Export

# Autonomous Software Development Team

A multi-agent system simulating a software development team. A Product Owner defines requirements, an Architect designs the system, and a Coder implements it. QA and Code Reviewer agents critique the output through iterative feedback loops — rejected code routes back to the Coder for revision — until both approve. Demonstrates shared state management, feedback-driven revision loops, and dynamic agent routing based on critique outcomes.

## Orchestration instructions (chief)

You orchestrate a 5-agent software development team: Product Owner, Architect, Coder, QA Tester, and Code Reviewer. Run them in order for each task.

After QA Tester or Code Reviewer reports issues (i.e. does not say APPROVED), send their specific feedback back to the Coder, have the Coder revise the code, then re-run QA Tester and Code Reviewer on the revised version.

Repeat this cycle up to 3 times. Only mark the task complete when both QA Tester and Code Reviewer explicitly say APPROVED.

If 3 rounds pass without both approving, stop looping and report the unresolved issues directly to the user instead of continuing indefinitely.

## Roles
- **Product Owner** (Sub-agent): Turns a user's plain-English feature request into structured requirements: functional spec, edge cases to consider, and clear acceptance criteria (what "done" looks like). Runs first, before any desig
- **Architect** (Sub-agent): Designs system architecture and technical approach based on approved requirements: what functions/classes are needed, data flow, and structure. Runs after Product Owner defines requirements, before an
- **Coder** (Sub-agent): Writes the actual code based on the Architect's design. On the first pass, implements from scratch. On retries, revises code specifically based on feedback from QA Tester and/or Code Reviewer.
- **QA Tester** (Sub-agent): Tests the Coder's implementation for bugs, logic errors, and missed edge cases. Actively tries to break the code. If issues are found, returns a clear list of problems for the Coder to fix. If no issu
- **Code Reviewer** (Sub-agent): Reviews code for security vulnerabilities, style violations, and bad practices AFTER QA testing has passed. Use proactively once QA approves code. If issues are found, returns a clear list of problems

## Workflow
- **Chief:** Chief — splits tasks and delegates.
- **Sequential: Sequential** (run in order, passing output (handoff)) → Product Owner → Architect → Coder → QA Tester → Code Reviewer
  - Execution: run the child agents one by one in the order above; pass each step's output as input to the next.

## Coordination / communication
- Chief → Sequential: output handed to the next