The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several system binaries for its primary media processing and automation tasks.
Scripts including qa.py, clipcheck.py, and mix-bgm.sh execute ffmpeg and ffprobe to generate frames, mix audio, and verify video metadata.
The pipeline-playbook.md defines workflows using yt-dlp for downloading reference footage and Playwright for automated evidence collection via browser screenshots.
All executions are consistent with the intended video production pipeline and use standard argument passing to avoid shell injection.
[EXTERNAL_DOWNLOADS]: The skill connects to well-known technology services to fulfill its functionality.
It integrates with the ByteDance Volcano Engine (Volcengine) API for high-quality text-to-speech synthesis.
It facilitates scraping trending topics and research data from major platforms such as Baidu, Toutiao, Zhihu, Bilibili, and 36kr.
These connections are documented for the user and target established service domains.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted news content and search results, creating a potential ingestion surface.
Ingestion points: News hot-lists and web search results enter the agent context through scraping tasks defined in pipeline-playbook.md.
Boundary markers: The workflow uses structured JSON manifests (sentences.json, timeline.json) to isolate processing steps.
Capability inventory: The skill has shell access to media tools and a browser engine (Playwright).
Sanitization: The skill contains explicit instructions for the AI to perform manual fact-checking and source cross-referencing in video-common/references/fact-check.md, mitigating the risk of obeying malicious instructions embedded in news sources.

investigation-video-skill