The Agent Skills Directory

[COMMAND_EXECUTION]: The execute_junos_command and execute_junos_command_batch tools permit execution of arbitrary CLI commands on target network devices. \n- [COMMAND_EXECUTION]: The load_and_commit_config tool allows pushing arbitrary configuration changes to a device's running configuration via NETCONF. \n- [CREDENTIALS_UNSAFE]: The skill manages sensitive authentication data including passwords and SSH private keys via a local JSON inventory file. While the get_router_list tool includes logic to filter these credentials from outputs, the storage and handling of these materials present an inherent risk surface. \n- [DYNAMIC_EXECUTION]: The render_and_apply_j2_template tool utilizes Jinja2 for dynamic configuration generation, which executes template logic at runtime based on provided variables. \n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted template and variable data that can influence device configuration. 1. Ingestion points: render_and_apply_j2_template parameters in SKILL.md. 2. Boundary markers: Absent from the prompt interpolation logic. 3. Capability inventory: Remote command execution and configuration application on JunOS hardware. 4. Sanitization: Implements regex-based filters (block.cmd and block.cfg) to prevent destructive commands and root credential modifications. \n- [EXTERNAL_DOWNLOADS]: Fetches the JunOS MCP server and required dependencies from Juniper's GitHub repository and official package registries.

junos-network