carousels
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes local Node.js scripts (
generate-image.mjsandrender.mjs) to process image generation requests and render the final slide layouts. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied source content (such as emails or scripts) to derive slide copy and image prompts, presenting a surface for indirect prompt injection if the source content contains malicious instructions.
- Ingestion points: Pasted text or local files processed in Phase 1 (SKILL.md).
- Boundary markers: None identified in the instructions for distilling source content.
- Capability inventory: Node.js script execution, local file writing, OpenAI API interaction, and headless browser rendering via Puppeteer.
- Sanitization: Content is distilled and written into HTML templates by the agent without explicit sanitization logic.
- [DYNAMIC_EXECUTION]: The skill generates HTML slide structures at runtime and uses Puppeteer to execute a headless browser instance for screenshotting and PDF assembly.
- [EXTERNAL_DOWNLOADS]: The skill performs network requests to the OpenAI API for image generation and utilizes Puppeteer, which may download official browser binaries for rendering tasks.
Audit Metadata