The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from multiple sources, creating a potential surface for indirect instructions.
Ingestion points: The skill accepts user-provided prompt values in the tool.execute calls. It also reads external data from the model_catalog via tool.get_info() and processes artifacts downloaded from the Atlas Cloud gateway.
Boundary markers: The instructions do not define delimiters or specific "ignore embedded instructions" warnings for the agent when handling the user prompt or external model metadata.
Capability inventory: The skill possesses significant capabilities, including performing network operations (uploading to Atlas, polling predictions, downloading files) and writing data to the local file system within the projects/ directory.
Sanitization: There is no evidence of sanitization, validation, or escaping of the user-provided prompt or the external data retrieved from the cloud provider before it is used in tool execution.

atlas-cloud