The package-owned launcher automatically writes persistent Claude Code hooks when the configured tool starts, creating a residual agent configuration risk. The inspected hooks report activity to the selected Parley chat, with no source evidence here of a concrete attack or malicious payload.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
Starting the configured Claude Code tool when its Parley chat server is present.
Impact
The settings change persists and affects subsequent Claude Code runs for the VM user; the inspected callbacks report activity to the selected Parley chat.
Mechanism
The launcher writes PreToolUse, UserPromptSubmit, Stop, and Notification callbacks into the user’s Claude settings while preserving other hook groups.
Evidence
config/launch/claude-code.shconfig/launch/claude-pulse.mjsconfig/tools.jsonserver/conversations.mjs~/.claude/settings.json