The Agent Skills Directory

[PROMPT_INJECTION]: The script scripts/analyze_ad.py interpolates user-provided product information directly into the prompt without sanitization or boundary markers.
Ingestion points: Data from the --product-info command-line argument in scripts/analyze_ad.py.
Boundary markers: Absent; user-provided content is placed directly into the prompt template.
Capability inventory: The skill can read local files, write reports, execute the ffmpeg utility via subprocess, and make network requests to Volcengine.
Sanitization: No validation or escaping is applied to the input JSON content.
[COMMAND_EXECUTION]: The script executes the ffmpeg utility via subprocess.run to compress video files. While the execution uses argument lists to prevent shell injection, it operates on user-provided file paths.
[DATA_EXFILTRATION]: The skill transmits video data to ark.cn-beijing.volces.com for processing. This is a legitimate connection to the Volcengine cloud service required for the Doubao model's operation.

ad-analytics