The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/video2text.py uses the subprocess module to execute system binaries yt-dlp and ffmpeg. While it uses argument lists to mitigate shell injection, it remains a high-privilege capability involving external input (video URLs).- [EXTERNAL_DOWNLOADS]: The skill downloads video and audio data from user-provided URLs via yt-dlp. It also requires the installation of the tencentcloud-sdk-python-asr package from a well-known cloud provider.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted video transcripts which are subsequently fed into the AI for proofreading and filing. This presents an attack surface where instructions embedded in the video audio could influence the agent's behavior.
Ingestion points: Video audio transcribed into text via Tencent Cloud ASR in scripts/video2text.py.
Boundary markers: The SKILL.md instructions for Step 2 do not specify delimiting the transcript content to prevent instruction confusion.
Capability inventory: The skill can write files to the local file system (05-选题研究/ directory) and execute shell commands via Python.
Sanitization: There is no evidence of sanitization or filtering of the transcript content before it is processed by the AI.

li-transcript