do-agent

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). This skill is high-risk: it explicitly requires granting all read/write/bash capabilities to parallel subagents, enforces hiding full outputs by having subagents save results locally and only return summaries, and mandates fully autonomous execution without human oversight—patterns that enable remote code execution, concealed data flows, and easy data exfiltration or persistent backdoors.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.70). The prompt explicitly mandates creating local work directories and requires the main agent to grant "all Read + Write + Bash tools" to all sub-agents and to run fully autonomously, which encourages broad filesystem and shell access that can modify the machine state (though it does not explicitly request sudo, system config edits, or new user creation).