

1. Surface WHAT

Capture the initial request clearly.

"Let me make sure I understand - you're asking for [X]?"

2. Discover WHY

This is the critical step most AI skips.

Go beyond WHAT to understand WHY:

Techniques for WHY:

Laddering - Dig into abstract goals:

"Dark mode" → "Why?" → "Eye strain" → "Why an issue?" → "Night shift workers"

5 Whys - Uncover root need:

"Export feature" → Why? → "Share reports" → Why? → "Stakeholder reviews" → Root need

Structuring Clarifications:

When presenting multiple questions, distinguish must-know from nice-to-know:

## Required Clarifications
1. [Critical question - blocks progress]
2. [Critical question - affects core approach]

## Optional Clarifications (if relevant)
3. [Nice-to-know - can assume reasonable default]

Note: Keep to 1-4 questions per round. Build on answers.

3. Surface Assumptions

This prevents "builds wrong thing."

AI always makes assumptions. Surface them explicitly:

"I'm assuming:
- This is for [platform/context]
- Users are [type]
- We need to support [X] but not [Y]
- [Other assumption]

Are these correct?"

Common hidden assumptions:

• Technology/platform
• User expertise level
• Scale/performance needs
• Integration requirements
• What's in vs out of scope

4. Informed WHAT

Now that WHY is clear, explore WHAT options:

"Given that you need [WHY], we could:
1. [Option A] - [trade-off]
2. [Option B] - [trade-off]
3. [Option C] - [trade-off]

Which fits your intent best?"

Key: Options should address the WHY, not just the surface WHAT.

5. Agree on Both

Confirm understanding of BOTH problem and solution:

## Understanding

**Problem (WHY)**: [What we're solving and why it matters]

**Solution (WHAT)**: [What we'll build/do]

**Key decisions**:
- [Decision 1]
- [Decision 2]

**Not included**: [Explicit scope boundaries]

Does this capture it correctly?

Only proceed after explicit confirmation.

Depth Check

How do you know understanding is deep enough?

Surface Understanding (NOT enough)

• Can repeat what user asked for
• Know the immediate request
• Haven't explored why

Deep Understanding (ENOUGH)

• Know WHY they want it, not just WHAT
• Know what problem it solves
• Assumptions are surfaced and validated
• Know who benefits and how
• Know what's explicitly out of scope
• Could explain it to someone else accurately
• User confirmed understanding is correct

Test: If you proceeded now and built something, would user say "yes, that's what I meant" or "no, you misunderstood"?

Assumption Categories

Surface assumptions in these areas:

Anti-Patterns

Tool Adaptation

Use whatever tools are available:

The skill describes WHAT to do. The agent uses available tools.

Output: Understanding Summary

Match formality to situation:

Quick (simple requests):

Got it: [WHAT] to solve [WHY]
Proceeding with [approach]. Confirm?

Standard (most cases):

## Understanding

**Problem (WHY)**: [Intent and problem being solved]
**Solution (WHAT)**: [What we'll do]
**Key points**: [Important details]
**Not included**: [Scope boundaries]

Ready to proceed?

Detailed (complex work): See references/summary-templates.md

Quick Reference

1. Surface WHAT → "You're asking for X?"
2. Discover WHY → "What problem does this solve?"
3. Surface assumptions → "I'm assuming A, B, C - correct?"
4. Informed WHAT → "Given WHY, we could do X, Y, or Z"
5. Confirm both → "So we're solving [WHY] by doing [WHAT]?"
6. Proceed → Only after explicit confirmation

Reference Files