AI Ticket Triage Agent — AI skill for Claude Code

AI Triage Agent

An AI-powered enterprise support ticket triage and incident investigation system built with **Python, FastAPI, Claude, and LangGraph**.

The application analyzes automation support tickets, classifies incidents, and uses an agentic workflow to gather additional evidence through tools before producing a response.

Project Overview

Traditional ticket classification systems make decisions using only the information provided in the ticket.

This project extends ticket classification into an **agentic incident investigation workflow**.

Instead of immediately generating a response, the AI agent can:

• Analyze an incoming support ticket
• Decide whether additional information is required
• Select an appropriate tool
• Generate tool arguments
• Execute the tool through the application
• Observe the tool result
• Continue reasoning using the new evidence
• Produce an evidence-based response

Current Architecture

Incoming Support Ticket
        |
        v
      FastAPI
        |
        v
   LangGraph Agent
        |
        v
     Claude LLM
        |
        v
  Need more evidence?
      /       \
    Yes        No
     |          |
     v          v
  Tool Node    END
     |
     v
Application Status Tool
     |
     v
 Tool Result
     |
     +--------> Agent

The graph allows the agent to follow a:

**Reason → Act → Observe → Reason**

execution cycle.

Current Features

• FastAPI REST API
• Pydantic data validation
• Claude LLM integration
• Structured ticket classification
• Severity classification
• LangGraph state management
• Conditional graph routing
• LLM tool selection
• Dynamic tool argument generation
• Application-status investigation tool
• Agent → Tool → Agent execution loop
• Evidence-based incident investigation
• Environment-variable based API key management

Severity Levels

The ticket classifier uses four severity levels:

**LOW** Enhancements, informational requests, future risks, or issues that automat