The verified postinstall creates or upgrades a first-party agent skill surface in the consumer project, which warrants a warning for agent extension lifecycle risk. The bundled entrypoints were unreadable, so this review cannot establish a clean verdict or resolve their behavior.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
npm installation, except when recognized CI indicators or AX_SKIP_SKILL_INSTALL cause the hook to skip.
Impact
Project-local agent behavior can be influenced by the newly installed or upgraded skills without a separate setup command.
Mechanism
The hook copies skill Markdown files into .claude/skills and upgrades existing files when the package version is newer.
Evidence
package.jsonscripts/postinstall.mjscli/index.mjsskills/ax-agent.md.claude/skills