llmops-demo-ts: Instructions file for Claude Code

llmops-demo-ts

TypeScript monorepo demonstrating LLMOps practices with LangGraph-based AI agents, Express backend, and Vue.js frontend.

Project Structure

packages/
  common/     — Shared types, interfaces, utilities (Logger, Agent, ChatRequest)
  agents/     — LangGraph AI agents (default, research, secure) with BaseAgent pattern
  backend/    — Express API with tsoa OpenAPI generation
  frontend/   — Vue.js 3 chat interface with Pinia state management

Key Commands

pnpm install          # Install dependencies
pnpm build            # Build all packages (common → agents → backend → frontend)
pnpm test             # Run all tests
pnpm lint             # Lint with trunk
pnpm start:backend    # Start backend dev server
pnpm start:frontend   # Start frontend dev server

Agent Architecture

Agents follow the pattern: BaseAgent → StateAnnotation → StateGraph → Nodes → ConditionalEdges

• BaseAgent (packages/agents/src/agents/baseAgent.ts): Abstract class implementing IAgent
• AgentFactory (packages/agents/src/agents/agentFactory.ts): Factory + cache for agent instantiation
• State: LangGraph Annotation.Root({}) for typed state management
• Nodes: Async functions that process state and return updates
• Tools: Declared as JSON schemas, dispatched via callTool nodes

Agent Team Configuration

This project uses a 3-tier agent architecture with a streamlined pipeline workflow:

Plan → Research (if needed) → Orchestrate (plan-only) → Parallel Execution

Agent Tiers

Tier 1 — Planning (read-only): Research and produce structured plans.

Agent	Role	Mode	Skill
planner	Task breakdown, roadmaps	Plan (read-only)	/plan-task
researcher	Library docs, APIs, specs	Plan (read-only)	/research
product-manager	Requirements, user stories	Plan (read-only)	/write-requirements