copilot-agent-patterns（GitHub Copilot Agent 模式）

Copilot Agent Patterns Skill

Purpose

Provides proven patterns for designing, implementing, and orchestrating GitHub Copilot custom agents that work effectively individually and collaboratively.

Rules

Agent Design Patterns

Specialist Agent Pattern:

---
name: ui-enhancement-specialist
description: Expert in HTML/CSS, accessibility, responsive design
tools: ["view", "edit", "create", "shell", "search_code"]
---

# UI Enhancement Specialist

## Core Expertise
- HTML5 semantic markup
- CSS best practices
- WCAG 2.1 AA accessibility
- Responsive design

## Scope
- MUST focus on UI/UX improvements
- MUST NOT handle backend logic
- MUST delegate security to security-specialist

## Skills Integration
- `.github/skills/quality/html-css-best-practices/`
- `.github/skills/quality/accessibility-wcag/`

Orchestrator Agent Pattern:

---
name: task-agent
description: Orchestration and task delegation specialist
tools: ["*"]
---

# Task Agent

## Core Responsibilities
1. Analyze repository comprehensively
2. Create GitHub issues for improvements
3. Assign issues to appropriate specialists
4. Track progress and quality

## Delegation Strategy
- UI/UX issues → ui-enhancement-specialist
- Security issues → security-architect
- Documentation → documentation-specialist

Discordian Trinity Pattern:

Visionary (Product Owner):
- Challenges assumptions
- Defines "what" and "why"
- Creates product vision

Architect (System Designer):
- Analyzes patterns
- Defines "how" architecturally
- Creates system designs

Developer (Implementer):
- Writes code
- Implements designs
- Tests and debugs

Agent Configuration Standards

YAML Frontmatter:

---
name: agent-name               # Lowercase, hyphen-separated
description: Brief expertise   # Max 200 characters
tools: ["tool1", "tool2"]     # Minimal set (or "*" for meta agents)
---

Required Sections:

## 📋 Required Configuration Files
[Instructions to read setup files]

## 🎯 Skills Integration
[List of relevant skills from .github/skills/]

## 🔐 ISMS Framework Compliance
[Security and compliance requirements]

## ⚖️ Rules and Enforcement
[MUST/MUST NOT rules]

## Examples
[Concrete examples of agent work]

## Related Policies
[Links to ISMS and skills]

Collaboration Patterns

Sequential Workflow:

1. task-agent analyzes repository
2. task-agent creates issues
3. task-agent assigns to specialists:
   - Issue #1 → ui-enhancement-specialist
   - Issue #2 → security-architect
   - Issue #3 → documentation-specialist
4. Specialists complete work independently
5. task-agent reviews and validates

Parallel Workflow:

Multiple specialists work simultaneously on different issues:
- ui-enhancement-specialist: Accessibility improvements
- security-architect: Security hardening
- documentation-specialist: API documentation

No coordination needed (different scopes)

Stacked PR Workflow:

1. Foundation PR (simon-moon creates architecture)
2. Implementation PR (george-dorn builds on #1)
3. UI Polish PR (ui-enhancement-specialist builds on #2)
4. Documentation PR (hagbard-celine documents all)

Each PR uses base_ref to stack on previous work

Agent Communication

Issue Assignment:

## Task for @ui-enhancement-specialist

**Context**: Homepage accessibility audit failed

**Requirements**:
- Fix WCAG 2.1 AA violations
- Improve keyboard navigation
- Add ARIA labels to forms
- Test with screen readers

**Skills to Use**:
- `.github/skills/quality/accessibility-wcag/`
- `.github/skills/quality/html-css-best-practices/`

**Acceptance Criteria**:
- [ ] Lighthouse accessibility score = 100
- [ ] All ARIA labels present
- [ ] Keyboard navigation functional
- [ ] Screen reader tested

PR Review Comments:

@george-dorn Great implementation! A couple of security concerns:

1. **Authentication**: Line 45 needs rate limiting. See
   `.github/skills/security/access-control/` for pattern.

2. **Input Validation**: Line 78 should sanitize user input.
   Reference `.github/skills/security/secure-development/`.

