GitHub Copilot Agents
Six specialized agents, each excelling at specific tasks.
Overview
| Agent | Specialization | Key Capability |
|---|---|---|
| @planner | Planning | Read-only analysis and planning |
| @orchestrator | Coordination | Multi-phase workflows |
| @codebase | Development | Multi-language implementation |
| @docs | Documentation | README, API docs, guides |
| @review | Quality | Security, performance, best practices |
| @em-advisor | Leadership | Strategy, team dynamics |
When to Use Each Agent
- @planner - Analyzes, plans, identifies risks (no code edits)
- @codebase - Writes code, runs tests, validates (multi-language)
- @orchestrator - Manages phases, delegates to specialists
- @review - Security audits, performance checks, best practices
- @docs - README, API docs, user guides
- @em-advisor - 1-on-1 prep, strategy, team issues
Agent Capabilities
Read & Search
All agents can:
- Read code and documentation
- Search by filename and content
- Analyze codebase patterns
- Find all usages of functions/classes
Edit & Create
Only some agents can:
- @codebase - Full edit access
- @orchestrator - Full edit access
- @docs - Documentation edit access
- @review - Read-only (no edits)
- @planner - Read-only (no edits)
- @em-advisor - Read-only (no edits)
Execute
Some agents can run commands:
- @codebase - Run tests, builds, linters
- @orchestrator - Run tests, builds, terminal commands
External Resources
All agents can:
- Fetch documentation from web
- Look up best practices
- Reference external APIs
Agent Workflow
Phase 1: Planning
Agent analyzes your request and proposes a plan.
Phase 2: Approval
You review and approve before any changes.
Phase 3: Execution
Agent implements step-by-step with validation.
Phase 4: Handoff
Agent suggests next steps (review, document, etc.).
Quick Examples
Example 1: Build Authentication
@orchestrator Build JWT authentication with:
- Login/logout endpoints
- Refresh tokens
- Rate limiting
- Comprehensive tests
- Security review
- API documentation
Flow:
- Plan created → You approve
- @codebase implements endpoints
- @review audits for security
- @codebase addresses findings
- @docs creates API docs
- Done!
Example 2: Fix a Bug
@codebase Fix: Users can't update their profile picture
- Identify root cause
- Implement fix
- Add tests to prevent regression
- Validate with existing tests
Example 3: Code Review
@review Audit the payment processing module for:
- Security vulnerabilities
- Performance issues
- Best practices violations
Profile Auto-Detection
@codebase automatically detects your project type:
| Detection | Language | Adaptations |
|---|---|---|
*.sln, *.csproj |
.NET | Clean Architecture, async/await, nullable types |
pyproject.toml, requirements.txt |
Python | Virtual env, type hints, pytest |
package.json, tsconfig.json |
TypeScript | Strict types, incremental builds |
| Mixed or unclear | Generic | Language-agnostic patterns |
Tip: Mention your language if detection is ambiguous:
@codebase Using TypeScript with React, create...
Agent Handoffs
Agents can hand off to each other seamlessly:
@codebase (implement)
↓ handoff
@review (audit)
↓ handoff
@docs (document)
Each handoff includes context, so the next agent understands the full picture.
Context Persistence
Agents automatically save important decisions to .github/copilot-instructions.md:
- Architectural patterns established
- Coding standards developed
- Project-specific conventions
- Technical constraints
This context is reused in all future sessions!
Next Steps
- View Workflows - See real-world examples
- Customization - Adapt agents to your project
- Getting Started - Quick setup guide