---

name: spec-validate description: Validation loop for speccing. Clarifies requirements through structured questioning before document creation.

Spec Validate Skill

Validate requirements through structured clarification. Produces validation data for spec-create.

Reference: See reference/issue-types.md for type definitions, reference/question-taxonomy.md for question templates, and reference/sdd-gates.md for pre-implementation gate definitions.

---

When to Use

Use for:

• New features before implementation
• Unclear requirements needing refinement
• Complex changes needing design exploration
• When multiple approaches seem viable

Don't use for:

• Simple bug fixes
• Well-specified changes
• Clear mechanical tasks (use Claude's native planning)

---

Workflow

Step 0: Check Native Plan Context

Before starting validation, check if the user has existing context from Claude's native /plan:

1. Check for context: Ask if /plan was used or if there's existing plan context
2. If present: Extract key elements:
   • Goal/objective → Seeds Scope taxonomy area
   • Approach/strategy → Seeds Integration/Architecture areas
   • Open questions → Become priority clarification targets
3. If absent: Proceed directly to Step 1

This step bridges native Claude planning with structured validation.

Step 0.5: Constitution Check (Initiatives Only)

For Initiative-type work, verify alignment with constitution before proceeding:

1. Read .claude/constitution.md
2. Check if proposed work aligns with core principles
3. If conflict detected:
   • Flag the conflict
   • Ask user to resolve or justify exception
   • Document exception in validation data

Skip for: Features, Tasks (constitution checked during task-dispatch for Initiatives)

Step 1: Issue Type Selection

FIRST QUESTION (Always) - Use AskUserQuestion:

CopyHeader: Work type
Question: What type of work is this?
multiSelect: false
Options:
- Initiative: Strategic coordination (months) - Multiple features toward business goal
- Feature: User-facing capability (weeks) - Deliverable value, multiple tasks
- Task: Implementation item (days) - Single concrete deliverable
- Exploratory: Not sure yet - Gather context first, then classify

This selection determines:

• Question limit: Tasks get 3, Features/Initiatives get 5
• Taxonomy areas: Tasks get minimal (3), Features/Initiatives get full (7)

If Exploratory: Gather context, ask 3 questions to understand scope, present classification recommendation, restart with correct type.

Step 2: Gather Context

1. Examine relevant files, docs, recent commits
2. Understand existing patterns and constraints
3. Initialize taxonomy tracking based on issue type

Taxonomy by type:

Step 2.5: Ambiguity Scan

Automatically scan gathered context for specification gaps across taxonomy areas.

Process:

1. For each taxonomy area (based on issue type), evaluate:

   • clear: Fully specified, no questions needed
   • partial: Some information present, gaps remain
   • missing: Not addressed at all

2. Populate ambiguity_scan section in validation data:

   Copyambiguity_scan:
     scope:
       status: clear | partial | missing
       gaps: ["gap description if partial/missing"]
     behavior:
       status: clear | partial | missing
       gaps: []
     # ... remaining areas
   

3. Route based on scan results:

   • No gaps (all clear): Proceed silently to Step 4 (skip validation loop)
   • Gaps found: Areas with partial or missing status become priority candidates for clarification questions in Step 3

Evaluation criteria per area:

Step 3: Validation Loop

Ask clarifying questions in taxonomy-based batches, with re-evaluation between rounds.

Process:

1. Identify uncovered taxonomy areas
2. Prioritize areas by (Impact × Uncertainty)
3. Group questions by taxonomy area into batches
4. Use AskUserQuestion with multiple questions from the same area
5. Receive answers
6. Re-evaluate remaining questions for relevance (skip questions invalidated by answers)
7. Update taxonomy coverage
8. Repeat with next taxonomy area until limit reached or all Primary areas covered

Batch format:

CopyAskUserQuestion with questions array (1-4 questions per batch):

Question 1:
  Header: [Area, max 12 chars]
  Question: [Clear question ending with ?]
  multiSelect: false
  Options: [2-4 options with implications]

Question 2:
  Header: [Same area]
  Question: [Related question]
  ...

Single question format (when only one question in area):

CopyHeader: [Area, max 12 chars]
Question: [Clear question ending with ?]
multiSelect: false
Options:
- Option A: [choice] - [implication]
- Option B: [choice] - [implication]
- Option C: [choice] - [implication]
- None: [default/skip]

Batching rules:

• One batch per taxonomy area (questions grouped by area)
• Multiple trigger fields within a batch combine with OR (any match)

Batch counts:

• Tasks: up to 3 batches (Scope, Behavior, Integration)
• Features/Initiatives: up to 7 batches (all taxonomy areas)
• Plus: additional clarification batches if "Other" answers need follow-up

Batch size limits:

• Tasks: up to 3 questions per batch
• Features/Initiatives: up to 4 questions per batch (AskUserQuestion max)

Re-evaluation between batches:

• After receiving answers, check if pending questions are still relevant
• Skip questions invalidated by previous answers
• If user provides ambiguous "Other" answer spanning areas, add follow-up to next batch

Step 3.5: Initiative-Specific Validation (Initiatives Only)

For Initiatives, ask about user story prioritization and implementation strategy:

CopyHeader: User Stories
Question: How should user stories be prioritized?
multiSelect: false
Options:
- MVP First (Recommended): P1 stories deliver standalone value, P2/P3 are incremental
- Parallel Tracks: Stories can be developed independently by different teams
- Sequential: Stories have strict dependencies, must complete in order

CopyHeader: Strategy
Question: What implementation approach fits best?
multiSelect: false
Options:
- MVP First (Recommended): Ship P1, iterate on P2/P3 based on feedback
- Incremental: Each phase adds value, all planned upfront
- Parallel Team: Multiple workstreams, integration points defined

Track selections in validation data for spec-create (populates Implementation Strategy section).

Step 3.6: SDD Section Opt-ins (Features Only)

For Features, offer opt-in for detailed SDD sections:

CopyHeader: SDD sections
Question: Which detailed sections do you want in the spec?
multiSelect: true
Options:
- Tech Decisions: Document technology choices and rationale
- API Contract: Define API endpoints and schemas
- Data Model: Document entities and relationships
- None: Keep spec lightweight

Track selections in validation data for spec-create.

Step 4: Propose Approaches

Use AskUserQuestion to present options:

CopyHeader: Approach
Question: Which approach should we take?
multiSelect: false
Options:
- Approach A: [brief] - Trade-off: [X]
- Approach B: [brief] - Trade-off: [Y] (Recommended)
- Approach C: [brief] - Trade-off: [Z]

Lead with your recommendation. Apply YAGNI ruthlessly.

Step 5: Generate Validation Data

Compile validation data for spec-create:

• Issue type: Selected in Step 1
• Ambiguity scan: Status per area (clear / partial / missing) with gap descriptions
• Questions used: N / limit
• Taxonomy coverage: Status per area (Covered / Gap / N/A)
• Clarification log: Question → Answer → Integration point
• Approach selected: From Step 4
• SDD opt-ins: Which sections selected (Features only)
• Gates status: For Initiatives: run gate checks; for Features/Tasks: mark n/a
• Markers: Any unresolved items identified during validation

This data passes to spec-create for validation.yaml.

---

Key Principles

---

Output

This skill produces validation data, not documents. The data flows to spec-create which generates:

• spec.md, context.md, tasks.md, dependencies.yaml, validation.yaml

---

Integration

Invoked by: /spec.create command (default behavior)

Standalone use: Can be invoked directly via spec-validate skill for validation without document creation.