The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute a wide range of shell commands on a remote pfSense firewall via SSH. While intended for management, the 'raw' command (python3 .claude/pfsense.py raw "<command>") allows the execution of any arbitrary shell command with administrative privileges.
[REMOTE_CODE_EXECUTION]: The skill includes a dedicated command to execute arbitrary PHP code on the firewall (python3 .claude/pfsense.py php "<code_snippet>"). Since pfSense uses PHP for its core configuration and logic, this provides 'God mode' access to the appliance's internals and data.
[DATA_EXFILTRATION]: The backup command (python3 .claude/pfsense.py backup) dumps the full config.xml file to stdout. In pfSense, this file contains the complete system configuration, including sensitive data such as password hashes, certificate private keys, VPN shared secrets, and API keys. Exposure of this file to the agent environment constitutes a high-risk data exposure.
[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection by instructing the agent to read and filter firewall logs (logs-filter) and Snort IDS alerts (snort). Malicious network traffic could be crafted to insert instructions into these logs that the agent might inadvertently execute if it lacks sufficient output sanitization or boundary markers.
Ingestion points: pfSense firewall logs, Snort alerts, and DHCP lease tables.
Boundary markers: None provided in the instructions.
Capability inventory: Full shell access (raw), PHP execution (php), and administrative SSH commands.
Sanitization: None specified for processing external log data.

pfsense