The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/supertoken_config.py uses subprocess.run to interact with system credential managers like the macOS Keychain (security utility) and Linux Secret Service (secret-tool). These calls use static argument lists without shell execution, following security best practices for platform-integrated secret management.
[EXTERNAL_DOWNLOADS]: The skill downloads image data from remote URLs. This process is protected by multiple security layers: it enforces HTTPS, validates that IP addresses are global and not targeting private or internal networks (preventing SSRF), pins connections to resolved IPs to prevent DNS rebinding, and strictly limits the size of downloaded content.
[INDIRECT_PROMPT_INJECTION]: As the skill processes user-supplied prompts and retrieves remote image content for editing, it possesses an inherent surface for indirect prompt injection.
Ingestion points: Prompts are passed via CLI arguments; image data is retrieved from URLs in supertoken_api.py.
Boundary markers: The system relies on the underlying LLM's safety filters for the image generation prompts.
Capability inventory: The skill can perform network requests to the SuperToken API and save image files to the local file system.
Sanitization: The skill includes comprehensive sanitization logic in supertoken_api.py that redacts API keys, base64 data, and sensitive URL components from all diagnostic and error outputs.

supertoken-gpt-image-2