The Agent Skills Directory

[Data Exposure & Exfiltration] (LOW): The skill performs network operations to api.worldlabs.ai and storage.googleapis.com which are not on the predefined whitelist for data exfiltration analysis. These calls are essential for the skill's primary purpose of API integration.
Evidence: Multiple curl examples and a Python requests script target https://api.worldlabs.ai/marble/v1 and https://storage.googleapis.com/.
[Indirect Prompt Injection] (LOW): The skill accepts user-provided text prompts and interpolates them directly into API requests, creating a surface for indirect prompt injection if the external service processes these inputs without further validation.
Ingestion points: world_prompt in SKILL.md and the prompt argument in the Python generate_world function.
Boundary markers: None detected; instructions are directly placed in the JSON body.
Capability inventory: Bash (curl) and WebFetch are authorized tools used to transmit this data.
Sanitization: No sanitization or escaping of the user-provided prompt is performed before interpolation.
[Command Execution] (LOW): The skill documentation provides and encourages the use of Bash and curl for interacting with the remote API, which is a standard but powerful capability.
Evidence: The allowed-tools section includes Bash and the body contains several curl command templates.

world-labs-api