Brief

This article details a Python pipeline designed to extract structured data from unstructured markdown documents using large language models. It emphasizes the limitations of traditional markdown parsers for semantic content extraction and proposes an LLM-based approach for greater resilience to formatting variations. The process involves defining a Pydantic schema for the desired JSON output, embedding this schema directly into prompts for the LLM, and implementing a robust extraction and validation layer to ensure the model returns only valid JSON. AI

A recent technical post details how to secure the OpenAI Agents SDK against memory poisoning attacks, a critical vulnerability known as OWASP ASI06. The method involves using Pydantic field validators within the SDK's architecture to scan and block malicious inputs before they enter an agent's context. This approach, validated by an OpenAI SDK maintainer, leverages the OWASP Agent Memory Guard library to detect various forms of prompt injection and data exfiltration attempts. AI

Prompt injection attacks, analogous to SQL injection for LLMs, pose a significant security risk by allowing malicious users to manipulate AI model behavior. These attacks can override system instructions, extract sensitive prompts, or exfiltrate data. Developers can defend against these threats using a multi-layered approach, starting with a fast, keyword-based blocklist to catch obvious attempts, followed by a more sophisticated method using a separate, isolated LLM to classify potentially malicious inputs. AI

This tutorial details a method for extracting structured data from unstructured text, specifically focusing on cybersecurity advisories. It outlines a process using the OpenAI API, Pydantic for schema definition and validation, and the `tenacity` library for retry logic. The guide covers system prompt design, few-shot examples, and handling ambiguous fields to reliably parse information like CVE IDs, affected products, and remediation steps into a JSON format. AI

Researchers have developed HarnessAPI, a Python framework designed to streamline the creation of tools for AI agents and traditional HTTP clients. This framework uses a typed skill folder as the single source of truth, automatically generating both a streaming HTTP endpoint with Server-Sent Events and an MCP tool registration for agent runtimes like Claude and Cursor. HarnessAPI aims to eliminate code duplication and ensure consistency between the two representations, reducing boilerplate code by 74% in tested scenarios. AI