The Agent Skills Directory

[COMMAND_EXECUTION]: The tool pyats_run_linux_command allows the agent to execute arbitrary shell commands on target Linux devices within the testbed environment.
[REMOTE_CODE_EXECUTION]: The pyats_run_dynamic_test tool enables the execution of Python AEtest scripts provided as strings. Although the instructions list several banned imports and functions (such as os, subprocess, and eval), this dynamic execution path remains a sensitive capability.
[DATA_EXFILTRATION]: Several tools, including pyats_show_running_config, pyats_show_logging, and pyats_list_devices, access and return highly sensitive infrastructure data, such as full device configurations and credential summaries.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the processing of untrusted device output.
Ingestion points: Data enters the context through pyats_run_show_command, pyats_show_running_config, and pyats_show_logging.
Boundary markers: There are no specified delimiters or instructions to ignore embedded commands within the device outputs.
Capability inventory: The skill possesses high-impact capabilities including device configuration (pyats_configure_device) and shell execution (pyats_run_linux_command).
Sanitization: The instructions do not mention any sanitization or filtering of data retrieved from network devices before it is processed by the agent.

pyats-network