interview
Interview me about anything in depth
When & Why to Use This Skill
The Deep Interview skill is a sophisticated AI agent tool designed to conduct structured, iterative discovery sessions. It excels at uncovering technical requirements, architectural tradeoffs, and hidden edge cases by engaging the user in a targeted Q&A process. Whether analyzing a file or a conceptual idea, it helps bridge the gap between high-level thoughts and detailed, actionable specifications.
Use Cases
- Technical Requirement Gathering: Use the skill to define the architecture of a new system by answering guided questions about data flow, sync/async patterns, and error handling.
- Design Document Refinement: Provide an existing technical document or file for the agent to analyze, identify ambiguities, and automatically update with refined insights and decisions.
- Complex Problem Solving: Deep dive into specific implementation challenges to evaluate alternatives and determine the best approach based on project constraints.
- Knowledge Transfer and Discovery: Conduct an in-depth exploration of a new technology or business domain to quickly identify key priorities and potential risks.
| name | interview |
|---|---|
| description | Interview me about anything in depth |
| argument-hint | [topic or file] |
| model | opus |
Deep Interview Skill
Conduct an in-depth interview about any topic, file, or idea the user provides.
Detecting Input Type
First, determine what the user wants to discuss:
If
$1looks like a file path (contains/,.md,.ts, etc.):- Read the file using the Read tool
- Interview about its contents
- At the end, update the file in-place with refined insights
If
$1is a topic or description:- Interview about that concept/idea
- At the end, summarize the key insights
Interview Process
Round-by-Round Approach
Interview iteratively - one round of questions at a time:
- Analyze what you know so far
- Identify the most important ambiguities, assumptions, or unexplored areas
- Ask 1-4 questions using
AskUserQuestiontool - Process the answers
- Repeat until the user says "done" or you've covered everything meaningful
Question Quality Rules
DO ask about:
- Implementation tradeoffs ("Should this be sync or async?")
- Edge cases ("What happens when the input is empty?")
- Scope boundaries ("Is X in scope for the first version?")
- User preferences ("Do you prefer explicit errors or silent fallbacks?")
- Architecture choices ("Should this be a separate service or integrated?")
- Constraints ("Are there performance requirements?")
- Alternatives ("Have you considered approach Y instead?")
DON'T ask:
- Obvious things ("Do you want tests?")
- Things you can infer ("What language?" when codebase is TypeScript)
- Yes/no validation questions ("Is this correct?")
- Surface-level stuff ("What's the feature name?")
- Questions answerable by quick code discovery (read configs/existing patterns first)
- Open-ended questions when multiple-choice would eliminate branches faster
Question Format
Always use AskUserQuestion with multiple choice options. Optimize for fast responses:
Standard format:
AskUserQuestion:
questions:
- question: "How should the system handle API failures?"
header: "Errors"
options:
- label: "Retry with backoff (Recommended)"
description: "Automatically retry failed requests with exponential backoff"
- label: "Fail fast"
description: "Return error immediately, let caller decide"
- label: "Not sure - you decide"
description: "Let me pick based on codebase patterns"
multiSelect: false
For power users: When asking multiple questions, add this footer:
Reply format:
1a 2bordefaultsto accept all recommended options
Always include:
- One clearly marked "(Recommended)" option when you have a strong opinion
- A "Not sure - you decide" escape hatch for low-stakes decisions
Interview Categories
Adapt questions based on context, but consider exploring:
Technical Implementation
- Architecture patterns
- Technology choices
- Integration points
- Data flow
User Experience
- Interaction patterns
- Error states
- Edge cases
- Feedback mechanisms
Constraints & Requirements
- Performance needs
- Security considerations
- Scalability requirements
- Compliance/regulatory
Scope & Priorities
- Must-have vs nice-to-have
- First version vs future iterations
- Dependencies and blockers
Risks & Concerns
- What could go wrong?
- What are you uncertain about?
- What alternatives exist?
Completion
When the interview is complete (user says "done" or all areas explored):
For file input:
- Summarize key decisions made during the interview
- Update the original file with refined information
- Add an "Interview Insights" or similar section if appropriate
- Preserve the original structure
For topic input:
- Provide a comprehensive summary of insights gathered
- List key decisions and preferences discovered
- Highlight any unresolved questions or areas for future exploration
Topic
$ARGUMENTS