<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Jose López · Writing</title><link>https://joselopez.pro/en/blog/</link><description>Notes from building secure AI systems with agents, in production and at home.</description><language>en</language><item><title>Be careful with Claude Code mods: they open a door to all your connectors</title><link>https://joselopez.pro/en/blog/silent-bridge/</link><guid>https://joselopez.pro/en/blog/silent-bridge/</guid><pubDate>Mon, 02 Nov 2026 09:00:00 -0000</pubDate><description>There are already 7,349 public Claude Code mods, nine in ten in repositories created in October, and 40% run programs on your machine. We built one and measured what it can do without you noticing, from the mode where Claude Code asks at every step to the one that asks nothing.</description></item><item><title>I trained a poisoned model in 3h 49m</title><link>https://joselopez.pro/en/blog/poisoned-model-3h49m/</link><guid>https://joselopez.pro/en/blog/poisoned-model-3h49m/</guid><pubDate>Tue, 27 Oct 2026 09:00:00 -0000</pubDate><description>A DeepSeek agent built and trained three variants on three RTX 4090s. What the earlier runs measured, and why V6 still needs a full-episode evaluation.</description></item></channel></rss>
