The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including prompts and media URLs, which are passed to remote generative models. This constitutes an attack surface for indirect prompt injection.
Ingestion points: Prompt strings and external URLs (e.g., image_url, audio_url) are passed as arguments to fal.subscribe() and fal.queue.submit() in SKILL.md.
Boundary markers: The documentation does not demonstrate the use of delimiters or specific boundary instructions to separate user data from model instructions.
Capability inventory: The skill possesses the capability to read local files via fs.readFileSync and perform network operations to transmit data to the service endpoint.
Sanitization: No sanitization, validation, or filtering of user-provided content is shown in the provided examples.
[EXTERNAL_DOWNLOADS]: The skill instructs users to download and install official client libraries from standard package registries (NPM and PyPI).
Evidence: npm install @fal-ai/client and pip install fal-client are listed as installation steps in SKILL.md.
[COMMAND_EXECUTION]: Code examples in the skill demonstrate local file system access for the purpose of selecting media files to upload to the service.
Evidence: The skill uses fs.readFileSync("audio.mp3") in the TypeScript example and fal_client.upload_file("./photo.png") in the Python example to prepare files for processing.

fal-ai