The source establishes a guarded but powerful agent-configuration and hook-execution capability. It does not establish malicious install-time behavior, credential theft, or a concrete attack against users.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
A user runs and confirms the package setup, selects MCP servers, or enables compatible agent hooks.
Impact
A user-approved setup can extend agent behavior and execute configured local tooling. No confirmed exfiltration, destructive action, or automatic npm-install mutation was found.
Mechanism
The command writes agent configuration under the project directory and can shell out to selected MCP installation commands; installed hooks dispatch local hook files.
Evidence
dist/setup.jsdist/mcp.jsdist/targets/hooks-export.jsstacks/_shared/hooks/run-hook.ts.claude/agents.claude/config/mcp-active.json.claude/hooks/svs-bridge.mjs