Agent Loop and Task Dispatch Guide

A simple way to perform tasks through the Agent Loop (run_agent_loop.bat) based on GitHub issues.

Workflow

1. Register Issue: Register the problem or task on GitHub as an issue. (e.g., #123 Login error analysis)
2. Request Task Creation: Ask the AI agent in the chat as follows:

   "Based on GitHub issue #123, create a task document in giipdb/dispatch. Set the role to Developer."

3. Agent Action: The agent reads the issue content and creates a giipdb/dispatch/TASK_[Date]_[ID].md file.
4. Automatic Execution: The currently running Agent Loop (run_agent_loop.bat) detects the new file and starts the task.

Batch Task Distribution (Using Orchestrator)

If there are multiple issues, you can delegate the overall management to the Orchestrator.

• Example Request:

  "Orchestrator, check all currently open GitHub issues and distribute them as giipdb/dispatch task files to appropriate roles (Developer, Error Analyst, etc.)."

• Effect: The Orchestrator analyzes the issues, breaks them down into independent task files, and creates files by assigning the most suitable roles.

Key Changes

• Task Path: .agent/dispatch (Hidden) → giipdb/dispatch (Accessible top-level folder)
• Synchronization Method: Simplified to a method where the user requests the agent as needed, instead of complex automatic scripts.

Help

You can also manually write tasks by referring to the giipdb/dispatch/TASK_TEMPLATE.md file. The agent picks up the task when its Status is Pending.

Troubleshooting

Symptom	Cause	Solution
Task file created but the agent does not pick it up	The task Status is not Pending	Set the task file's Status to Pending
New task is not detected	run_agent_loop.bat is not currently running	Start the Agent Loop first, then create the task
A created task is ignored	File placed in the old .agent/dispatch (hidden) path	Create the task file under giipdb/dispatch
Orchestrator assigns an unsuitable role	The issue content is ambiguous	Specify the role (Developer, Error Analyst, etc.) in your request