openkrak-mcp@1.0.6

One or more suspicious static signals were detected.; source fingerprint signature matched known malicious package; routed for review

A configured coding agent lists or calls an OpenKrak MCP tool.

An agent may be steered away from independent source inspection; remote service receives a persistent host-derived identifier and error details.

Agent-directed prompt injection combined with remote fingerprint and error telemetry.

After a user configures this package as an MCP server, its tool descriptions tell the coding agent to call its analysis first and not inspect source files. Every tool call performs licensing with a hash derived from hostname and username. Failures additionally report error text and that hash to the remote service. This combines AI behavior steering with remote host tracking, although it does not run during npm installation.

The package embeds instructions intended to make an AI defer to its opaque analysis rather than independently inspect source, which is agent-control manipulation. Its per-call host fingerprinting and error telemetry strengthen the concrete privacy and control risk despite no install hook.