Stats
Actions
Tags
From my-skills
Conducts discovery conversations to uncover user intent, validate assumptions, and agree on approach before acting. Use for recommendations, brainstorming, clarifications, or ambiguous requests.
How this skill is triggered — by the user, by Claude, or both
Slash command
/my-skills:discovering-intentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Prevent building the wrong thing. Discover user's intent (WHY), validate assumptions, and agree on approach (WHAT) before taking action.
Prevent building the wrong thing. Discover user's intent (WHY), validate assumptions, and agree on approach (WHAT) before taking action.
AI builds the wrong thing because it:
This skill ensures:
Surface WHAT → Discover WHY → Surface Assumptions →
Informed WHAT → Agree on Both → Proceed
| Phase | Purpose | Example |
|---|---|---|
| Surface WHAT | Capture initial request | "Add dark mode" |
| Discover WHY | Uncover intent/problem | "Eye strain for night workers" |
| Surface Assumptions | Expose AI's hidden assumptions | "Assuming web app, not mobile" |
| Informed WHAT | Solution options based on WHY | "Dark mode + auto-brightness + schedule" |
| Agree on Both | Confirm problem AND solution | "Solving eye strain via dark mode with auto-switch" |
| Trigger | Example |
|---|---|
| Clarification requested | "Let's clarify", "Help me think through this" |
| Request could be misunderstood | Ambiguous, complex, or multi-part requests |
| Recommendations needed | "What should I use for..." |
| Brainstorming | "Help me think through..." |
| High-stakes work | Where wrong output wastes significant effort |
Don't over-trigger: Simple, clear requests don't need full discovery.
Gather available context before asking questions:
| Source | Gather |
|---|---|
| Conversation | User's stated request, prior context |
| Available Context | Information already shared in session |
| Skill References | Question patterns from references/ |
Capture the initial request clearly.
"Let me make sure I understand - you're asking for [X]?"
This is the critical step most AI skips.
Go beyond WHAT to understand WHY:
| Ask | To Discover |
|---|---|
| "What problem does this solve?" | The real need |
| "Why now?" | Urgency and context |
| "What happens if we don't do this?" | Stakes and priority |
| "Who benefits and how?" | Users and value |
| "What led to this request?" | Background and triggers |
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.
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:
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.
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.
How do you know understanding is deep enough?
Test: If you proceeded now and built something, would user say "yes, that's what I meant" or "no, you misunderstood"?
Surface assumptions in these areas:
| Category | Example Assumptions |
|---|---|
| Context | Platform, environment, existing systems |
| Users | Who they are, expertise level, needs |
| Scale | Volume, performance requirements |
| Scope | What's included vs excluded |
| Quality | Standards, constraints, requirements |
| Timeline | Urgency, phases, dependencies |
| Anti-Pattern | What Happens | Fix |
|---|---|---|
| Skip WHY | Build wrong solution | Always ask why before how |
| Hidden assumptions | Surprise misalignment | Surface and validate explicitly |
| Accept surface request | Miss real need | Dig deeper with laddering/5 whys |
| Proceed without confirm | Waste effort | Get explicit "yes, proceed" |
| Over-question simple requests | Annoy user | Match depth to complexity |
Use whatever tools are available:
| Goal | Approach |
|---|---|
| Ask questions | Interactive tools if available, otherwise conversation |
| Research context | Web search if needed and available |
| Present options | Structured choices if available |
The skill describes WHAT to do. The agent uses available tools.
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
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
| File | Purpose |
|---|---|
references/question-patterns.md | Techniques for discovering WHY and surfacing assumptions |
references/anti-patterns.md | Common mistakes that lead to building wrong thing |
references/summary-templates.md | Output formats for different situations |
npx claudepluginhub rehan-ul-haq/my-skills --plugin my-skillsConducts a one-question-at-a-time interview to surface the user's actual intent when requests are underspecified. Use before planning or coding when you lack clarity on who, why, or what success looks like.
Interviews the user one question at a time to extract true intent from underspecified requests, achieving ~95% confidence before any plan or code exists.
Orchestrates structured Socratic interviews to clarify ambiguous requirements using a dedicated interviewer agent and Ambiguity Score. Useful for vague ideas; invoke via /deep-interview or keywords.