The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources such as device configurations, command outputs, and job details. If this data contains malicious instructions, the agent might execute them using its powerful capabilities.
Ingestion points: Tools like get_device_configuration, describe_inventory, describe_job, and run_service (stdout/stderr) fetch data from the network environment.
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: High-risk capabilities include apply_device_configuration, run_command, run_service, and start_workflow.
Sanitization: No explicit sanitization or validation of the fetched content is described before the agent processes it.
[COMMAND_EXECUTION]: Several tools allow for the execution of commands and configurations on network infrastructure, which is a high-privilege action.
Evidence: run_command executes CLI commands across devices; apply_device_configuration deploys configuration changes; run_service executes gateway services.

itential-automation