User Modeling
Build just enough understanding of your users to make better product decisions.
Why This Exists
Creates behavior-based user models that reveal what users need and how they'll behave, not marketing personas with stock photos.
Input Requirements
This skill works best with:
problem-framingoutput (problem statement, target user, JTBD)- Any existing research (interviews, surveys, support tickets, Reddit threads, reviews)
Can also work from assumptions if no research exists—but flags that these need validation.
Workflow
Step 1: Gather Context
Ingest upstream artifacts or ask:
- Who are you building this for?
- What do you know about them already?
- Have you talked to any potential users?
- Any data sources—reviews, forums, support tickets?
Step 2: Identify User Segments
Look for meaningful differences in:
- Goals — What are they trying to accomplish?
- Context — When/where do they encounter the problem?
- Constraints — What limits their options?
- Skill level — How sophisticated are they?
- Frequency — How often do they face this problem?
Not every difference matters. Focus on differences that change *what you'd build*.
Step 3: Build Personas
For each meaningful segment, create a lightweight persona. Limit to 2-3 personas max—more than that dilutes focus.
Step 4: Define Scenarios
For each persona, define 2-3 concrete scenarios where they'd use the product. These become the basis for user stories and flows.
Step 5: Identify Insights
Surface patterns that inform product decisions:
- What do all personas have in common?
- Where do they diverge?
- What would you build differently for each?
Automatically save the output to design/02-user-modeling.md using the Write tool while presenting it to the user.
Output Format
# User Modeling: [Project Name]
## Context
[Brief summary of the problem space and what we know]
**Research basis:**
- [Source 1: what it told us]
- [Source 2: what it told us]
- [Or: "Based on assumptions—needs validation"]
---
## Personas
### Persona 1: [Name/Label]
*[One-line description of who they are]*
**Goals:**
- [Primary goal]
- [Secondary goal]
**Context:**
- [When they encounter the problem]
- [Where they encounter it]
- [What else is going on]
**Pain points:**
- [Frustration 1]
- [Frustration 2]
**Current behavior:**
- [How they solve this today]
- [Tools they use]
- [Workarounds they've developed]
**Constraints:**
- [Time/budget/skill/access limitations]
**What success looks like:**
- [How they'd know the problem is solved]
**Quote:** *"[Something they might say that captures their mindset]"*
---
### Persona 2: [Name/Label]
*[One-line description]*
[Same structure]
---
### Persona 3: [Name/Label]
*[One-line description]*
[Same structure]
---
## Scenarios
### Persona 1 Scenarios
**Scenario 1.1: [Name]**
- **Situation:** [Context—what's happening]
- **Trigger:** [What prompts them to act]
- **Goal:** [What they're trying to accomplish]
- **Current approach:** [How they handle it today]
- **Frustration:** [What's broken about current approach]
**Scenario 1.2: [Name]**
[Same structure]
### Persona 2 Scenarios
**Scenario 2.1: [Name]**
[Same structure]
---
## Key Insights
### Commonalities
[What all personas share—these are table-stakes features]
- [Insight 1]
- [Insight 2]
### Divergences
[Where personas differ—these inform prioritization]
- [Persona 1] needs [X], while [Persona 2] needs [Y]
- [Persona 1] is [context], while [Persona 2] is [different context]
### Design Implications
[How this should influence what you build]
- [Implication 1]
- [Implication 2]
- [Implication 3]
---
## Validation Needed
[What assumptions need testing]
- [ ] [Assumption to validate]
- [ ] [Assumption to validate]Adaptation Guidelines
Minimal (single obvious user type):
- One persona, 2-3 scenarios
- Skip Divergences section
- 1 page total
Standard (2-3 user types):
- Full structure as shown
- 2-3 pages total
Research-heavy (actual user data):
- Include research summary
- Add quotes from real users
- Link to source data in appendix
What Makes a Good Persona
Good persona:
- Defined by *goals and behaviors*, not demographics
- Reveals something that changes what you'd build
- Based on patterns, not individuals
- Specific enough to make decisions against
Bad persona:
- Stock photo + age + job title + hobbies
- So generic it could be anyone
- Based on one interview or pure assumption
- Doesn't inform any product decisions
Anti-Patterns to Avoid
- The Kitchen Sink — Don't add demographics unless they matter
- The Clone Army — If personas don't differ meaningfully, merge them
- The Wishful Thinker — Model who users *are*, not who you wish they were
- The Edge Case Collector — Focus on primary users, not every possible user
Handoff
After presenting the personas, ask:
"Want to move to/solution-scopingto prioritize features, or straight to/prd-generation?"
Note: File is automatically saved to design/02-user-modeling.md for context preservation.