Add Anthropic SDK code sample demonstrating ASI03 privilege escalation

[DanielMendozaC]

Add Anthropic SDK code sample demonstrating ASI03 privilege escalation

Key Changes:

• New code sample demonstrating ASI03 (Identity & Privilege Abuse)
• Uses Anthropic Python SDK with tool calling
• Includes Dockerfile as required by CONTRIBUTING.md

Added:

• insecure_agent.py - Over privileged file management agent
• README.md - Vulnerability writeup, OWASP mapping, attack walkthrough, mitigations
• Dockerfile, requirements.txt, .env.example
• data/ - Test files including one with embedded prompt injection

What this adds

New code sample under agentic_top_ten/frameworks/anthropic/privilege_escalation/ demonstrating ASI03: Identity & Privilege Abuse.

The example is a file management agent built with the Anthropic Python SDK. It has deliberate security flaws: over privileged tools, no authorization checks, no input sanitization, and no audit logging. A prompt injection embedded in a data file tricks the agent into deleting files it should not touch.

Files

• insecure_agent.py - The vulnerable agent (~200 lines)
• README.md - Vulnerability writeup with OWASP ASI03 and LLM Top 10 mapping, attack walkthrough, and mitigations
• Dockerfile - Run locally with Docker
• requirements.txt - Python dependencies
• data/ - Test files including the poisoned report

Why

The agentic_top_ten/frameworks/ section currently has two examples (pydantic, mastra), both covering the same vulnerability. This adds a third framework (Anthropic SDK) covering a different risk category (ASI03). It also includes a Dockerfile as required by CONTRIBUTING.md.

Tested

Ran the agent locally. The prompt injection in report_q3.txt successfully triggers unauthorized file deletions, confirming the vulnerability works as documented.