longform-to-content

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow ingests outsider-authored free text when generating/packaging content from the user’s own spoken recording transcript and rendering those words into captions/metadata without any step that filters out injected instructions embedded in the user-provided content (e.g., Gate 2 transcribe→“READ the entire packed transcript before cutting,” then CAPTIONS.md + YOUTUBE.md use that transcript-derived text for SRT/cues and packaging).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs agents at runtime to clone and/or read the repository (git clone https://github.com/crealwork/longform-to-content and fallback raw docs at https://raw.githubusercontent.com/crealwork/longform-to-content/master/), and the setup requires installing and running external code via git clone of https://github.com/browser-use/video-use and https://github.com/ggml-org/whisper.cpp, all of which fetch remote code that the agent will execute or use to control its prompts/workflow.