The postinstall hook installs the package’s own AI-agent skill into the project by default, creating an agent instruction surface without an explicit setup command. The reviewed content is package-related and the setup has opt-outs, so this is suspicious rather than a confirmed malicious control hijack.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
Package installation outside CI, unless automatic skill setup is disabled.
Impact
The installed agent instructions can affect later Claude interactions without an explicit setup command; the reviewed files identify the skill as package documentation.
Mechanism
The postinstall hook reads project configuration and writes the bundled ax-llm skill and reference guides under .claude/skills.
Evidence
package.jsonscripts/postinstall.mjsscripts/skill-config.mjsscripts/skill-install.mjsskills/ax-llm.md<project>/.claude/skills/ax-llm/SKILL.md<project>/.claude/skills/ax-llm/references/*