- likes
- 21
- comments
- 1
Post
If an AI skill can shape an agent’s behavior, it definitely shouldn’t hold all the keys. Teams should be able to adopt skills from public repos or MCP without turning every new skill into a data leak. Here’s how Glean handles it: 🔹 Catch prompt injections, suspicious destinations, and dangerous tool usage upfront. 🔹 Sandboxes keep untrusted code contained, while credential brokers enforce destination policy. 🔹 Bottom-up creation meets top-down admin guardrails. See how we keep skills safe to scale: https://lnkd.in/ephGP85x