The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes network management tasks via Python scripts triggered by environment variables ($MCP_CALL, $PYATS_MCP_SCRIPT, etc.). This is a standard pattern for tool invocation within specific agent environments to interact with localized pyATS, ServiceNow, and GAIT integrations.
[DATA_EXFILTRATION]: The skill accesses sensitive network device information, including running configurations and system logs. This data access is essential for its primary function of capturing pre-change baselines and performing post-change diffing to ensure network stability.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external network devices (logs, command output) to verify change success, which represents a potential injection surface. However, this risk is mitigated by the skill's professional design, which requires a multi-phase workflow, ServiceNow approval gates, and human oversight for change documentation.
Ingestion points: Data is pulled into the agent context through device show commands and log retrieval in SKILL.md.
Boundary markers: The instructions do not define specific text delimiters for untrusted device output.
Capability inventory: The skill utilizes capabilities to modify device configurations and update ServiceNow change requests.
Sanitization: The skill does not specify sanitization steps for the data returned from the network devices.

pyats-config-mgmt