The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/extract_phase.py uses subprocess.run to call ffmpeg for audio manipulation. This is a legitimate and expected use of the tool within the skill's context for media processing. The script includes checks for the ffmpeg binary and uses fixed command structures.
[EXTERNAL_DOWNLOADS]: The skill mentions external AI video generation services (e.g., doubao-seedance, kling, sora, veo) and refers to other internal skills (e.g., qingyun-api, video-lyrics-subtitle). These references are instructional and do not involve unauthorized remote code execution or unknown package downloads.
[PROMPT_INJECTION]: The skill provides structured templates and 'rules' for prompt construction. These instructions are directed at improving AI output quality and do not contain patterns aimed at bypassing safety filters or overriding system instructions.
[DATA_EXFILTRATION]: Analysis of the Python and Bash scripts shows that file operations are limited to local project directories for reading assets (audio, SRT) and writing output. There is no evidence of network operations sending sensitive data to external servers.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external data (lyrics, scene descriptions) to build prompts, it uses these for creative generation within a controlled workflow. Standard security measures for handling such data in the context of LLM agents are recommended if integrated into automated pipelines.

cinematic-video-gen