The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches installation scripts and configuration from well-known services. It references official installation methods for Tailscale and provides version-pinned download links for core networking tools like Nginx, Caddy, and HAProxy from their respective official repositories.
[COMMAND_EXECUTION]: Employs standard Linux networking utilities for system configuration and troubleshooting. Commands involving ip, nft, wg, and systemctl are used to manage interfaces, firewalls, and services. The skill emphasizes safety by requiring syntax validation (nginx -t, caddy validate) and timed rollbacks before applying permanent changes.
[INDIRECT_PROMPT_INJECTION]: Processes untrusted network data through capture tools like tcpdump and tshark. This presents a surface for potential instruction injection via malicious network packets. The skill includes instructions to verify observed state and topology before making changes, reducing the risk of accidental obedience. Evidence chain:
Ingestion points: Tool outputs from tcpdump, tshark, dig, and ss (all files).
Boundary markers: Present in SKILL.md (AI Self-Check requiring topology verification).
Capability inventory: Subprocess calls to nft, systemctl, wg, ip, and file-write operations like install and tee (all files).
Sanitization: Use of placeholders for credentials and mandatory syntax checks before reloading services.
[CREDENTIALS_UNSAFE]: Demonstrates secure handling of sensitive data by enforcing the use of placeholders for secrets like VPN private keys and VRRP passwords. It explicitly warns against hardcoding credentials and provides patterns for secure key generation using restricted file permissions (umask 077).

networking