The Agent Skills Directory

[COMMAND_EXECUTION]: Executes extensive system auditing commands to gather host context, including network listening status (ss, lsof), OS identification (uname, sw_vers, /etc/os-release), and firewall configurations (ufw, firewall-cmd, nft, socketfilterfw, pfctl). These are used for read-only assessment before presenting a plan to the user.
[COMMAND_EXECUTION]: Utilizes vendor-specific CLI tools (openclaw security audit, openclaw update status) for deep security scanning and version management of the OpenClaw installation.
[EXTERNAL_DOWNLOADS]: Performs version checks via npm view and the vendor's update status command to ensure the system is running current software.
[PROMPT_INJECTION]: Contains an indirect prompt injection surface (Category 8) as the agent ingests untrusted system data (e.g., process names, firewall rules, command outputs) during audits.
Ingestion points: System command outputs from uname, ss, lsof, and openclaw audit logs (SKILL.md).
Boundary markers: Absent; the agent directly interprets the output of the executed commands.
Capability inventory: Extensive shell execution capabilities including firewall modification, SSH configuration changes, and persistence creation via openclaw cron (SKILL.md).
Sanitization: Not explicitly implemented for system outputs. However, the risk is mitigated by a strict requirement for explicit user approval before any state-changing command is executed.

healthcheck