TrustSkills Blogs
Field notes for the AI agent security frontier
Every post on this page is written to support a security decision. We cite the reporting, standards, and vendor documentation we relied on so readers can validate the analysis themselves.
EU AI Act and AI agent skills: what organizations need to know before August 2026
August 2 is not just another compliance date. It is when the European Commission gains the authority to issue fines for GPAI model violations — and the supply chain that includes your agent skills is in scope.
Data exfiltration in AI agent skills: how attackers steal credentials through ClawHub
The most damaging ClawHub attacks are not the flashy ones. They are the skills that quietly read your credentials and send them to an attacker-controlled server while appearing to do something useful.
How to scan an AI agent skill for malware before you install it
Most users install ClawHub skills without reviewing the code. Here is a repeatable, five-minute process that catches the threats that visual review and star ratings miss.
What is MCP server security?
MCP servers extend what AI agents can do. That is also what makes them a security boundary. Understanding the threat surface is essential before you connect one to a model with real capabilities.
ClawHavoc explained: the supply chain attack that put 1,184 malicious skills on ClawHub
ClawHavoc placed over 1,000 malicious skills on the ClawHub marketplace in early 2026. Understanding the attack pattern is the first step to defending against the next campaign like it.
ClawJacked shows why localhost is not a security boundary
The ClawJacked disclosure is a strong reminder that a local gateway is still reachable from a malicious browser tab if the surrounding trust model is weak.
8 best practices before you install an AI agent skill
Installing an AI skill is not like installing a harmless theme. You are often extending a control plane that can read data, reach services, and trigger real actions on your behalf.
What is prompt injection?
Prompt injection is not just a clever string. It is any input that changes a model's behavior in a way the system designer did not intend, especially when the model can reach tools, data, and accounts.
The OpenClaw inbox incident is a security lesson, not a meme
The reported OpenClaw inbox wipe did not just expose model unreliability. It showed why approvals, identity separation, and destructive-action controls must live outside the prompt.