@.claude/shared-imports/constitution.md @.claude/templates/clarification-checklist.md

Specification Clarification

Purpose: Systematically eliminate ambiguity from specifications through structured questioning before implementation planning.

Constitutional Authority: Article IV (Specification-First Development), Article V (Template-Driven Quality)

Quick Reference

Workflow	Key Activities	Output
Clarification	Load spec → Scan ambiguities → Prioritize → Ask questions → Update incrementally	spec.md (resolved)
Validation	Verify consistency → Update coverage → Report completion	Readiness status

Workflow Files

Detailed Workflows:

•@.claude/skills/clarify-specification/workflows/clarification-workflow.md - Complete Phases 1-4 (load, prioritize, clarify, validate)

Examples:

•@.claude/skills/clarify-specification/examples/clarification-example.md - E-commerce checkout feature walkthrough

References:

•@.claude/skills/clarify-specification/references/failure-modes.md - 10 common failure modes with solutions

Clarification Workflow (Overview)

See: @.claude/skills/clarify-specification/workflows/clarification-workflow.md

Summary:

Phase 1: Load Specification and Detect Ambiguities

Step 1.1: Read current specification from specs/<feature>/spec.md

Step 1.2: Scan against 10+ ambiguity categories:

•Functional Scope & Behavior
•Domain & Data Model
•Interaction & UX Flow
•Non-Functional Requirements (performance, scale, security)
•Integration & Dependencies
•Edge Cases & Failure Scenarios
•Constraints & Tradeoffs
•Terminology & Definitions
•Permissions & Access Control
•State & Lifecycle

Step 1.3: Assess coverage for each category:

•Clear (10 points) - Well-defined, no ambiguity
•Partial (5 points) - Some aspects defined, others unclear
•Missing (0 points) - Not addressed in specification

Coverage Formula:

code

Coverage := ∑(c_i) where c_i ∈ {clear: 10, partial: 5, missing: 0}
Readiness := Coverage / (10 × num_categories) ≥ 70%

Phase 2: Prioritize Clarification Questions

Step 2.1: Extract existing [NEEDS CLARIFICATION] markers (Article IV limit: max 3)

Step 2.2: Prioritize by impact (Article IV order):

•Scope (highest impact) - Affects what gets built
•Security - Affects risk and compliance
•UX Flow - Affects user experience
•Technical (lowest impact) - Implementation details

Maximum 5 Questions Per Iteration (Article IV requirement)

Step 2.3: Generate questions with structured format:

•Context: Why this matters
•Question: Specific, focused inquiry
•Options: 2-3 recommendations based on common patterns
•Impact: What depends on this answer
•Intelligence Evidence: project-intel.mjs findings or MCP query results

Phase 3: Interactive Clarification

Step 3.1: Present questions sequentially (ONE AT A TIME for complex topics)

Step 3.2: Capture user response with rationale and additional context

Step 3.3: Update specification incrementally AFTER EACH answer:

•Edit spec.md to incorporate answer
•Remove or resolve [NEEDS CLARIFICATION] marker
•Add functional requirement with answer
•Verify no contradictions introduced

Critical: Incremental updates prevent contradictions and lost context

Phase 4: Validation and Completion

Step 4.1: Verify consistency (no conflicts between new and existing requirements)

Step 4.2: Update clarification checklist with resolved categories

Step 4.3: Report completion:

code

✓ Clarification complete: N questions resolved
✓ Updated specification with specific requirements
✓ Remaining ambiguities: M markers (≤3 per Article IV)

Readiness Gate: Coverage ≥ 70% AND ≤ 3 [NEEDS CLARIFICATION] markers remaining

Next Step: Use create-implementation-plan skill to define HOW

Phase 5: Re-Clarification (If Needed)

Trigger clarification again if:

•Implementation planning reveals new ambiguities
•User requests changes to requirements
•New [NEEDS CLARIFICATION] markers added during planning

