ai-image-ppt
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/image_ppt_runtime.pyusessubprocess.runto manage a local virtual environment, install dependencies viapip, and verify imports. These operations are standard setup procedures for local agent skills and are necessary for the skill's functionality. - [EXTERNAL_DOWNLOADS]: During the setup phase, the skill downloads well-known Python packages (
python-pptx,Pillow,openai) from the official Python Package Index (PyPI). It also communicates with established image generation endpoints, includingapi.atlascloud.aiand OpenAI-compatible APIs. - [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text content (articles, reports, and outlines) to generate slide images. While the skill processes untrusted input, it utilizes extensive constraints in its prompts to guide the AI model's output.
- Ingestion points: Untrusted user-supplied content such as articles and reports are processed in the workflow described in
SKILL.mdanddocs/outline-style-and-sample.md. - Capability inventory: The skill has the ability to perform file system operations (writing images and presentations) and network requests to third-party image APIs.
- Boundary markers: The system uses structured prompts with explicit constraints (e.g., 'Render Chinese text exactly', 'No watermark') to delimit the generation task.
- Sanitization: No explicit sanitization of the input text is performed before interpolation into image generation prompts.
Audit Metadata