ergo-remotion-video
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied articles (article.md) to generate video scripts and metadata. This provides an attack surface where instructions embedded in the source article could potentially influence the agent's behavior during content preparation.
- Ingestion points: The skill explicitly reads and processes article.md as described in the '流程与验收' section of SKILL.md.
- Boundary markers: The skill establishes specific truncation points to remove promotional content and 'ignore' specific trailing segments of text.
- Capability inventory: The skill calls local Python scripts that perform file system writes, network requests to the Volcengine TTS API, and execute video rendering via Remotion.
- Sanitization: The instructions include precise patterns for identifying and removing specific text blocks (e.g., WeChat public account promotions) from the ingested articles.
- [COMMAND_EXECUTION]: The skill uses local Python-based command-line tools to perform its core functions, including project initialization and audio verification.
- Evidence: Calls to tools such as init_project.py, doctor.py, gen_audio.py, and align_words.py are initiated from the Docs/src/ai/script/shared/tools/ directory within the repository.
Audit Metadata