aliyun-qwen-ocr
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data from image sources and user-defined prompts, creating a surface for indirect prompt injection where adversarial content in images could influence model behavior.
- Ingestion points:
imageandpromptarguments inscripts/prepare_ocr_request.py. - Boundary markers: None present to delimit untrusted input within generated requests.
- Capability inventory: The script writes generated JSON payloads to the local filesystem.
- Sanitization: No content validation or sanitization is performed on inputs.
- [CREDENTIALS_UNSAFE]: The documentation identifies a sensitive local path (~/.alibabacloud/credentials) for API key storage. While this is instructional for user configuration and the provided script does not programmatically access this path, the explicit mention of sensitive credential locations is noted.
Audit Metadata