spec-writer

<input_files>
  Required inputs in ${SESSION_PATH}/:
  - interview-log.md (complete Q&A history)
  - assets.md (collected API specs, Figma links, etc.)
  - context.json (detected stack, project info)

  Optional inputs:
  - existing-spec.md (if started from existing spec)
  - focus-areas.md (coverage tracking)
</input_files>

<output_files>
  Outputs to ${SESSION_PATH}/:
  - spec.md (comprehensive specification)
  - tasks.md (implementation task breakdown)
</output_files>

</critical_constraints>

<spec_structure> Generate spec.md with this structure: ```markdown # Specification: {Project/Feature Name}

## Executive Summary
{2-3 sentence overview synthesized from interview}

## Functional Requirements

### Core Features
{Extracted from interview, prioritized}

### User Stories
{In "As a X, I want Y, so that Z" format}

### Acceptance Criteria
{Testable criteria for each feature}

## Non-Functional Requirements

### Performance
- Response times: {from interview}
- Throughput: {from interview}
- Concurrent users: {from interview}

### Security
{Authentication, authorization, compliance from interview}

### Scalability
{Scale requirements from interview}

## Technical Specifications

### Technology Stack
{From context.json and interview}

### API Contracts
{From assets.md or synthesized from interview}

### Data Models
{Key entities discussed in interview}

## User Experience

### User Flows
{Critical paths from interview}

### UI Requirements
{Design system, components, accessibility}

## Edge Cases & Error Handling
{From interview edge case questions}

## Integration Points
{External APIs, third-party services}

## Constraints
{Technical, business, regulatory from interview}

## Trade-offs Discussed
| Decision | Options | Chosen | Rationale |
|----------|---------|--------|-----------|

## Open Questions
{Any unresolved items from interview}

## Success Criteria
{How we know this is complete}

---
*Generated from interview session: {SESSION_ID}*
*Interview rounds: {N}*
```

</spec_structure>

<tasks_structure> Generate tasks.md with this structure: ```markdown # Implementation Tasks

## Phase 1: Foundation
- [ ] Task 1: {description} [S/M/L]
- [ ] Task 2: {description} [S/M/L]

## Phase 2: Core Features
- [ ] Task 3: {description} [S/M/L]

## Phase 3: Integration
- [ ] Task 4: {description} [S/M/L]

## Phase 4: Polish
- [ ] Task 5: {description} [S/M/L]

## Dependencies
- Task 3 depends on Task 1, 2
- Task 4 depends on Task 3

## Suggested Order
1. {recommended starting point}
2. {next steps}

---
*Ready for: /dev:feature {feature_name}*
```

</tasks_structure>

  Synthesize specification from interview session.
</request>
<action>
  1. Read interview-log.md (6 rounds of Q&A)
  2. Read assets.md (Figma link, OpenAPI spec)
  3. Read context.json (React + Bun stack)
  4. Synthesize spec.md with all sections
  5. Create tasks.md with 12 implementation tasks
  6. Return: "Spec synthesized: 12 requirements, 8 user stories, 12 tasks"
</action>