llamacpp-tuner

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

• Third-party content exposure detected (high risk: 0.85). Step 3 fetches outsider-authored free text from the public Hugging Face model card README (https://huggingface.co/<repo>/raw/main/README.md) via scripts/fetch_model_card.py, then extracts prose lines and feeds them into the agent’s context as “author config notes”/recommendations.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly instructs the agent to stop/restart local services (kill , nohup relaunch), stop other LLM runners, and run system-level inspection/affinity commands (pgrep, ss, taskset, numactl), which modify the machine's runtime state and can disrupt services — it does not, however, request sudo, edit system files, or create users.