The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to execute FFmpeg and FFprobe for tasks like scene detection, frame extraction, and storyboard generation. These calls use list-based arguments rather than shell strings, which effectively mitigates common command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection.
Ingestion points: Processes user-provided video files (via understand.py) and researcher-authored background_research.json files (via briefing/context.py).
Boundary markers: Uses structured section headers (e.g., 【挂者】, 【偵标签】) in LLM prompts defined in vlm.py and prompt-templates.md to separate evidence from analysis.
Capability inventory: Includes the ability to execute system commands via FFmpeg and perform network API calls to external services.
Sanitization: Implements parameter escaping for FFmpeg filters in storyboard.py and redacts sensitive keys from API error logs in lib.py.
[EXTERNAL_DOWNLOADS]: The skill communicates with api.xiaomimimo.com and related subdomains to perform ASR and VLM analysis. These network operations are necessary for the skill's primary function and target legitimate service endpoints.

video-understanding