epub2podcast-standalone
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it ingests untrusted text from EPUB files and interpolates it into LLM prompts for script and slide generation. Ingestion occurs in 'src/services/epubService.ts' and content is processed in 'src/services/scriptService.ts'. While the skill uses XML-like boundary markers (e.g., <full_content>) in 'src/prompts/script.ts' to mitigate accidental instruction obedience, it does not implement specific sanitization to filter out adversarial prompt patterns within the book content.
- [COMMAND_EXECUTION]: The tool relies on 'ffmpeg' and 'ffprobe' system utilities for audio merging, video rendering, and duration calculations. Commands are executed via the 'execFile' and 'fluent-ffmpeg' libraries in 'src/services/audioService.ts' and 'src/localPipeline.ts'. Arguments are constructed from internal file paths and numeric metadata, minimizing command injection risks.
- [EXTERNAL_DOWNLOADS]: The skill interacts with several external AI and TTS service providers to perform its core functions. It makes network requests to OpenRouter (openrouter.ai), ElevenLabs (elevenlabs.io), Minimax (minimaxi.com), and Volcengine (openspeech.bytedance.com). These are documented in their respective service files as part of the intended pipeline functionality.
Audit Metadata