video-feed-ocr

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The LLM OCR mode explicitly instructs "Extract all text exactly as seen" and "Do not paraphrase," which forces verbatim output of on-screen text — therefore any secrets (API keys, tokens, passwords) present in video frames would be reproduced by the model.

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

• Potentially malicious external URL detected (high risk: 0.90). The script calls hardcoded LLM endpoints at runtime (e.g. http://192.168.1.161:1234/v1/chat/completions, http://192.168.1.161:1234/api/v1/models, and http://192.168.1.161:1234/api/v1/models/load), sending image+prompt payloads and relying on the returned completions as the OCR output and issuing model-load requests, so these URLs are runtime dependencies that directly control the skill's outputs and remote model behavior.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly instructs "Always run this skill in escalation permission mode," asking for elevated/privileged execution (sudo-like escalation) which can compromise the machine state.