---

name: spec description: Create specifications directly from clear requirements - uses spec-kit tools to create formal, executable specs following WHAT/WHY principle (not HOW)

Direct Specification Creation

Overview

Create formal specifications directly when requirements are clear and well-defined.

Use this instead of brainstorm when:

• Requirements are already clear
• User provides detailed description
• Feature scope is well-defined
• No exploratory dialogue needed

This skill creates specs using spec-kit tools and ensures WHAT/WHY focus (not HOW).

When to Use

Use this skill when:

• User provides clear, detailed requirements
• Feature scope is well-defined
• User wants to skip exploratory dialogue
• Requirements come from external source (PRD, ticket, etc.)

Don't use this skill when:

• Requirements are vague or exploratory → Use sdd:brainstorm
• Spec already exists → Use sdd:implement or sdd:evolve
• Making changes to existing spec → Use sdd:spec-refactoring

Prerequisites

Ensure spec-kit is initialized:

{Skill: spec-kit}

If spec-kit prompts for restart, pause this workflow and resume after restart.

Critical: Specifications are WHAT and WHY, NOT HOW

Specs define contracts and requirements, not implementation.

✅ Specs SHOULD include:

• Requirements: What the feature must do
• Behaviors: How the feature should behave (user-observable)
• Contracts: API structures, file formats, data schemas
• Error handling rules: What errors must be handled and how they should appear to users
• Success criteria: Measurable outcomes
• Constraints: Limitations and restrictions
• User-visible paths: File locations, environment variables users interact with

❌ Specs should NOT include:

• Implementation algorithms: Specific sorting algorithms, data structure choices
• Code: Function signatures, class hierarchies, pseudocode
• Technology choices: "Use Redis", "Use React hooks", "Use Python asyncio"
• Internal architecture: How components communicate internally
• Optimization strategies: Caching mechanisms, performance tuning

📋 Example: What belongs where

SPEC (WHAT/WHY):

Copy## Requirements
- FR-001: System MUST validate email addresses before storing
- FR-002: System MUST return validation errors within 200ms
- FR-003: Invalid emails MUST return 422 status with error details

## Error Handling
- Invalid format: Return `{"error": "Invalid email format", "field": "email"}`
- Duplicate email: Return `{"error": "Email already exists"}`

PLAN (HOW):

Copy## Validation Implementation
- Use regex pattern: `^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$`
- Cache validation results in Redis (TTL: 5 min)
- Database query: `SELECT COUNT(*) FROM users WHERE email = ?`

Why this matters:

• Specs remain stable - Implementation details change, requirements don't
• Implementation flexibility - Can change HOW without changing WHAT
• Clearer reviews - Easy to see if requirements are met vs implementation quality
• Better evolution - When code diverges from spec, know which to update

The Process

1. Gather Requirements

Extract from user input:

• What needs to be built
• Why it's needed (purpose/problem)
• Success criteria
• Constraints and dependencies
• Error cases and edge conditions

Ask clarifying questions (brief, targeted):

• Only if critical information is missing
• Keep questions focused and specific
• Don't turn this into full brainstorming session

If requirements are vague: Stop and use sdd:brainstorm instead.

2. Check Project Context

Review existing specs:

Copyls -la specs/features/
# Or: ls -la specs/[NNNN]-*/

Check for constitution:

Copycat .specify/memory/constitution.md

Look for related features:

• Similar functionality already specced
• Integration points
• Shared components

3. Create Specification

Use spec-kit tools:

Copy# Interactive spec creation using spec-kit template
speckit specify "[feature description]"

# Or use spec-kit scripts directly
.specify/scripts/bash/create-new-feature.sh --json "[feature description]"

This will:

• Create feature directory (e.g., specs/0001-feature-name/)
• Initialize spec.md from template
• Set up directory structure (docs/, checklists/, contracts/)

Fill in the spec following template structure:

• Purpose - WHY this feature exists
• Functional Requirements - WHAT it must do
• Non-Functional Requirements - Performance, security, etc.
• Success Criteria - Measurable outcomes
• Error Handling - What can go wrong
• Edge Cases - Boundary conditions
• Constraints - Limitations
• Dependencies - What this relies on
• Out of Scope - What this doesn't do

Follow WHAT/WHY principle:

• Focus on observable behavior
• Avoid implementation details
• Use user/system perspective
• Keep technology-agnostic where possible

4. Validate Against Constitution

If constitution exists:

Copycat .specify/memory/constitution.md

Check alignment:

• Does spec follow project principles?
• Are patterns consistent with constitution?
• Does error handling match standards?
• Are architectural decisions aligned?

Note any deviations and justify them in spec.

5. Review Spec Soundness

Before finishing, validate:

Use sdd:review-spec skill to check:

• Completeness (all sections filled)
• Clarity (no ambiguous language)
• Implementability (can generate plan from this)
• Testability (success criteria measurable)

If review finds issues:

• Fix critical issues before proceeding
• Document any known gaps
• Mark unclear areas for clarification

6. Commit Spec

Create git commit:

Copygit add specs/[feature-dir]/
git commit -m "Add spec for [feature-name]"

Spec is now source of truth for this feature.

Next Steps

After spec creation:

1. Review spec soundness (if not already done):

   CopyUse sdd:review-spec
   

2. Implement the feature:

   CopyUse sdd:implement
   

3. Or refine spec further if issues found

Remember

Spec is contract, not design doc:

• Defines WHAT and WHY
• Defers HOW to implementation
• Remains stable as code evolves
• Is source of truth for compliance

Keep specs:**

• Technology-agnostic
• User-focused
• Measurable
• Implementable

The goal: A clear, unambiguous specification that serves as the single source of truth for implementation and validation.