The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to read from a .env file to retrieve the BFL_API_KEY. While accessing .env is identified as a sensitive operation, its usage here is for standard local configuration of the API client. Evidence: grep BFL_API_KEY .env 2>/dev/null in references/api-key-setup.md.
[EXTERNAL_DOWNLOADS]: The skill provides instructions and scripts to download generated images from external URLs provided in the API response. Evidence: curl -s -o output.png "IMAGE_URL" in references/code-examples/curl-examples.sh and similar logic in the Python and TypeScript client examples.
[COMMAND_EXECUTION]: The documentation includes shell commands that the agent is encouraged to run to verify environment variables and manage secrets. Evidence: echo $BFL_API_KEY and export BFL_API_KEY=... in SKILL.md and references/api-key-setup.md.
[INDIRECT_PROMPT_INJECTION]: The skill serves as a proxy for user-supplied prompts and image URLs which are forwarded to external API endpoints, creating a potential attack surface. 1. Ingestion points: User input via the prompt and input_image parameters in SKILL.md and multiple reference scripts. 2. Boundary markers: The skill uses JSON structuring for API payloads but lacks explicit instructions for the agent to sanitize or ignore malicious content within user prompts. 3. Capability inventory: The skill possesses network communication capabilities and file system write access for saving images. 4. Sanitization: Dimension validation is present in the client examples, but prompt text content is passed to the API without filtering.

bfl-api