How samtusick/teamproject6_ai_agent works
This repository is an unfinished meeting-transcript action-item agent. Its documented workflow is to read a transcript, classify candidate items with an NRP model, verify quoted evidence, request human approval, then write approved items through an MCP filesystem server and read them back. The source files contain TODOs rather than those implementations, so the graph shows documented intended responsibilities—not verified runtime calls. Live meetings, external task/calendar integrations, and cross-meeting memory are explicitly out of scope.
Main entry points: agent.py Workflow Agent · mcp_client.py MCP Client · llm_client.py Item Classifier · output Approved Items
Components 10
Action Workflow
- Workflow Agent
agent.py - Transcript File
- Human Approval
agent.py - Approved Items
output
Agent Components
- Item Classifier
llm_client.py - Quote Guardrail
verify.py - MCP Client
mcp_client.py
External Integrations
- NRP Model
- Filesystem MCP Server
Other components
- Transcript User
Connections 11
- Transcript User → Workflow Agent: provides transcript
- Workflow Agent → Transcript File: reads
- Workflow Agent → Item Classifier: requests classification
- Item Classifier → NRP Model: classifies items
- Workflow Agent → Quote Guardrail: checks quotes
- Workflow Agent → Human Approval: requests decision
- Human Approval → Workflow Agent: returns decisions
- Workflow Agent → MCP Client: writes approved items
- MCP Client → Filesystem MCP Server: calls tools
- MCP Client → Filesystem MCP Server: reads back
- Filesystem MCP Server → Approved Items: stores items