The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the automated ingestion of data from external cloud storage providers (AWS S3 and Google Cloud Storage).
Ingestion points: Data is pulled from external buckets into the Roboflow workspace via datasource_trigger and connect_cloud_storage.
Boundary markers: The instructions do not define specific delimiters or warnings for the agent when processing the content or metadata of the mirrored images.
Capability inventory: The skill allows triggering mirror jobs, defining glob rules for selection, and validating bucket access. While the skill itself manages the pipeline, the imported data is intended for downstream processing by other agent tools.
Sanitization: There is no evidence of sanitization for object keys, metadata, or file content that might contain instructions targeting the agent.
[SAFE]: The skill implements strong credential security measures.
Credential Management: It explicitly avoids passing secrets like API keys or service account JSON through tool arguments. Instead, it utilizes MCP elicitation for out-of-band secret collection, ensuring sensitive data never enters the model context or chat transcript.
Data Masking: The credentials_list tool is designed to return only masked, non-secret fields for existing credentials.

roboflow-cloud-storage