image-gen
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DATA_EXFILTRATION]: The script
scripts/generate_image.pyreads sensitive credentials from~/.openclaw/openclaw.json. The implementation retrieves the first available API key without verifying its association with the target service, potentially leading to credential exposure to third-party endpoints.\n- [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it lacks sanitization for inputs used in sensitive operations.\n - Ingestion points: Input prompt (
-p) and output path (-o) arguments inSKILL.mdare passed directly to the script.\n - Boundary markers: No delimiters or safety instructions are present to mitigate injection in the processed data.\n
- Capability inventory: The skill performs network requests and writes binary data to the local filesystem.\n
- Sanitization: The output path argument is not validated, allowing for potential arbitrary file writes or directory traversal if the agent is manipulated by untrusted external data.\n- [EXTERNAL_DOWNLOADS]: Fetches generated image content from Alibaba Cloud's DashScope API service.
Recommendations
- AI detected serious security threats
Audit Metadata