Interview
Interview user in-depth to gather requirements for a plan or objective
---
description: Interview user in-depth to gather requirements for a plan or objective
argument-hint: [topic or feature description]
allowed-tools: AskUserQuestion, Read, Glob, Grep
---
# /local:interview
Interview the user in-depth about a feature or task, then return a structured summary oriented toward creating a plan or objective.
## Usage
```bash
# With a topic hint
/local:interview authentication system
# Without arguments (will ask about the topic)
/local:interview
Agent Instructions
Execute the interview workflow in four phases: explore, interview, scope determination, and summary.
Phase 1: Explore Codebase
1.1 Parse arguments
Extract the topic from $ARGUMENTS:
- If non-empty, use it as the initial topic
- If empty, proceed to ask about the topic in Phase 2
1.2 Identify relevant code
If a topic was provided, explore the codebase before asking questions:
Search for relevant files using Glob:
- Look for files related to the topic (e.g., if topic is "authentication", search for
**/*auth*) - Look for configuration files, test files, and implementation files
- Look for files related to the topic (e.g., if topic is "authentication", search for
Search for relevant code using Grep:
- Search for key terms from the topic
- Identify main entry points, APIs, and interfaces
Read key files:
- Read 2-4 most relevant files to understand current implementation
- Note: Keep exploration focused - don't read more than necessary
1.3 Build initial context
Create a mental model of:
- What exists already (if anything)
- Where the work would likely happen
- What patterns/conventions are used in this area
Phase 2: Interview Rounds
Conduct 4-8 rounds of questions, with 1-4 questions per round.
Interview structure:
Round 1: Topic and goals
- If no topic was provided, start by asking what the user wants to work on
- Ask about the high-level goal and motivation
- Ask what problem this solves or what value it provides
Round 2: Requirements
- Ask about must-have requirements
- Ask about nice-to-have features
- Ask what's explicitly out of scope
Round 3-N: Deep dive (continue as needed)
- Between each round, read additional code files if needed based on user answers
- Ask about specific behavior, edge cases, and design decisions
- Ask about constraints (performance, compatibility, security)
- Ask about acceptance criteria and definition of done
- Probe areas where the user's answers raise new questions
Question guidelines:
- Ask 1-4 questions per round (prefer fewer, more focused questions)
- Let user responses guide the next round
- Read additional code between rounds if user mentions specific files/areas
- Continue until you have enough detail to write a plan or objective
Between-round processing:
After each round of answers:
- Summarize what you learned (internally, not to user)
- Identify gaps or ambiguities
- Read additional code files if needed
- Prepare next round's questions
Phase 3: Scope Determination
During a late interview round (round 4-6), ask about scope:
Based on what you've described, this work could be:
A) A single PR (plan-sized): Can be implemented in one focused pull request
B) Multiple PRs (objective-sized): Requires breaking into multiple independent PRs
Which approach makes sense for this work?
Listen for signals:
- Plan-sized: "This is straightforward", "Should be one PR", "It's focused"
- Objective-sized: "This is complex", "Multiple parts", "Will take several PRs"
Phase 4: Structured Summary
After completing the interview, output a structured summary to the conversation.
Output format:
## Interview Summary: [Topic]
### Context
[1-2 paragraphs summarizing what exists now and what's being requested]
### Requirements
**Must have:**
- [Bulleted list of must-have requirements]
**Nice to have:**
- [Bulleted list of nice-to-have features]
**Out of scope:**
- [Bulleted list of explicitly excluded items]
### Behavior
[Description of expected behavior, user flows, or interaction patterns]
### Edge Cases & Constraints
[Bulleted list of edge cases to handle and constraints to respect]
### Design Decisions
[Key architectural or design choices that should be made]
### Acceptance Criteria
[Bulleted checklist of how to verify the work is complete]
### Open Questions
[Bulleted list of any remaining ambiguities or decisions needed]
---
### Recommendation
**Scope assessment:** This is [plan-sized / objective-sized] work.
**Reasoning:** [1-2 sentences explaining why]
**Next step:** [Enter plan mode / Run `/erk:objective-create`]
Important:
- The summary should be comprehensive enough to start planning without re-asking questions
- All key information from the interview should be captured
- The recommendation should be clear and actionable
Key Principles
- Explore before asking - Read relevant code first to ask informed questions
- Listen and adapt - Let user responses guide subsequent questions
- Read between rounds - Consult code when user mentions specific areas
- Be thorough - Gather enough detail for effective planning
- No side effects - This command only reads and asks questions; it doesn't create files or enter plan mode itself
Error Cases
| Error | Message |
|---|---|
| Codebase search fails | Continue with interview (exploration optional) |
| User declines to answer | Note the gap in "Open Questions" |
| Unclear scope assessment | Default to plan-sized, note uncertainty |
Examples
Example: Plan-sized work
User: /local:interview add logout button
[Agent explores auth-related files, asks 4-6 questions about placement, behavior, confirmation]
[Agent outputs summary with recommendation: "This is plan-sized work. Next
Maintain Interview?
Let people know it's listed here — add the badge (live metrics, light/dark aware) or a plain link to your README or docs.
[Interview on getagentictools](https://getagentictools.com/loops/dagster-io-local-interview?ref=badge)