media-notebooklm-slide-prompt
Pass
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied markdown files to generate complex prompts for other AI systems.
- Ingestion points: The skill ingests "세션 본문 MD" (Session body Markdown) either as a file or direct text attachment in Phase 1 of its workflow.
- Boundary markers: The prompt template defined in SKILL.md lacks explicit boundary delimiters or instructions to ignore embedded commands within the ingested markdown content, which could lead the agent to follow instructions maliciously placed inside a legitimate-looking lecture text.
- Capability inventory: The skill generates text meant to be used as prompts for external platforms (NotebookLM and Gemini/Nano Banana). It does not perform file writes, network requests, or shell executions within the host environment.
- Sanitization: The skill recommends chaining to
moai-coworker:ai-slop-reviewerat the end of the process to clean up text, but this is an optional formatting step rather than a security sanitization measure. - [EXTERNAL_DOWNLOADS]: The skill references several official Google documentation and resource URLs for NotebookLM, DeepMind, and Google AI for Developers. These are documented as official sources for prompt engineering guidelines and do not involve executable code downloads.
Audit Metadata