image-compose-api
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied image URLs and base64 data, creating a potential surface for indirect prompt injection.
- Ingestion points: Processes
image_urlsand image data (URLs or base64) provided by users as described inSKILL.md. - Boundary markers: Absent; the skill does not define specific delimiters or instructions to disregard potential commands embedded in external data.
- Capability inventory: The skill utilizes network capabilities via
curl, Pythonrequests, and Node.jsfetch, and it has the ability to read and write to the local file system (e.g., the API key file). - Sanitization: Absent; there is no mention of validating or sanitizing the external content before it is processed.
- [EXTERNAL_DOWNLOADS]: The skill is instructed to fetch extended schemas and parameter definitions from the vendor's website at
https://www.pixazo.ai/models/image-compose-api.md. - [COMMAND_EXECUTION]: The skill documentation includes an
npxcommand intended to download and add further components from the vendor's repository (Pixazo-AI/skills). - [CREDENTIALS_UNSAFE]: The skill manages a
PIXAZO_API_KEYby checking environment variables or reading from a local file (~/.pixazo/api-key). It instructs the agent to save user-provided keys to this file using restricted permissions (chmod 600), which is a standard secret management practice for the service.
Audit Metadata