auto-updater

To safely review skill updates, you need a process that checks pinned community skills for newer commit SHAs and generates per-skill diffs for human trust review. This enforces explicit approval before overwriting installed files.

Security guardrails for updating hooks and MCP configurations must enforce fail-closed checks for hook execution, MCP credential access, and tool permission changes. A fresh-reference scan gate should run prior to overwriting installed skill files.

Immutable commit SHA checks prevent unreviewed changes by fetching the latest source commit SHA and comparing it to the installed version. This avoids mutable tags and ensures only explicitly approved updates are applied to your skills.

Freshness verification detects stale skills by tracking their verified freshness window. When a skill exceeds this window, the system prompts re-validation to ensure the installed version remains trusted and secure against upstream behavior changes.

You should require explicit approval for high-surface security changes whenever scan or regression rules indicate increased risk. The system refuses updates automatically if these fail-closed security checks detect unauthorized hook execution or tool permission modifications.