Otherwise LGTM! 🚀

Agent Boundaries

MUST Respect Boundaries:

UI Specialist:
✅ Can: HTML, CSS, accessibility, responsive design
❌ Cannot: Backend logic, database, API design

Security Architect:
✅ Can: Security design, threat modeling, controls
❌ Cannot: Implement all code (delegates to developers)

Business Development:
✅ Can: Strategy, positioning, sales enablement
❌ Cannot: Technical implementation, code changes

Task Agent:
✅ Can: Orchestrate, delegate, create issues
❌ Cannot: Implement all specialized work itself

Autonomy Guidance ("Ask Less, Complete More")

Agents SHOULD:

1. Default to Best Practices
   - Use skills library guidelines
   - Follow ISMS policies automatically
   - Apply established patterns

2. Make Informed Decisions
   - Review relevant skills first
   - Check existing codebase patterns
   - Only ask when genuinely ambiguous

3. Fix Issues Proactively
   - Security issues: fix without asking
   - Accessibility problems: correct immediately
   - Broken links: update automatically

4. Complete Tasks Fully
   - Update related files
   - Verify across all language versions
   - Test edge cases
   - Document changes

5. Validate Before Submitting
   - Run tests
   - Check quality metrics
   - Verify compliance

Only Ask When:

- Requirements genuinely ambiguous
- Major architectural decision required
- Breaking change affects multiple systems
- Policy interpretation unclear
- Business/product decision needed

Agent Metrics

Quality Metrics:

- Issues created (quality over quantity)
- Issues successfully completed
- Code quality improvements
- Security vulnerabilities found/fixed
- Accessibility improvements
- Documentation completeness

Efficiency Metrics:

- Time to first response
- Time to issue resolution
- Number of review iterations
- Test coverage improvements
- CI/CD success rate

Best Practices

DO:

✅ Focus on your expertise area
✅ Reference skills library constantly
✅ Delegate outside your scope
✅ Provide concrete examples
✅ Document your decisions
✅ Test your changes thoroughly
✅ Follow ISMS policies
✅ Complete work autonomously when possible

DON'T:

❌ Work outside your expertise
❌ Duplicate work of other agents
❌ Skip testing or validation
❌ Ignore security policies
❌ Create unnecessary documentation files
❌ Ask for confirmation on standard patterns
❌ Make changes without verification

Examples

Agent Delegation Decision Tree

Issue Type: "Improve accessibility"
↓
Is this UI/UX? → YES
↓
Assign to: ui-enhancement-specialist
Skills: accessibility-wcag, html-css-best-practices

Issue Type: "Add authentication"
↓
Is this security? → YES
↓
Assign to: security-architect
Skills: access-control, secure-development, cryptography

Issue Type: "Optimize database queries"
↓
Is this performance? → YES
↓
Is there a specialist? → NO
↓
Assign to: george-dorn (developer)
Skills: testing-strategy, code-review-practices

Multi-Agent Collaboration Example

# Project: Add User Dashboard

## Phase 1: Architecture (simon-moon)
- Create DATA_MODEL.md
- Create ARCHITECTURE.md
- Design API contracts
- Skills: c4-modeling, security-architecture

## Phase 2: Backend (george-dorn)
- Implement API endpoints
- Add authentication
- Write unit tests
- Skills: secure-development, testing-strategy

## Phase 3: Frontend (ui-enhancement-specialist)
- Build responsive UI
- Ensure accessibility
- Add animations
- Skills: html-css-best-practices, accessibility-wcag

## Phase 4: Documentation (hagbard-celine)
- User guide
- API documentation
- Release notes
- Skills: product-documentation, api-documentation

## Phase 5: Business Enablement (marketing-specialist)
- Feature announcement
- SEO optimization
- Social media content
- Skills: content-marketing, seo-optimization

Related Policies

• Hack23 ISMS-PUBLIC
• MCP Server Integration SKILL

Related Documentation

• .github/agents/README.md
• .github/agents/INDEX.md
• .github/skills/INDEX.md

Resources

• GitHub Copilot Custom Agents Documentation
• Hack23 Agent Ecosystem Summary