https://blog.navcore.io Last 10 notes on 0xIvan Ivan blog.navcore.io https://blog.navcore.io/AI-Agents/Giving-a-Pi.dev-Agent-Hands-on-a-Flipper-Zero https://blog.navcore.io/AI-Agents/Giving-a-Pi.dev-Agent-Hands-on-a-Flipper-Zero I wanted to stop asking my AI agent to summarise documents and start asking it to poke at the real world. The Flipper Zero was the obvious first stop, it talks SubGHz radio, NFC, RFID, infrared, GPIO, and USB HID, and it plugs into a laptop like a boring serial device. Fri, 17 Apr 2026 00:00:00 GMT https://blog.navcore.io/AI-Agents/DSPy-RLM-with-MCP-Code-Mode https://blog.navcore.io/AI-Agents/DSPy-RLM-with-MCP-Code-Mode DSPy’s RLM (Reinforcement Language Model) module lets LLMs write Python code instead of picking tools one at a time. Combined with mcp_use’s code_mode, this creates a powerful pattern: the LLM writes code that directly calls MCP tools as async Python functions. Mon, 26 Jan 2026 00:00:00 GMT https://blog.navcore.io/AI-Agents/Replicating-Anthropic-Agent-Skills-with-DSPy https://blog.navcore.io/AI-Agents/Replicating-Anthropic-Agent-Skills-with-DSPy Anthropic recently released their Agent Skills specification, a framework for teaching AI agents new capabilities through modular, reusable skill packages ... Thu, 08 Jan 2026 00:00:00 GMT https://blog.navcore.io/AI-Security/Prompt-Injection/Building-an-AI-Guardrail-with-Embeddings-Part-2 https://blog.navcore.io/AI-Security/Prompt-Injection/Building-an-AI-Guardrail-with-Embeddings-Part-2 In Building an AI Guardrail with Embeddings, we built a basic prompt injection detector using BGE embeddings and three MLP classifier heads. It worked, but we used toy data and a simple architecture. Sat, 03 Jan 2026 00:00:00 GMT https://blog.navcore.io/AI-Security/Prompt-Injection/Building-an-AI-Guardrail-with-Embeddings https://blog.navcore.io/AI-Security/Prompt-Injection/Building-an-AI-Guardrail-with-Embeddings LLMs are powerful, but they’re vulnerable. Prompt injection attacks can trick models into ignoring instructions, leaking data, or doing things they shouldn’t. Fri, 02 Jan 2026 00:00:00 GMT https://blog.navcore.io/Malware/LokiBot-Analysis https://blog.navcore.io/Malware/LokiBot-Analysis Brief Introduction The initial delivery was via email, however this post is about analyzing the delivery stages, malware and some SECOPS fails from the LokiBot threat actors ... Sun, 08 May 2022 00:00:00 GMT https://blog.navcore.io/ https://blog.navcore.io/ DSPy RLM with MCP Code Mode Giving a Pi.dev Agent Hands on a Flipper Zero Replicating Anthropic Agent Skills with DSPy LokiBot Analysis Building an AI Guardrail with Embeddings Part ... Mon, 01 Jan 1990 00:00:00 GMT