Each iteration: Max 5 new questions, focus on highest-priority gaps, update incrementally

Anti-Patterns to Avoid

DO NOT:

•Ask more than 5 questions per iteration (Article IV limit)
•Ask open-ended questions without recommendations
•Present all questions at once (use sequential for complex topics)
•Make assumptions instead of asking
•Skip updating specification after each answer
•Accept ambiguous answers (press for specifics)

DO:

•Prioritize by impact (scope > security > UX > technical)
•Provide 2-3 options with recommendations
•Use intelligence queries (project-intel.mjs) for context
•Update spec incrementally (after each answer)
•Verify consistency after updates
•Limit [NEEDS CLARIFICATION] markers to max 3

Example: E-Commerce Checkout Clarification

See: @.claude/skills/clarify-specification/examples/clarification-example.md

Summary:

Input: Specification with 4 [NEEDS CLARIFICATION] markers (exceeds limit), vague acceptance criteria

Process:

•Scanned 10 categories → 31.25% coverage (FAIL, < 70%)
•Prioritized 5 questions by impact (Scope > Integration > Non-Functional)
•Asked questions sequentially with options and evidence
•Updated spec after each answer
•Validated consistency

Output:

•81.25% coverage (PASS, > 70%)
•1 marker remaining (deferred low-priority technical detail)
•All ACs specific and testable
•Ready for implementation planning

Time Investment: 15-20 minutes of clarification saved hours of rework

Prerequisites

Before using this skill:

•✅ spec.md exists (created by specify-feature skill)
•✅ [NEEDS CLARIFICATION] markers present in spec OR user mentions ambiguity
•✅ Feature directory structure exists: specs/<feature>/
•⚠️ Optional: clarification-checklist.md (for category coverage tracking)
•⚠️ Optional: project-intel.mjs (for evidence-based recommendations)

Dependencies

Depends On:

•specify-feature skill - MUST run before this skill (creates initial spec.md)

Integrates With:

•create-implementation-plan skill - Uses clarified spec.md as input (typical successor)
•specify-feature skill - May trigger this skill if ambiguities detected

Tool Dependencies:

•Read tool (to load spec.md and templates)
•Write, Edit tools (to update spec.md incrementally)
•project-intel.mjs (optional, for evidence-based recommendations)

Next Steps

After clarification completes, typical progression:

If all ambiguities resolved:

code

clarify-specification (resolves ambiguities)
    ↓
create-implementation-plan (user invokes /plan or auto-triggered)
    ↓
generate-tasks (auto-invoked)
    ↓
/audit (auto-invoked)

If new ambiguities discovered during planning:

code

create-implementation-plan (finds gaps)
    ↓
clarify-specification (invoked again)
    ↓
create-implementation-plan (continues after resolution)

User Action Required:

•Answer clarification questions (max 5 per iteration)
•Provide specific answers, not vague responses
•Confirm specification updates after each answer

Outputs Modified:

•specs/$FEATURE/spec.md - Updated incrementally with clarifications
•Removed [NEEDS CLARIFICATION] markers
•Added functional requirements with clarified details

Commands:

•/plan spec.md - After clarification complete, create implementation plan
•/clarify - User-facing command that invokes this skill

Agent Integration

This skill operates in the main conversation context but may be invoked by other agents when they encounter ambiguities.

Invocation Patterns

User-Initiated (most common):

code

User notices ambiguity → runs /clarify → clarify-specification skill executes

Agent-Initiated (during planning):

code

implementation-planner agent (finds ambiguity during plan creation)
    ↓ invokes
clarify-specification skill via instruction
    ↓ returns
Updated spec.md with resolved ambiguity

Code Analyzer Support (Optional)

When: If clarification requires understanding existing codebase patterns for evidence-based recommendations

Agent: code-analyzer

Example Task Tool Invocation:

python

# If clarification needs code analysis for evidence
Task(
    subagent_type="code-analyzer",
    description="Analyze existing authentication patterns",
    prompt="""
    @.claude/agents/code-analyzer.md

    Analyze existing authentication in codebase to inform
    clarification question about auth strategy.

