paidf-cosmos-predict

SUSPICIOUS. The skill’s purpose, file access, Docker/GPU usage, and VLM API flow are broadly consistent with media generation, and the scanner’s command-injection/secret flags are benign documentation artifacts. The main risk is execution trust: the PAIDF container image is not verifiably identified in the skill text, yet it receives host mounts plus HF_TOKEN and optional VLM_API_KEY. That combination makes the skill high risk even without evidence of confirmed malicious intent.