Framework detector — identify agent frameworks in codebase

[rodchalski]

Goal

Detect which AI agent frameworks are used in a scanned project. This feeds into the permission gap analyzer (issue #3).

Supported Frameworks (v1)

Detect via package.json, requirements.txt, pyproject.toml, go.mod, pom.xml, and import statements:

Framework	Package/Import	Language
LangChain	langchain, @langchain/*	Python, TS
LangGraph	langgraph, @langchain/langgraph	Python, TS
CrewAI	crewai	Python
OpenAI Agents SDK	openai-agents, openai/agents	Python
MCP	@modelcontextprotocol/*, mcp	TS, Python
Vercel AI SDK	ai, @ai-sdk/*	TS
Amazon Bedrock	@aws-sdk/client-bedrock*, boto3 (bedrock)	TS, Python
AutoGen	autogen, pyautogen	Python
Permission Protocol	@permission-protocol/sdk, permission-protocol	TS, Python

Output Interface

interface DetectionResult {
  frameworks: Array<{
    name: string;
    language: "typescript" | "python" | "java" | "go";
    version?: string;
    source: string; // e.g. "package.json", "requirements.txt"
  }>;
  ppInstalled: boolean; // Is Permission Protocol SDK already present?
  entryPoints: Array<{
    file: string;
    type: "agent" | "tool" | "prompt" | "mcp-server" | "mcp-client";
    framework: string;
  }>;
  files: {
    scanned: number;
    relevant: number;
  };
}

Detection Strategy

1. Manifest scan — Check dependency files for known packages
2. Import scan — Grep source files for framework imports to catch unlisted deps
3. Entry point identification — Find files that instantiate agents, define tools, or configure MCP servers

Acceptance Criteria

• Detects all 9 frameworks listed above
• Identifies whether PP SDK is already installed
• Finds agent/tool entry points with file paths
• Works on TypeScript and Python projects (minimum)
• Returns structured DetectionResult object
• Unit tests for each framework detection

Depends On

• Issue Repo scaffold + CLI package setup #1 (repo scaffold)