spec-writer（规范编写者）

Specification Writer

This skill provides guidance for writing high-quality technical specifications that communicate complex systems, design decisions, and implementation strategies effectively.

Core Capabilities

This skill supports:

1. System Architecture Specs - Document system design, component interactions, data flow
2. Feature Specifications - Define requirements, user stories, acceptance criteria
3. Technical Proposals - Present solutions, trade-offs, implementation strategies
4. Design Documents - Capture design decisions, rationale, alternatives considered
5. Process Documentation - Describe workflows, procedures, automation pipelines
6. Mapping Specifications - Document conversion strategies between systems

When to Use This Skill

Use this skill when:

• Writing architectural decision records (ADRs)
• Creating system design documents
• Proposing new features or enhancements
• Documenting complex integrations or migrations
• Explaining technical trade-offs to stakeholders
• Creating implementation roadmaps
• Mapping between different systems or frameworks

Specification Structure Template

Frontmatter (YAML/Metadata)

Every specification should begin with structured metadata:

---
title: Descriptive Title
version: 1.0.0
date: YYYY-MM-DD
status: draft | review | approved | deprecated
authors:
  - Name/Entity
categories:
  - category1
  - category2
related:
  - path/to/related/doc1
  - path/to/related/doc2
---

Purpose: Makes specs discoverable, trackable, and navigable.

Document Sections

1. Executive Summary

Length: 2-4 paragraphs Purpose: High-level overview for decision-makers

Include:

• What is being proposed/documented
• Why it matters (business/technical value)
• Key outcomes or decisions
• Critical success factors

2. Problem Statement / Context

Length: 1-3 sections Purpose: Establish shared understanding of the problem

Include:

• Current state analysis
• Pain points or limitations
• User impact or business drivers
• Constraints or requirements

3. Proposed Solution / Design

Length: Variable (main body of spec) Purpose: Detailed technical approach

Include:

• High-level architecture diagram (mermaid/ASCII)
• Component breakdown with responsibilities
• Data flow and interactions
• Key design decisions with rationale
• Technology choices (with alternatives considered)

4. Implementation Plan

Length: 1-2 sections Purpose: Actionable roadmap

Include:

• Phased approach (Phase 1, 2, 3...)
• Dependencies and prerequisites
• Timeline estimates
• Risk mitigation strategies
• Success metrics

5. Trade-offs and Limitations

Length: 1 section Purpose: Honest assessment of constraints

Include:

• What this solution doesn't address
• Known limitations or caveats
• Performance/scalability considerations
• Security implications
• Cost analysis (time, resources, complexity)

6. Alternatives Considered

Length: 1 section Purpose: Show due diligence

Include:

• Alternative approaches (2-3)
• Why each was rejected
• Comparative analysis when relevant

7. Open Questions / Future Work

Length: Short list Purpose: Surface unknowns and next steps

Include:

• Unresolved technical questions
• Areas requiring more research
• Future enhancements out of scope
• Follow-up specifications needed

8. References

Length: List Purpose: Citations and further reading

Include:

• Related documentation
• External resources
• API documentation
• Research papers or blog posts

Writing Style Guidelines

Use Imperative/Infinitive Form

Good: "To accomplish X, configure Y" Bad: "You should configure Y to accomplish X"

Good: "Process files in batches" Bad: "The system will process files in batches"

Rationale: Direct, action-oriented language is clearer for both humans and AI agents.

Be Specific and Concrete

Vague: "The system should be fast" Specific: "Process 1000 requests/sec with <100ms p99 latency"

Vague: "Use appropriate error handling" Specific: "Retry with exponential backoff (3 attempts, 2x multiplier)"

Show, Don't Just Tell

Include:

• Code examples for key interfaces
• Diagram for complex flows
• Sample input/output for transformations
• Before/after comparisons for changes

Layer Information Progressively

Structure from high-level to low-level:

1. Abstract overview
2. Component-level design
3. Implementation details
4. Edge cases and optimizations

Readers can stop at appropriate depth for their needs.

