The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation includes shell commands for local lab environment setup.
Evidence: Commands such as docker compose up -d and bash scripts/verify.sh are provided for managing the FRR testbed.
[PRIVILEGE_ESCALATION]: The skill requires administrative privileges for network configuration and protocol participation.
Evidence: Instructions for the lab environment include sudo bash scripts/setup-gre.sh to configure tunnels.
Evidence: Safety rules explicitly note that raw sockets for BGP (TCP/179), OSPF (IP/89), and GRE (IP/47) protocol speakers require root privileges.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external routing protocols which could theoretically contain malicious instructions or malformed data influencing agent behavior.
Ingestion points: bgp_get_rib and ospf_get_lsdb fetch data from external routing peers and protocol databases.
Boundary markers: Not identified in the instructions.
Capability inventory: The skill has tools to modify routing states via bgp_inject_route, bgp_withdraw_route, and ospf_adjust_cost.
Sanitization: No explicit sanitization or validation of the ingested routing data is documented.
[TIME_DELAYED_CONDITIONAL]: The skill uses a configuration flag to conditionally relax safety requirements.
Evidence: The environment variable NETCLAW_LAB_MODE=true is used to bypass mandatory ServiceNow Change Request approval processes designed for production safety.

protocol-participation