The install hook automatically writes the package's own Claude skill into eligible projects, which is an agent extension lifecycle risk. The inspected skill is product documentation, with no evidence of permission changes, concealed behavior, or credential exfiltration.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
Package installation as a direct dependency, unless skipped by the hook's conditions or disabled in project configuration.
Impact
This creates an agent extension without a separate explicit setup command, a lifecycle risk; the inspected skill content is product documentation and shows no agent control hijack.
Mechanism
The hook writes or upgrades the ax-llm skill and its bundled guides under the project's .claude/skills directory.
Evidence
package.jsonscripts/postinstall.mjsscripts/skill-install.mjsskills/ax-llm.md.claude/skills/ax-llm/SKILL.md.claude/skills/ax-llm/references/