| name | spec |
| description | Living Specifications for Claude Code - consolidate development artifacts into AI-maintainable spec files with AI-DLC principles. |
| argument-hint | [command] or [feature description] |
| disable-model-invocation | false |
| user-invocable | true |
| allowed-tools | Read, Write, Edit, Grep, Glob, Bash, Task, TaskCreate, TaskUpdate, TaskList, AskUserQuestion |
/spec - Living Specifications
Multi-agent orchestration for AI-maintainable specifications.
Usage
/spec # Start new or continue existing
/spec <feature> # Create spec for specific feature
/spec status # View all specs and phases
/spec drift # Check spec-code drift
/spec view <role> # Role-based view (developer|manager|qa|architect)
First-Time Activation
MANDATORY: Use AskUserQuestion on first trigger:
{
"questions": [{
"header": "Approach",
"question": "Which spec approach fits your project?",
"options": [
{"label": "A) Living Spec Only (Recommended)", "description": "MVPs, small teams, rapid iteration"},
{"label": "B) Living Spec + Feature Specs", "description": "Multiple features, growing projects"},
{"label": "C) Feature Specs Only", "description": "Clear feature boundaries, simple projects"}
],
"multiSelect": false
}]
}
Then ask Greenfield/Brownfield via AskUserQuestion.
Core Files
Always load: ~/.claude/skills/steering/workflow.md (core logic)
| File | Load When |
|---|
steering/workflow.md | Always |
steering/template.md | Creating new spec |
steering/ears-reference.md | Writing requirements |
steering/drift-detection.md | /spec drift |
agents/INDEX.md | Spawning agents |
Multi-Agent Orchestration
See agents/INDEX.md for complete agent directory and spawning patterns.
Planning Phase: Spawn in parallel via Task tool:
- requirements-analyst, architecture-reviewer, risk-assessor
Building Phase: Spawn relevant specialists:
- database, api, frontend, security, test specialists
Quality Gates: spec-critic, comprehension-gate
AI-DLC Phases
| Phase | Icon | Purpose |
|---|
| Planning | 🔵 | WHAT and WHY |
| Building | 🟢 | HOW |
| Operating | 🟡 | RUN and MEASURE |
Tiered Approval
| Tier | Type | Examples |
|---|
| 1 | Autonomous | Timestamps, status icons, drift |
| 2 | Notify | Component Map, tech debt, decisions |
| 3 | Blocking | Requirements, architecture, phase transitions |
EARS Format
See steering/ears-reference.md for syntax. Key templates:
- Ubiquitous: THE system SHALL [response]
- Event: WHEN [trigger] THE system SHALL [response]
- Unwanted: IF [condition] THEN THE system SHALL [response]
Directory Structure
project/
├── CLAUDE.md # Project instructions + spec context (auto-loaded)
└── .specs/
├── 00-project.living.md # Main spec
└── feature-*/ # Feature specs (Option B)
├── requirements.md
├── design.md
└── tasks.md
Quick Reference
Greenfield: Questionnaire → Architecture approval → Building
Brownfield: Parallel analysis → Auto-populate → Verify → Building
Phase Transitions: Always require comprehension gate verification
Behavior Rules
- Never skip approach selection on first activation
- Always load workflow.md for core logic
- Greenfield: Block on questionnaire until complete
- Brownfield: Auto-populate, non-blocking verification
- Approval gates mandatory at phase transitions
- Comprehension gates verify understanding
- Use parallel agents for analysis
- Use EARS format for requirements
- Add spec context directly to CLAUDE.md (do NOT create separate steering files)
- Sub-agents must return text output only - do NOT let them write files. The main orchestrator writes all spec files to
.specs/
CLAUDE.md Spec Section Template
When setting up specs, add this to the project's CLAUDE.md:
## Specification
- **Main Spec:** `.specs/00-project.living.md`
- **Approach:** [A/B/C]
- **Phase:** [🔵 Planning | 🟢 Building | 🟡 Operating]
## Key Paths
[List important file paths]
## Development Guidelines
1. Before coding: Check spec for requirements
2. New features: Create feature spec in `.specs/feature-<name>/`
3. After changes: Update spec if architecture changes
## Tiered Approval
| Change Type | Approval |
|-------------|----------|
| Bug fixes, typos | Autonomous |
| New components, endpoints | Notify (update spec) |
| Architecture changes | Blocking (discuss first) |
Status Icons
| Icon | Meaning |
|---|
| ⬚ | Not started |
| 🔄 | In progress |
| ✅ | Complete |