The automatic postinstall writes an AI-agent skill into detected user directories and, for Claude, adds a persistent session-start hook to user-level settings. This is an unconsented AI-agent control-surface modification, so it warrants blocking under the install-time agent configuration rule.
Static reason
High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review; source fingerprint signature matched known malicious package; routed for review
Trigger
npm install when a supported agent client is detected; Claude hook wiring also requires a newly installed skill and Python availability.
Impact
Changes agent configuration outside the package directory and causes the package’s orientation validator to run at Claude session start.
Mechanism
Copies the package skill into the agent’s home directory and writes a command entry into the user-level Claude settings.json SessionStart hooks.
Policy narrative
During npm postinstall, the package detects agent clients and copies its skill into their home directories. For a newly installed Claude skill, it also writes a SessionStart command into the user-level Claude settings file, causing the package’s validator to run on later sessions. This install-time agent configuration change happens without a separate request to enable the hook.
Evidence
package.jsonscripts/postinstall.js