The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The README.md file provides installation instructions that include piping a remote script directly to a shell: curl -fsSL https://www.modelscope.cn/skills/install.sh | bash. This pattern allows for the execution of arbitrary code from the remote server on the user's system.
[COMMAND_EXECUTION]: The Python scripts scripts/ai_2md2marp2av.py and scripts/md2marp2av.py use subprocess.run() to execute external binary commands, specifically the marp and npx marp command-line interfaces, to render Markdown files into images and HTML.
[PRIVILEGE_ESCALATION]: The references/environment.md file contains instructions for the agent to install system-level dependencies using apt-get install -y ffmpeg, which requires administrative (root/sudo) privileges.
[EXTERNAL_DOWNLOADS]: The skill's setup process involves downloading multiple external dependencies via pip install (e.g., edge-tts, moviepy), npm install (e.g., @marp-team/marp-cli), and curl, reaching out to various third-party registries and servers.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown files to generate narration and slide content, creating an attack surface where embedded instructions in the source file could influence the agent's output or the narration generated by the TTS engine.
Ingestion points: Markdown files provided by the user and processed by scripts/md2video.py and scripts/ai_2md2marp2av.py.
Boundary markers: None explicitly defined in the scripts to prevent the execution of instructions embedded within the text processed for TTS.
Capability inventory: File system write access, network communication (via edge_tts), and shell command execution (subprocess calls for Marp).
Sanitization: The sanitize_text_for_display function in scripts/md2video.py removes specific HTML tags and emojis, but does not sanitize against prompt-based instruction injection.

md2video-audio