micro-models-agent@0.2.22

One or more suspicious static signals were detected.; previous stored version diff introduced dangerous source

User runs mma with an agent prompt, invokes /init, or configures MCP/hooks.

A compromised or prompt-injected configured model could execute commands with the invoking user's environment; explicit /init can alter project agent instructions.

LLM-directed shell execution plus user-configured extension commands.

Source inspection does not support a malicious verdict or publish block. The package nonetheless exposes unguarded LLM-directed command execution and explicit agent-instruction mutation, so it warrants a warning under the dangerous-capability and explicit user-command control-surface policy.