clip
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- External Downloads (LOW): The skill instructs the user to install the CLIP package directly from a remote Git repository using
pip install git+https://github.com/openai/CLIP.git. This is an external download of executable code. Because the source is a trusted GitHub organization (OpenAI), the severity is downgraded to LOW per the trusted source policy. - Indirect Prompt Injection (LOW): The skill is designed to ingest and process untrusted data (images and text labels) through its vision-language model.
- Ingestion points:
Image.open()andclip.tokenize()are used to ingest external files and strings inSKILL.mdandapplications.md. - Boundary markers: Absent; there are no instructions to the agent to ignore potential adversarial commands embedded in the processed data.
- Capability inventory: The provided scripts only perform classification, embedding generation, and printing. There are no subprocess calls, file writes, or network operations triggered by the model's output.
- Sanitization: Absent; the content is passed directly to the model for inference.
Audit Metadata