devils-advocate

SUSPICIOUS. The core skill behavior is a benign critique framework, and its stated ColdIQ use is plausibly related to validating GTM claims. However, it extends beyond pure analysis by instructing installation of an external MCP via unpinned `npx @latest` and forwarding an API key to that process, with only partially verified package provenance. Data flows to same-org ColdIQ domains appear coherent, so this is not clearly malicious, but the install and credential-forwarding footprint is disproportionate enough to warrant caution.