The package automatically updates and supplies a publisher-owned skill to Claude and Codex, which meets the warning-level agent extension lifecycle risk category. Installation itself is an explicit CLI action, and the inspected source does not show a concrete attack or broad agent configuration write.
Static reason
High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review
Trigger
The native host starts; Codex receives the skill when its app server initializes, and Claude receives it when a session is spawned.
Impact
Updated publisher-controlled skill guidance can enter agent context automatically; the inspected code does not establish credential theft or an unconsented write to user agent configuration.
Mechanism
The Claude host fetches skill files from raw.githubusercontent.com and stores them in its bundled skill directory, then supplies that directory to Claude and registers the bundled root with Codex.
Evidence
src/host/claude-host.mjssrc/host/codex-host.mjssrc/host/skills/lizard/SKILL.mdsrc/host/skills/lizard/README.mdsrc/host/skills/lizard/skills.sh.json