Interview

Overview

This skill provides a structured interview framework to systematically gather information before starting work. It helps reduce ambiguity, ensure comprehensive understanding, and produce actionable outputs. The interview adapts to different contexts: requirements definition, debugging, architecture decisions, security review, and more.

When to Use

• Starting a new feature implementation

• Investigating bugs or issues

• Reviewing architecture or security

• Creating documentation or reports

• Any task requiring clarification of scope, constraints, or expectations

Workflow

Phase 1: Preparation (Silent)

Before asking questions, gather background context silently.

Parallel Fan-Out Pattern: The following three preparation tasks have no dependencies on each other and SHOULD be executed in parallel. This reduces Phase 1 completion time by approximately 66%.

Reference: Google ADK Multi-Agent Patterns

Codebase Exploration: (parallel task 1/3)

• Project structure - Identify key directories, config files, entry points

• Recent changes - Check git history if applicable

• Related code - Search for relevant patterns or implementations

Documentation Review: (parallel task 2/3)

• Read any referenced files or URLs provided by user

• Check for existing specifications, ADRs, or design documents

• Review related issues or PRs if applicable

Web Research (if applicable): (parallel task 3/3)

• Best practices for the domain

• Similar implementations or patterns

• Security considerations

For detailed exploration commands per environment (Claude Code, Codex, API), see references/exploration-patterns.md .

Phase 2: Interview

Conduct the interview in stages, adapting questions based on interview type.

Stage 1: Goal Confirmation

Start by confirming the overall objective:

Environment-specific approach:

• Claude Code: Use AskUserQuestion tool for structured choices

• Other environments: Present numbered options and ask user to reply with number or description

Core Questions:

• What is the primary goal or outcome expected?

• Who are the stakeholders or users affected?

• What does success look like?

Interview Type Selection:

Ask the user to select the interview type to tailor subsequent questions:

Type Use Case

Requirements New feature, specification, API design

Investigation Bug analysis, performance issue, incident

Architecture Design review, technology selection, refactoring

Security Security audit, vulnerability assessment

Documentation Report creation, knowledge transfer

General Open-ended exploration, brainstorming

Stage 2: Deep Dive

Based on the selected interview type, ask targeted questions.

For detailed question frameworks and output templates per interview type, read references/interview-types.md .

Stage 3: Confirmation and Prioritization

Before concluding:

• Summarize Understanding - Restate key points for confirmation

• Identify Gaps - Note any undecided or unclear items

• Prioritize - Classify requirements as Must/Should/Could

• Confirm Scope - Agree on what is in/out of scope

Deep Dive Strategies

Use these techniques to reduce ambiguity:

Technique When to Use

"Specifically?" When details are vague

"Why?" When motivation is unclear

"What else?" When list seems incomplete

"For example?" When concept needs illustration

"What if...?" When edge cases need exploration

Phase 3: Output

Generate a structured summary document.

Output Template

[Interview Type]: [Topic]

Summary

[1-2 sentence overview]

Goal

• Objective: [Primary goal]
• Stakeholders: [Who is affected]
• Success Criteria: [How to measure success]

Requirements / Findings

Must Have

• [Item 1]
• [Item 2]

Should Have

• [Item 1]

Could Have

• [Item 1]

Constraints

• [Technical constraints]
• [Business constraints]
• [Timeline constraints]

Undecided / Open Questions

• [Question 1]
• [Question 2]

Next Steps

1. [Action item 1]
2. [Action item 2]

References

• [Link or file reference 1]
• [Link or file reference 2]

Affected Files / Components

• path/to/file1
• path/to/file2

Usage Examples

For detailed usage examples with sample outputs, see references/usage-examples.md .

Quick reference:

• /interview Add user authentication → Requirements interview

• /interview Investigate 504 errors → Investigation interview

• /interview Review database schema → Architecture interview

Guidelines

• Adapt to Context - Adjust question depth based on task complexity

• Avoid Overwhelming - Ask 2-3 questions at a time, not all at once

• Be Specific - Reference actual code, files, or examples when possible

• Document Everything - Capture decisions and their rationale

• Identify Gaps Early - Surface undecided items for follow-up

• Output Language - Follow user's language preference (check conversation history)

User Input Methods

Claude Code Environment

Use the AskUserQuestion tool for structured choices:

Question: "What type of interview is this?" Header: "Type" Options:

• "Requirements (Recommended)" - New feature or specification
• "Investigation" - Bug or issue analysis
• "Architecture" - Design or technology review
• "Security" - Security assessment

This enables efficient selection with clickable options.

Other Environments (Codex, API, etc.)

Present numbered options in plain text and wait for user response:

What type of interview is this?

1. Requirements - New feature or specification (Recommended)
2. Investigation - Bug or issue analysis
3. Architecture - Design or technology review
4. Security - Security assessment
5. Documentation - Report creation, knowledge transfer
6. General - Open-ended exploration

Please reply with a number (1-6) or describe your needs:

Wait for the user's response before proceeding. Accept both:

• Number selection (e.g., "1", "2")

• Free-form description (e.g., "I need to investigate a performance issue")