fal-video-generation
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to fetch OpenAPI specification files from the vendor domain
fal.aiand process them using a Python script. This is used to dynamically resolve model contracts and parameter constraints. As the domain belongs to the authorized vendor infrastructure, this is a legitimate operational pattern for the skill's purpose. - [EXTERNAL_DOWNLOADS]: The instructions permit the installation of publicly documented dependencies required for media rendering and processing. The skill mandates that these installations must occur only within the user's authorized environment and requires the agent to stop and report if a dependency is missing rather than attempting unverified installation.
- [DYNAMIC_EXECUTION]: The production guide provides an embedded Python snippet used for parsing JSON data from the vendor's API. This script allows the agent to inspect model input schemas at runtime to ensure compatibility and parameter correctness.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data including user prompts, documentation files, and vendor-provided API schemas. This represents a typical vulnerability surface for media generation agents.
- Ingestion points: User-supplied motion briefs and prompts in SKILL.md, the supporting production-guide.md, and dynamic OpenAPI JSON data fetched from fal.ai.
- Boundary markers: The skill relies on structured API payloads and internal routing logic to manage data flow, though it does not mandate specific natural language delimiters for user input.
- Capability inventory: Use of
curlfor network requests,python3for data parsing,genmediafor model interaction, andffprobefor media inspection across both analyzed files. - Sanitization: The skill emphasizes strict schema validation and parameter checking against vendor specifications to ensure generated requests conform to expected model contracts.
Audit Metadata