The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves downloading and processing subtitle files (SRT) from external URLs. These files are then passed directly into a translation prompt for the AI to process.
Ingestion points: Subtitles are fetched via youtube-downloader and stored in /tmp/video-translation before being read as text (SKILL.md, Workflow step 2).
Boundary markers: The skill uses a specific 'Translation Prompt' to guide the LLM, but it lacks strict XML-style delimiters or explicit instructions to ignore control sequences potentially embedded in the subtitle text.
Capability inventory: The skill has access to shell execution (ffmpeg, bash scripts) and file system writes. If a malicious subtitle file successfully injects commands that influence the agent's next steps, it could lead to unauthorized file manipulation.
Sanitization: There is no evidence of sanitization or validation of the subtitle content before it is sent to the LLM for translation.
[EXTERNAL_DOWNLOADS]: The requirements section instructs the user or agent to clone/copy external skills from GitHub repositories (github.com/crazynomad/skills and github.com/NoizAI/skills) to fulfill dependencies like youtube-downloader and tts. These are documented neutrally as required resources for the skill's functionality.

video-translation