Agent Skills: Intent clarification

Intent clarification

Guide the user through a structured conversation to deeply understand their intent before creating specifications or solutions.

Purpose

Ensure alignment on:

• What problem we're solving
• Who we're solving it for
• What constraints exist
• What success looks like
• How the code should be written

Process

Introduction

Start by asking the user to describe their idea:

"Before we dive into specifics, could you describe your idea or the problem you're trying to solve in your own words?"

Listen actively and acknowledge their response before proceeding.

Step 1: Core intent

Ask each question one at a time, wait for response, then acknowledge:

Q1: Core problem/need

• Ask: "What is the core problem or need that this solution should address?"
• Example prompt: "For instance: 'We need a tool that helps new team members onboard more quickly by centralising our documentation.'"
• If vague, ask follow-up questions for specificity

Q2: Primary users and expectations

• Ask: "Who will be the primary users, and what do they expect to achieve?"
• Example prompt: "For example: 'New hires who need to find onboarding materials without asking multiple people.'"
• Probe for specific user personas if needed

Q3: Constraints and conditions

• Ask: "Are there any constraints we should be aware of?"
• Example prompt: "Such as: 'Must work on mobile and integrate with our SSO system.'"
• Cover: technical, budget, timeline, organisational constraints

Summary checkpoint:

• Synthesise into 2-3 clear bullet points
• Ask: "Based on what you've shared, here's what I understand... Does this accurately capture your intent?"
• Iterate until aligned

Step 2: Success criteria

Q4: One-sentence goal

• Ask: "If you had to describe the overall goal in one sentence, what would it be?"
• This forces clarity and reveals priorities

Q5: Measurable outcomes

• Ask: "How will we know when this is successful? What would we measure?"
• Push for specific, testable criteria
• Examples: "Users complete onboarding in under 2 hours", "Zero support tickets about finding documentation"

Step 3: Coding standards (for new projects)

When starting a new project, gather coding preferences:

Q6: Programming paradigm

• Ask: "Do you have a preference for how the code should be structured?"
• Options: functional, object-oriented, or mixed
• If unsure, recommend based on project type

Q7: Language and framework

• Ask: "What language and frameworks should we use?"
• Note any existing codebase constraints

Q8: Patterns and practices

• Ask: "Are there specific patterns you want to follow or avoid?"
• Examples to offer: "Repository pattern, dependency injection, event sourcing"
• Also ask: "Any patterns or practices to explicitly avoid?"

Q9: Naming conventions

• Ask: "Do you have naming conventions for files, functions, variables?"
• If none, suggest sensible defaults for the chosen language

Step 4: Final confirmation

Summarise the complete intent:

• Core problem
• Target users and needs
• Constraints
• Success criteria
• Coding standards

Ask: "Does this fully reflect what you want to achieve?"

Output

Once confirmed, create structured output for the manifest:

intent:
  problem_statement: |
    Clear articulation of the core problem being solved.

  users:
    - persona: "New hire"
      needs:
        - "Find onboarding materials quickly"
        - "Complete setup without asking colleagues"

    - persona: "HR manager"
      needs:
        - "Track onboarding progress"
        - "Update materials easily"

  constraints:
    technical:
      - "Must integrate with existing SSO"
      - "Mobile-responsive required"
    business:
      - "Budget: internal team only"
      - "Timeline: Q1 delivery"
    organisational:
      - "Must comply with data retention policy"

  success_criteria:
    - metric: "Onboarding completion time"
      target: "< 2 hours"
    - metric: "Support tickets about documentation"
      target: "Zero in first month"
    - metric: "User satisfaction score"
      target: ">= 4.5/5"

  coding_standards:
    paradigm: "functional"
    language: "TypeScript"
    framework: "Next.js"
    patterns:
      - "Repository pattern for data access"
      - "Dependency injection"
      - "Event-driven for notifications"
    forbidden:
      - "No class inheritance deeper than 2 levels"
      - "No mutable global state"
    naming:
      files: "kebab-case"
      functions: "camelCase"
      types: "PascalCase"
      constants: "SCREAMING_SNAKE_CASE"

Save to .sdlc/manifest.yaml under the intent key.

Tone and style

• Be conversational and patient
• Don't rush through questions
• Validate and acknowledge each response
• Ask clarifying questions when ambiguous
• Use examples to illustrate expected detail level

Checkpoint

Intent is a configured checkpoint. After completing:

1. Present the full intent summary
2. Ask for explicit approval
3. Only proceed to requirements phase after confirmation

Tips

• One question at a time; don't overwhelm
• Silence is okay; let users think
• Vague answers need follow-up, not acceptance
• If user can't articulate success criteria, they may not be ready to build
• Coding standards can be skipped for existing projects with established patterns