The Agent Skills Directory

Indirect Prompt Injection (SAFE): The create_video_script.py script possesses a data ingestion surface that processes untrusted markdown content.
Ingestion points: markdown_path read via Path.read_text() in create_video_script.py.
Boundary markers: Absent; slide content and speaker notes are interpolated directly into the output markdown file.
Capability inventory: Limited to local file system read (Path.read_text) and write (Path.write_text). No command execution or network operations are present in the script.
Sanitization: No sanitization of the input markdown is performed; however, the output is a review script intended for human or agent consumption and does not trigger dangerous downstream tools in the provided scope.
Data Exposure & Exfiltration (SAFE): The README provides instructions for users to set their OPENAI_API_KEY using a placeholder (sk-...). No hardcoded credentials or sensitive file access patterns were found in the code.
Unverifiable Dependencies (SAFE): The requirements.txt file specifies standard, well-known libraries (openai, google-genai, Pillow). No suspicious or unversioned packages were detected.

markdown-video