Implement agent workflow system with four chained custom agents

[Copilot]

Created a complete agent workflow system for development task orchestration: develop → test → document → review with feedback iteration support.

Agent Definitions

Four specialized custom GitHub Copilot agents in .github/agents/:

• develop-agent - Code implementation
• test-agent - Independent validation
• document-agent - Documentation generation
• review-agent - Quality gate with approval/iteration control

Each agent defines: role, workflow position, instructions, and structured output format for handoffs.

Workflow Infrastructure

• workflow-orchestrator.md - Agent handover protocols for custom agents
• workflow-example.md - End-to-end scenario walkthrough
• workflow-architecture.md - Visual diagrams and data flow

Architecture

Workflow is orchestrated through custom GitHub Copilot agents stored in .github/agents/. These are custom agent definitions that are invoked through GitHub Copilot's custom agent system, not standard GitHub @mentions.

Each custom agent completes its work and provides context for the next agent in the chain:

Custom agent: develop-agent → implement feature
Custom agent: test-agent → test the implementation
Custom agent: document-agent → document the feature
Custom agent: review-agent → review all work

Important: These agents are invoked through GitHub Copilot's custom agent system. The @agent-name syntax used in documentation examples is shorthand representing these custom agents, but they must be invoked through GitHub Copilot, not as standard GitHub @mentions in issues or PRs.

Review agent closes loop: approved workflows complete, change requests restart from develop-agent with specific feedback.

All agents are placeholder implementations ready for concrete task definitions.

Original prompt

Create me serveral custom-agents and help me chain them together. Eventually have a full workflow of develop -> test -> document -> review

• the develop agent will do any kind of developement, this will be more concrete later, for now it is just a placeholder
• the test will test what the develop-agent build but independently, placehodler
• the document wil write documentation, placeholder
• the review-agent reviews and instructs the develop-agent if something is off. also provide a summary

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[UncleBats]

@copilot why did you write the shell script, cant we orchestrate this using handovers

[Copilot]

@copilot why did you write the shell script, cant we orchestrate this using handovers

You're absolutely right! I've removed the shell script and updated the workflow to use proper agent handovers instead.

The workflow now orchestrates through @mentions where each agent explicitly hands off to the next agent with context:

@develop-agent implement feature
  ↓ completes & hands off
@test-agent test the implementation
  ↓ completes & hands off
@document-agent document the feature
  ↓ completes & hands off
@review-agent review all work

See the updated workflow-orchestrator.md for detailed handover examples with context passing.

Commit: 8da0557

[UncleBats]

@copilot so when using the @ mentions now described, would that trigger by using github issues and github pr? thought it would only trigger when using the @copilot

[Copilot]

@copilot so when using the @ mentions now described, would that trigger by using github issues and github pr? thought it would only trigger when using the @copilot

Correct - these are custom GitHub Copilot agents, not standard GitHub @mentions. They won't work in regular GitHub issues or PRs.

These agent definitions in .github/agents/ are designed to be invoked through GitHub Copilot's custom agent system. The @agent-name syntax in the docs is shorthand, but you would actually invoke them as custom agents through GitHub Copilot (similar to how you use @copilot).

Updated all documentation to clarify this distinction.

Commit: bd6b7a5

[UncleBats]

@copilot can we do a test run of this flow or do we need to merge to main first

[Copilot]

@copilot can we do a test run of this flow or do we need to merge to main first

You can test the workflow right now on this branch! The custom agent definitions in .github/agents/ are available to use immediately - no need to merge to main first.

To test the flow:

1. You can invoke any of the custom agents defined in .github/agents/ (develop-agent, test-agent, document-agent, review-agent) through the GitHub Copilot agent system while on this branch

2. Try a simple test:

   • Invoke the develop-agent with a simple task
   • Have it hand off context to test-agent
   • Continue through the chain

The agents are currently placeholders, so they'll follow the instructions in their definition files but won't have specific implementation logic yet. You can add concrete tasks to make them more functional as needed.