The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents an interface for processing user-supplied natural language prompts and image references, which constitutes an indirect prompt injection attack surface. Ingestion points: User-provided values for the prompt and image parameters defined in SKILL.md. Boundary markers: The instructions provide structured prompt templates and note the presence of provider-side content safety filters. Capability inventory: The skill documents capabilities for network communication with API endpoints and local file system writes for artifact storage. Sanitization: Example implementations include validation of magic bytes, image dimensions, and download size limits, alongside recommendations for human review and rights clearance.- [SAFE]: The skill follows security best practices by recommending the use of secret managers for API tokens instead of hardcoding credentials.- [SAFE]: All network operations documented target official, well-known infrastructure belonging to the vendor and its global subsidiaries.- [SAFE]: The provided code examples prioritize security, implementing host allowlisting and credential isolation to prevent the leaking of authorization headers during artifact retrieval.

bytedance-seedream