thumbnail-creator
Fail
Audited by Gen Agent Trust Hub on Jun 17, 2026
Risk Level: HIGHCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Persistence mechanism through shell profile modification.
- Evidence: The instructions guide the user to persist an environment variable by appending it to shell profiles:
echo 'export GEMINI_API_KEY="your-key-here"' >> ~/.zshrc. - [COMMAND_EXECUTION]: Local script generation and execution.
- Evidence: The skill relies on a local script
generate_image.py. It instructs the agent to create the script if missing and subsequently execute it viapython3 generate_image.pywith various command-line arguments. - [EXTERNAL_DOWNLOADS]: Installation of third-party dependencies.
- Evidence: The skill requires
pip install google-generativeai Pillow requests. While these are well-known packages from a trusted service (Google), they represent external code dependencies necessary for execution. - [PROMPT_INJECTION]: Indirect prompt injection surface.
- Ingestion points: Article copy or external URLs (Step 1 of execution).
- Boundary markers: Absent. The skill does not define delimiters to separate untrusted article content from its own instructions.
- Capability inventory: Execution of a Python script with network access to the Gemini API and local file system access (writing to
./thumbnails/). - Sanitization: No instructions provided for sanitizing or escaping the content fetched from URLs before it is processed by the agent to create compositions.
Recommendations
- AI detected serious security threats
Audit Metadata