Agent Topology

A Claude Code skill that inspects multi-agent orchestration code — LangGraph, AutoGen, or custom routing loops — and generates production-ready Mermaid.js system architecture diagrams directly inside your editor, without any external tooling.

flowchart LR
    subgraph skill [Agent Topology Skill]
        A([Python orchestration code]) --> B[[agent-topology]]
        B --> C(Flow Diagram)
        B --> D(Sequence Diagram)
    end

Loading

---

What it does

Point the skill at any Python file or directory that contains agent definitions, routers, or workflow graphs and it will:

• Scan for nodes, edges, entry points, and state schemas across LangGraph, AutoGen, and custom patterns
• Output a flowchart showing the static topology (who routes to whom, under what conditions)
• Output a sequence diagram showing the runtime message-passing order
• Fall back to an inline SVG for graphs with more than 30 nodes
• Never touch your source files — strictly read-only

---

Mermaid rendering in VS Code

Native rendering (VS Code 1.90+)

Mermaid diagrams now render natively in VS Code's built-in Markdown preview, Notebooks, and Copilot Chat — no extension required. Simply open any .md file containing a ```mermaid ``` block and hit Ctrl+Shift+V / Cmd+Shift+V to preview.

This native support is also active in:

• GitHub Markdown files and pull request descriptions
• Obsidian (built-in renderer)
• GitLab Markdown preview
• Any Claude Code chat response

Critical: uninstall legacy Mermaid extensions

If you have installed a third-party Mermaid preview extension — especially the legacy standalone version by Matt Bierner (bierner.markdown-mermaid) — uninstall it before using this skill.

The old extension predates VS Code's native Mermaid support and conflicts with it directly, causing:

• Diagrams that flash or fail to render on first open
• Double-rendering artefacts (the native renderer and the extension both fire)
• Blank preview panes when diagrams contain newer Mermaid syntax (flowchart, sequenceDiagram with loop/alt blocks)

To uninstall:

1. Open the Extensions panel (Ctrl+Shift+X / Cmd+Shift+X)
2. Search for Markdown Preview Mermaid Support
3. Click Uninstall
4. Reload VS Code

Once removed, the native renderer takes over automatically.

---

Installation

Clone this repository directly into your global Claude skills directory:

git clone https://github.com/SoniaMehta14/agent-topology.git \
  ~/.claude/skills/agent-topology

That's it. Claude Code discovers skills in ~/.claude/skills/ automatically — no config file edits, no restart required.

Verify installation

In any Claude Code session (terminal or VS Code), type:

/agent-topology

Claude will load the skill and prompt you for a target file or directory.

---

Usage

Slash command

/agent-topology

Claude will ask which file or directory to analyze. You can also pass a target inline:

/agent-topology src/agents/

Natural language triggers

The skill auto-loads when your message matches any of these phrases:

Phrase	Example
map the architecture	"map the architecture of my LangGraph pipeline"
visualize agents	"visualize the agents in research_graph.py"
trace system routing	"trace the system routing in my AutoGen setup"
diagram the workflow	"diagram the workflow in src/orchestrator.py"
show agent topology	"show the agent topology for this project"
draw the agent graph	"draw the agent graph for the coordinator module"
map data flow	"map data flow between my workers"

---

Supported frameworks

Framework	Patterns detected
LangGraph	StateGraph, add_node, add_edge, add_conditional_edges, set_entry_point, END
AutoGen	AssistantAgent, UserProxyAgent, GroupChat, GroupChatManager, initiate_chat
Custom	@node decorators, class *Agent, def *_agent(), return {"next": ...} router patterns

---

Example output

Given a LangGraph research pipeline with a search → draft → critique loop:

graph TD
    accTitle: Agent Topology Mapping Flow
    accDescr: A flowchart showing the routing between a supervisor and specialized worker agents with a validation memory loop.

    User([User Prompt]) --> Supervisor[Orchestrator / Supervisor]
    
    subgraph Multi-Agent Workspace
        Supervisor -->|Classify Intent| IntentClassifier{Router}
        IntentClassifier -->|Billing Query| WorkerA[Billing Agent]
        IntentClassifier -->|Technical Support| WorkerB[Technical Agent]
        IntentClassifier -->|Refund Request| WorkerC[Refund Processor]
        
        WorkerA --> Memory[State / Short-Term Memory]
        WorkerB --> Memory
        WorkerC --> Memory
        
        Memory -->|Validation Loop| Supervisor
    end

    Supervisor --> FinalResponse([Final Action / Response])

Loading

---

Repository structure

agent-topology/
├── README.md                   ← this file
└── agent-topology/
    └── SKILL.md                ← skill definition (frontmatter + instructions)

The nested agent-topology/ directory matches Claude Code's skill discovery convention: skills are loaded from <skills-root>/<skill-name>/SKILL.md.

---

License

MIT