The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system-level utilities including 'ip tunnel show' and 'ip addr show' via the 'gre_tunnel_status' tool to monitor network interfaces.
[COMMAND_EXECUTION]: Participation in routing protocols like BGP (TCP/179) and OSPF (IP/89) requires the use of raw sockets, which necessitates root or administrative privileges on the host system.
[COMMAND_EXECUTION]: The provided installation and lab setup workflows require the execution of shell scripts with elevated permissions, specifically 'sudo bash scripts/setup-gre.sh'.
[COMMAND_EXECUTION]: The skill provides tools for direct manipulation of the host's routing table and network topology, such as 'bgp_inject_route' and 'ospf_adjust_cost'. It includes a 'NETCLAW_LAB_MODE' environment variable that intentionally bypasses safety checks like ServiceNow CR approval.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by ingesting untrusted routing data from external network peers. 1. Ingestion points: 'bgp_get_rib' and 'ospf_get_lsdb' fetch routing attributes and link-state advertisements from the network. 2. Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions for this ingested data. 3. Capability inventory: The agent has high-impact capabilities including route injection, metric manipulation, and system command execution. 4. Sanitization: No sanitization or validation of the ingested routing data is mentioned, potentially allowing an attacker to inject instructions into text-based routing fields.

protocol-participation