Diagrams and Visualizations

Mermaid Diagrams

Use mermaid for:

• Flowcharts (graph TB)
• Sequence diagrams (sequenceDiagram)
• State machines (stateDiagram-v2)
• Entity relationships (erDiagram)

Example:

graph TB
    A[Input] --> B{Validate?}
    B -->|Valid| C[Process]
    B -->|Invalid| D[Error]
    C --> E[Output]

ASCII Diagrams

Use for simple structures:

Input → [Validator] → [Processor] → [Storage]
            ↓              ↓
        [Logger]      [Metrics]

Architecture Patterns

Use standard patterns when applicable:

• Pipes and Filters (Unix philosophy)
• Event-Driven Architecture
• Layered Architecture
• Microservices
• MVC/MVP/MVVM

Specification Types and Templates

1. System Architecture Spec

Focus on:

• Component boundaries
• Data flow and state management
• API contracts
• Scalability and performance
• Failure modes and recovery

2. Feature Specification

Focus on:

• User stories and use cases
• Acceptance criteria
• UI/UX considerations
• API changes
• Testing strategy

3. Migration/Mapping Spec

Focus on:

• Current state → Target state
• Conversion strategy (1:1, hybrid, workaround)
• Fidelity assessment (what's preserved, what's lost)
• Validation and testing
• Rollback plan

4. Integration Spec

Focus on:

• External systems and protocols
• Authentication and authorization
• Rate limiting and quotas
• Error handling and retries
• Monitoring and alerting

5. Process/Workflow Spec

Focus on:

• Step-by-step procedures
• Input/output for each step
• Automation opportunities
• Error recovery
• Metrics and success criteria

Best Practices

1. Start with "Why"

Always explain the motivation before diving into "how". Readers need context to evaluate solutions.

2. Use Consistent Terminology

Define terms once, use consistently throughout. Create a glossary for complex domains.

3. Be Honest About Trade-offs

Every design has limitations. Acknowledging them builds credibility and helps stakeholders make informed decisions.

4. Include Success Metrics

Define measurable criteria for success:

• Performance targets
• Quality metrics
• User satisfaction indicators
• Business outcomes

5. Version and Date Specs

Technical decisions evolve. Timestamps and versions help readers understand when decisions were made and which is canonical.

6. Make Specs Searchable

Use:

• Clear headings (H1, H2, H3)
• Keywords in frontmatter
• Descriptive filenames
• Internal links for cross-references

7. Keep Specs Living Documents

Update specs when:

• Implementation reveals new insights
• Requirements change
• Alternatives are discovered
• Lessons are learned post-launch

Mark deprecated sections clearly.

Anti-Patterns to Avoid

❌ Vague Requirements

Bad: "The system should be secure and fast" Good: "Encrypt data at rest (AES-256) and in transit (TLS 1.3). Support 10k req/s with <200ms p95 latency."

❌ Implementation Without Rationale

Bad: "Use Redis for caching" Good: "Use Redis for caching because it provides atomic operations, pub/sub, and persistence options. Considered Memcached (faster but lacks persistence) and Hazelcast (better for distributed scenarios but adds complexity)."

❌ Assuming Knowledge

Bad: "Configure the usual K8s setup" Good: "Deploy as StatefulSet with 3 replicas, PersistentVolumeClaim per pod, headless service for DNS-based discovery."

❌ No Examples

Provide concrete examples for:

• API requests/responses
• Configuration files
• Command-line usage
• File formats

❌ Ignoring the Reader

Consider your audience:

• Engineers: Need implementation details
• Architects: Need system-level view
• Product: Need feature/benefit explanation
• Executives: Need ROI and timeline

Write sections for each or use progressive disclosure.

