The Agent Skills Directory

[Prompt Injection] (SAFE): The skill contains only instructional guidance for a developer persona. No bypass markers, DAN-style injections, or attempts to extract system prompts were detected.
[Data Exposure & Exfiltration] (SAFE): No sensitive file paths, hardcoded credentials, or unauthorized network operations were found. References to image I/O are within standard library usage.
[Obfuscation] (SAFE): The file is written in clear, human-readable Markdown with no Base64, zero-width characters, or hidden Unicode tags.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): Dependencies listed (OpenCV, PyTorch, NumPy, etc.) are standard, reputable packages in the data science ecosystem. There are no patterns for piping remote URLs to interpreters.
[Privilege Escalation] (SAFE): No use of sudo, chmod, or administrative commands detected.
[Persistence Mechanisms] (SAFE): No attempts to modify shell profiles, cron jobs, or system services identified.
[Metadata Poisoning] (SAFE): Metadata fields accurately describe the skill's functionality as a computer vision guidance tool.
[Indirect Prompt Injection] (SAFE): While the skill instructions cover the ingestion of external data (images/video), the skill definition itself does not implement an automated pipeline with missing boundary markers or unsafe interpolation.
[Time-Delayed / Conditional Attacks] (SAFE): No time-based or environment-based conditional logic detected.
[Dynamic Execution] (SAFE): No use of eval(), exec(), or runtime compilation techniques.

computer-vision-opencv