Skills
Agent frontmatter hooks now require the agent file's own folder to be workspace-trusted
Claude Code 2.1.218 hardened a real escalation path: hooks declared in an agent file's frontmatter no longer execute unless that file's own folder has accepted workspace trust, closing the gap where a cloned repo or vendored plugin directory could run hooks on load. The same release fixed `/context` reporting stale token usage after compaction, and relaxed skill/plugin frontmatter booleans to accept `yes`/`no`/`on`/`off`/`1`/`0` case-insensitively. If you distribute agents or plugins, expect silent hook no-ops in untrusted subfolders and document the trust step.
↳ Follow the thread