cloudrelay-imagegen
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill implements a self-update mechanism that fetches release metadata and assets from the vendor's official GitHub repository (api.github.com/repos/CloudRelay-Code/cloudrelay-imagegen-skill).
- [REMOTE_CODE_EXECUTION]: The update process downloads compressed archives from GitHub, validates them, and replaces the skill's own executable scripts. Security measures are included, such as SHA-256 integrity checks, path traversal prevention, and syntax validation before files are overwritten.
- [DATA_EXFILTRATION]: The skill manages and stores API keys for GPT, Gemini, and Grok services in local configuration files (e.g., ~/.config/cloudrelay/) or the Windows Registry. These keys are transmitted to the vendor's API endpoint (cloudrelay.cn) to perform image generation. The skill includes logic to redact these keys from console output and logs.
- [DYNAMIC_EXECUTION]: During the update process, the scripts/_update_common.py module uses the compile() function to verify the syntax of downloaded Python scripts as a safety check before they are integrated into the runtime directory.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a configuration handoff process in SKILL.md that invites users to provide API keys directly via the chat interface if terminal access is unavailable. The agent is instructed to pass this input to a local script via stdin for secure processing, creating an ingestion surface for sensitive data.
Audit Metadata