Templates

Quick Spec Template

---
title: [Feature/System Name]
version: 1.0.0
date: YYYY-MM-DD
status: draft
---

# [Title]

## Summary
[2-3 sentences: what, why, impact]

## Problem
[Current pain point or opportunity]

## Solution
[High-level approach]

## Implementation
1. Phase 1: [description]
2. Phase 2: [description]
3. Phase 3: [description]

## Trade-offs
- Pro: [benefit]
- Con: [limitation]

## Success Metrics
- Metric 1: [target]
- Metric 2: [target]

Architecture Spec Template

---
title: [System Name] Architecture
version: 1.0.0
date: YYYY-MM-DD
---

# [System Name] Architecture

## Overview
[2-3 paragraphs: purpose, scope, key components]

## Architecture Diagram
[mermaid or ASCII diagram]

## Components

### Component A
- **Responsibility**: [what it does]
- **Interface**: [API/contracts]
- **Dependencies**: [what it needs]

### Component B
[repeat structure]

## Data Flow
[Describe end-to-end flow]

## Key Design Decisions

### Decision 1: [title]
- **Context**: [why this decision needed]
- **Chosen**: [selected approach]
- **Alternatives**: [what was considered]
- **Rationale**: [why chosen over alternatives]

## Non-Functional Requirements
- **Performance**: [targets]
- **Scalability**: [growth plan]
- **Reliability**: [uptime, recovery]
- **Security**: [auth, encryption, compliance]

## Implementation Roadmap
[Phased plan with timeline]

## Risks and Mitigations
[Known risks with mitigation strategies]

Mapping Spec Template

---
title: [Source] → [Target] Mapping
version: 1.0.0
date: YYYY-MM-DD
---

# [Source] to [Target] Mapping Specification

## Executive Summary
[Viability, strategy, fidelity]

## Current State (Source)
[Structure, features, examples]

## Target State
[Target structure, constraints, capabilities]

## Mapping Strategy
[Direct, hybrid, workaround]

## Conversion Process

### Phase 1: [name]
[Steps, tools, validation]

### Phase 2: [name]
[Steps, tools, validation]

## Fidelity Analysis

### High Fidelity (>90%)
- Feature X maps directly
- Feature Y has equivalent

### Medium Fidelity (70-90%)
- Feature Z requires adaptation

### Low Fidelity (<70%)
- Feature W not supported

## Limitations
[What can't be mapped and why]

## Validation
[How to verify successful mapping]

## Automation
[Scripts, tools, continuous sync]

Examples

Example 1: Architecture Spec

See: specs/cursor-skills-mapping.md

• Clear viability assessment
• Detailed component mapping
• Trade-offs documented
• Implementation phases defined
• Success metrics included

Example 2: Process Spec

See: specs/enhanced-skill-generator-pipeline.md

• Unix philosophy applied
• Atomic composability explained
• Pipeline visualized with mermaid
• Multiple usage examples provided
• Success criteria enumerated

Example 3: Feature Spec

When documenting a new feature:

1. User stories (what users want to do)
2. Acceptance criteria (how to verify)
3. API/interface changes
4. Migration path for existing users
5. Testing strategy

Integration with Skills

This skill complements:

• architecture - For system design analysis
• skill-creator - For skill specification documents
• document-skills - For formatting and templates
• mcp-builder - For MCP server specs

Validation Checklist

Before finalizing a spec, verify:

• [ ] Frontmatter complete with title, version, date, status
• [ ] Executive summary present (2-4 paragraphs)
• [ ] Problem/context clearly stated
• [ ] Solution explained with diagrams
• [ ] Implementation plan with phases
• [ ] Trade-offs and limitations documented
• [ ] Alternatives considered section included
• [ ] Success metrics defined
• [ ] References provided
• [ ] Consistent terminology throughout
• [ ] Code examples for key interfaces
• [ ] Diagrams for complex flows
• [ ] Actionable for intended audience

References

• Technical Writing Guide
• Architecture Decision Records
• Markdown Style Guide
• Mermaid Documentation

Quick Reference

Command: Write clear, structured specifications Output: Actionable technical documentation Audience: Engineers, architects, stakeholders Format: Markdown with YAML frontmatter Tools: Diagrams (mermaid), examples, templates