The install hook makes an unprompted change to the consumer's Claude skill directory and deletes any existing same-named skill before copying its own. The inspected skill content is package usage guidance, so this supports a warning for agent-extension lifecycle risk rather than a confirmed malicious attack.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
The package postinstall hook runs during installation.
Impact
An existing local skill at that destination can be lost, and Claude receives the package's skill guidance without a separate setup action.
Mechanism
The hook resolves the consumer project root, targets .claude/skills, removes the destination skill directory recursively, then copies the package's skill into it.
Evidence
package.jsonpostinstall.mjsagents/claude/evoke-astro-core/SKILL.md.claude/skills/evoke-astro-core