presentation-master
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (Topic, Audience, Purpose) and external data (WebSearch results) to generate slide content, which is then interpolated into prompts for an image generation skill.
- Ingestion points: User input for Topic/Audience/Purpose (SKILL.md, analyze-context.js), and external content retrieved via WebSearch and Obsidian knowledge extraction (Phase 2 workflow).
- Boundary markers: The prompt generation logic in
scripts/suggest-images.jsuses template literals to build prompts for theimage-genskill without employing delimiters or specific instructions to ignore embedded commands within the slide content. - Capability inventory: The skill has access to
Bash(to run internal adapter scripts),Write(to save files and metadata to Obsidian),WebSearch, and the ability to call other skills includingimage-genandpptx. - Sanitization: While the skill performs basic string manipulation for filename generation, it lacks robust sanitization or escaping for slide content before it is used to construct prompts for downstream AI tools.
Audit Metadata