The Agent Skills Directory

[METADATA_POISONING]: The author field in the SKILL.md frontmatter identifies the author as "Black Forest Labs," which contradicts the provided skill author information. While common for API wrappers, this mismatch is technically misleading.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data which could lead to indirect prompt injection if malicious content is embedded in inputs or API responses.
Ingestion points: User-provided prompt strings in generation requests and sample URLs returned by the BFL API endpoints.
Boundary markers: No explicit delimiter or instruction-guarding markers are implemented for the prompt inputs in the provided examples.
Capability inventory: The skill facilitates network operations (via curl, requests, and fetch) and file system writes (via curl -o or language-specific file writing methods).
Sanitization: The Python and TypeScript clients include validation for image dimensions (multiples of 16, 4MP limit), but do not perform deep sanitization of the prompt strings or returned URLs.
[COMMAND_EXECUTION]: The skill provides numerous shell commands for the agent to execute, including API requests and file downloads. These are directed at the official api.bfl.ai domain and associated Azure storage for image delivery.

bfl-api