The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages sensitive authentication credentials by reading and writing OAuth tokens to a hardcoded local path at ~/.codex/auth.json (shared with the Codex CLI) and resolving provider-specific API keys from environment variables and local .env files.
[EXTERNAL_DOWNLOADS]: The implementation performs remote fetching of image content from external provider-generated URLs, as seen in the BytePlus and DashScope integration logic.
[DATA_EXFILTRATION]: The skill transmits user-supplied prompts and reference images to various third-party AI platforms including OpenRouter, Google Gemini, and OpenAI to perform its primary generation and analysis tasks.
[PROMPT_INJECTION]: The skill processes untrusted input data (text prompts and image files) through external vision and generation models, presenting a surface for indirect prompt injection. * Ingestion points: create_image.py (prompt input) and read_image.py (image file path). * Boundary markers: No explicit instructions for the agent to ignore embedded commands in processed data. * Capability inventory: Network access via the requests library and local file system write permissions. * Sanitization: Filename slugification is used for output management.

media-tools