The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill's configuration utility in scripts/utils.py embeds the sensitive BLOCKEDEN_ACCESS_KEY directly into the URL path for all API requests (https://api.blockeden.xyz/gemini/{api_key}). This is a poor security practice as API keys in URLs are likely to be recorded in server logs, proxy logs, and error messages, leading to credential exposure.
[COMMAND_EXECUTION]: In SKILL.md, the agent is instructed to download reference images provided by the user via a shell command: curl -L -o /tmp/ref_image.jpg "<url>". Because the <url> originates from untrusted user input and is passed directly to a shell command without validation or sanitization, it creates a surface for command injection or Server-Side Request Forgery (SSRF) attacks.
[DATA_EXFILTRATION]: While no user data is directly exfiltrated, the systematic exposure of the API key in URL strings constitutes an exfiltration of credentials to any intermediary log-recording systems between the agent and the API provider.
[SAFE]: The Python scripts (scripts/generate_image.py, scripts/imagen3.py, scripts/gemini_flash.py) use standard libraries for handling JSON and network requests, avoiding dynamic execution like eval() or exec().

image-gen-blockeden-gemini