| name | LLM Integration Patterns |
| description | How to identify and analyze LLM integration points across popular frameworks — OpenAI SDK, Anthropic SDK, LangChain, LlamaIndex, Vercel AI SDK |
| version | 1.0.0 |
LLM Integration Patterns
Purpose
Teaches the recon-agent how to find system prompts, tool definitions, guardrails, and data flows in codebases using different LLM frameworks.
When to Use
- During code reconnaissance (recon-agent)
- When the user asks "where are the LLM integration points?"
- When analyzing a new codebase for the first time
Framework Detection
| Framework | Detection Patterns |
|---|
| OpenAI | import openai, from openai, ChatCompletion, chat.completions.create |
| Anthropic | import Anthropic, from anthropic, messages.create, @anthropic-ai/sdk |
| LangChain | from langchain, ChatOpenAI, ConversationChain, AgentExecutor, @langchain |
| LlamaIndex | from llama_index, VectorStoreIndex, QueryEngine, ServiceContext |
| Vercel AI | import { streamText }, import { generateText }, useChat, from 'ai' |
Methodology
- Detect framework — Search for import patterns
- Find system prompts — Framework-specific locations (see reference files)
- Find tool definitions — Framework-specific registration patterns
- Find guardrails — Content filters, validators, middleware
- Trace data flows — User input → prompt → LLM → output
Reference Files