ppt-agent
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill consumes untrusted data that is later interpolated into generated files.
- Ingestion points: User input (Phase 1) and external data from
WebSearch(Phase 3) inSKILL.md. - Boundary markers: The skill uses a JSON wrapper
[PPT_OUTLINE]for internal data exchange, but the final output instructions for SVG/HTML do not specify the use of delimiters to ignore embedded instructions. - Capability inventory: File system write access for generating multiple SVG files (
NN-slug.svg) and apreview.htmlviewer inSKILL.mdandreferences/svg-generation.md. - Sanitization: The instructions lack explicit directives to sanitize or escape potentially malicious tags (e.g.,
<script>,onloadattributes) when converting untrusted web search results into SVG/HTML content. - [COMMAND_EXECUTION]: The skill requires the agent to perform file system operations based on its own generated output.
- Evidence: Phase 4 in
SKILL.mdandreferences/svg-generation.mdinstructs the agent to generate multiple independent SVG files and apreview.htmlfile to facilitate a slide-viewing experience. This is the primary intended behavior of the skill.
Audit Metadata