The package contains guarded but automatic install-hook mutation and runtime AI-agent configuration mutation. The code supports a warning for dangerous capability, not a malicious block, because no concrete attack behavior was found.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
Installing the package activates postinstall; constructing a plan executor activates configuration restoration.
Impact
This is a guarded automatic install-time privilege-sensitive mutation and a runtime capability to change project agent configuration. No confirmed malicious payload or data theft was identified.
Mechanism
The hook calls the sandbox configurator non-interactively, and the configurator can use chown and chmod. The executor reads backup files and writes or removes the corresponding project configuration files.
Evidence
chrome-sandbox.claude/settings.json.mcp.json