openagent

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs running remote code via npx (e.g. the referenced package and its Git fallback) so the git repo references can be fetched and executed at runtime — see "github.com/5dive-ai/openagent" and "github:5dive-ai/openagent#" which appear in the docs.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs using "sudo 5dive agent import ..." to provision a new running agent and describes an automated self-provisioning flow ("no human in the loop"), which directly encourages obtaining sudo privileges and modifying the host system state.