The postinstall can automatically add a recurring package-owned hook to Claude’s user-level settings, which warrants a warning. Its traced behavior is limited to local project-document orientation, with no confirmed attack behavior.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
npm postinstall detects Claude Code, skill installation reports success, and Python and valid settings are available.
Impact
The hook runs as part of later agent sessions and places project-document context into the local agent session. No external recipient or credential transfer was established.
Mechanism
The installer adds a SessionStart command to user-level Claude settings. The bundled command reads a fixed set of confined project documents and emits bounded orientation to the agent.
Evidence
package.jsonscripts/postinstall.jsscripts/lib.jsskills/course-creator/scripts/sdlc_core.py