local-llm

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

• Third-party content exposure detected (medium risk: 0.30). [process_command] builds a system prompt and then sends first-party runtime user_input (“user” content) directly into Ollama via chat(), so outsider-authored text delivered as user_input into the workflow is read by the LLM without selecting a specific item.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runs a remote install script via curl -fsSL https://ollama.com/install.sh | sh (SKILL.md:19-19), which fetches and executes remote code as part of the required Ollama installation.

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 running system-level service commands with sudo (sudo systemctl enable/start) and pipes an external install script (curl | sh), which directs the agent to perform privileged modifications to the host system.