Content
# Developer Support System — Plan 
> **Note:** Built on **ADK (Agent Development Kit)**, focused on debugging a **Django + MySQL system**.
## 1) Purpose
Help users debug issues by providing details (screenshots, Django yellow error pages, HTML, images, or free‑form descriptions). The AI agent will understand the problem and follow a standard, step‑by‑step debugging procedure, delegating tasks to sub‑agents/tools as needed.
## 2) Scope (Current)
* **Agentic Framework:** ADK (core orchestration layer)
* **Target Framework for Debugging:** Django
* **Database:** MySQL
* **Context Inputs:**
* Screenshots
* Yellow error page text (raw HTML, image, or plain text)
* Free‑form problem details
* Links/URLs inside the app (when available)
## 3) Debugging Workflow (Existing)
1. **If a link is provided:**
* Extract file/function location from the link.
2. **If file or function path is known:**
* Read the function docstring (`__doc__`) first, then the full function code.
* If sub‑functions are referenced inside, read them too and understand their behavior.
* Use this to reconstruct the logic of the function for deeper understanding.
3. **Use database models when relevant:**
* Understand what data enters the function and where it goes.
4. **Understand the data flow:**
* Rely on any available documentation (acknowledging it may be limited).
## 4) Tools (Existing)
* **ADK as Orchestration Layer**
* Hosts the agent network; coordinates sub‑agents and tool calls.
* **GenAI Toolkit Toolbox**
* Connect to **MySQL** (initially **read‑only**; later, controlled write access will be enabled after verification).
* **Custom‑Built Tool (to be converted to ADK’s MCP Server)**
* Lists the different links in the system.
* Given a link for debugging, returns key details (URL, file path, name, etc.) with schema guaranteed for locating the code.
* **File/Function Reading Tools**
* Via **GitHub MCP** and/or a **custom code‑parsing tool** (both may be used; GitHub MCP is ideal but limited on alternate branches).
## 5) Assumptions & Constraints (As Stated)
* System documentation exists but is limited.
* The agent should prioritize understanding function logic and data flow from real code and models.
* MySQL: start with **read‑only access**. Controlled writes added later once reliability is proven.
* The ADK agent network is separate from Django, with only database + tool permissions for debugging.
* Links inspector guarantees correct schema for locating code.
## 6) Supporting Context (New)
* **Notion Documentation** will be provided, containing scenarios, problems, and mitigation strategies. This gives the agent context on system logic and expected flows.
* **Agentic Pipelines in ADK:** ADK will orchestrate multiple specialized agents working together in a pipeline.
* **Audit Trail (Future To‑Do):** Logging every debugging step is **very important**, but will be added after the system stabilizes.
## 7) Current Status Summary
* Planning a multi‑agent debugging flow using ADK as the orchestration framework.
* Flow starts from user-provided artifacts (links, error pages, screenshots, text) and drills into code, sub‑functions, and models to map data flow and logic.
* Tooling to resolve links → code locations and to read code is in scope.
* MySQL connectivity via GenAI toolkit is in scope (read‑only first).
* Notion docs and audit trail are part of the future roadmap.