    Use project-intel.mjs to find auth patterns.
    Output: What auth patterns exist (OAuth, JWT, sessions, etc.)
    """
)

Use Case: When user asks "What auth should we use?", analyzer provides evidence from existing patterns

Integration with Planner

Typical Flow:

code

clarify-specification (resolves all ambiguities)
    ↓ updates
spec.md (all [NEEDS CLARIFICATION] removed)
    ↓ ready for
create-implementation-plan skill

Iterative Flow (if planner discovers new ambiguities):

code

implementation-planner (finds gap while planning)
    ↓ invokes
clarify-specification (targeted question on gap)
    ↓ user answers
    ↓ updates spec.md
    ↓ returns to
implementation-planner (continues planning)

Task Tool Usage

This skill typically does NOT use Task tool directly. It:

•Runs in main conversation context (needs user interaction)
•Updates spec.md incrementally based on user answers
•May suggest analyzer agent if code evidence needed (but doesn't invoke directly)

Design Rationale:

•Clarification requires user dialog (can't run in isolated agent)
•Incremental updates more efficient than agent round-trips
•User must approve spec changes (can't delegate to agent)

Failure Modes

See: @.claude/skills/clarify-specification/references/failure-modes.md

Summary of Common Failures:

•Too many [NEEDS CLARIFICATION] markers (> 3) - Violates Article IV → Prioritize and defer low-priority details
•Open-ended questions without options - Wastes time → Always provide 2-3 options with recommendations
•Asking > 5 questions per iteration - Violates Article IV → Prioritize by impact, ask top 5 only
•Not updating spec after each answer - Introduces contradictions → Update incrementally after EACH answer
•Accepting ambiguous answers - Defeats purpose → Press for specifics, offer more options
•No prioritization - Wastes effort → Always use Article IV order (Scope > Security > UX > Technical)
•Introducing contradictions - Breaks spec consistency → Validate after EACH update
•No intelligence evidence - Recommendations ignore codebase → Query project-intel.mjs before recommending
•Iterating forever - Analysis paralysis → Stop at ≥70% coverage, ≤3 markers
•Not tracking coverage - Can't measure completion → Use clarification-checklist.md matrix

Diagnostic Workflow: Check Article IV compliance → Question quality → Update process → Answer quality → Evidence → Stopping conditions

Related Skills & Commands

Direct Integration:

•specify-feature skill - Creates spec.md that this skill refines (required predecessor)
•create-implementation-plan skill - Uses clarified spec.md as input (typical successor)
•/clarify command - User-facing command that invokes this skill

Workflow Context:

•Position: Phase 1.5 of SDD workflow (between specification and planning)
•Triggers: [NEEDS CLARIFICATION] markers OR user mentions "unclear requirements"
•Output: Updated spec.md with resolved ambiguities

Quality Gates:

•Pre-Planning: Ensures spec is unambiguous before creating plan (Article IV)
•Max 3 Markers: Article IV limits [NEEDS CLARIFICATION] markers to 3 max
•Max 5 Questions: Article IV limits clarification questions to 5 per iteration

Workflow Diagram:

code

specify-feature (creates spec.md with 0-3 [NEEDS CLARIFICATION] markers)
    ↓ (if markers exist OR ambiguities detected)
clarify-specification (resolves ambiguities, max 5 questions/iteration)
    ↓ (when all resolved)
create-implementation-plan (can proceed without specification gaps)

Re-Clarification Trigger: If create-implementation-plan discovers new gaps, it can trigger clarify-specification again for iterative refinement

Version: 1.1.0 Last Updated: 2025-10-23 Change Log:

•v1.1.0 (2025-10-23): Refactored to progressive disclosure pattern (<500 lines)
•v1.0.0 (2025-10-22): Initial version