1. Source code for analysis and documentation is often very large. When an Agent (for example, GitHub Copilot) reads the source, common issues include:
1.1 Incomplete or inaccurate retrieval, and very fast context window overflow.
1.2 Agent outputs may contain inferred assumptions (even with careful prompting), without sufficient evidence from the source code.
1.3 Agents do not always clearly identify which parts require human review or approval.

2. A project usually involves multiple roles: PM, BA, DEV, and TEST. Each role has its own tasks and typically uses an Agent as an assistant.
3. Task management and task-status tracking are not handled in a unified way.
4. There is no shared, standardized context (accurate and complete) for all participating Agents.
5. There are no standardized output templates for all Agent-generated deliverables.

1. Provide complete and well-structured context so Agents can analyze and search source code effectively.
2. Provide a complete process for PM/BA/DEV roles participating in the project.
3. Provide step-by-step workflows for supporting sub-agents (PM/BA/DEV roles) and enforce process compliance.
4. Provide shared templates and unified context for all participating sub-agents.
5. Ensure all sub-agent outputs clearly mark areas that require human review and confirmation.

1. Use VS Code.
2. Install GitHub Copilot.
3. Install Ollama.