Agent skills shift AI agents toward procedural tasks with skill.md steps; progressive disclosure reduces context window bloat ...
Malicious AI browser extensions collected LLM chat histories and browsing data from platforms such as ChatGPT and DeepSeek.