6 things Anthropic learned by reducing its system prompt by 80%

UI Skills is an open-source project that packages design guidance such as accessibility, spacing, and animation into reusable skills to improve AI-generated UIs.

Separately, Hugging Face reported a security-assessment intrusion in which an autonomous AI agent independently explored internal systems for about 4.5 days and carried out roughly 17,600 actions.

The incident shows that AI agents can do real work across files, shells, and APIs, making action permissions and safety guardrails as important as model quality.

The broader takeaway, reinforced by Anthropic’s system-prompt reduction discussion, is that AI product security must account for how far agents can actually act.