The source supports a warning for automatically refreshed first-party agent instructions loaded into Claude. The installer is user-invoked and the skill is scoped to individual Claude processes, so the evidence does not support a malware block.
Static reason
High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review
Trigger
Starting the Claude host, followed by a Claude process using the installed host.
Impact
Upstream skill changes can alter instructions available to Claude in later sessions.
Mechanism
When the local skill is older than twelve hours, the host fetches skill files from the project's GitHub repository and writes them to its runtime directory; Claude receives that directory as a plugin.
Evidence
src/host/claude-host.mjssrc/host/skills/lizard/SKILL.mdsrc/host/install.mjs
Network endpoints1
raw.githubusercontent.com/lizard-build/skill/main