5dive-cli

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow sends arbitrary user-supplied free text into agents via the host’s shared CLI/queue surfaces, especially 5dive task add / 5dive task set-body (task queue contents are group-writable and readable at runtime by agent-* users) which the LLM then processes as task body text.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). This skill repeatedly instructs the agent to run sudo 5dive commands (e.g., sudo 5dive agent create) which create Linux users and systemd units and grant scoped sudo actions, i.e. it explicitly pushes the agent to perform privileged, state-changing operations on the host.