The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes host-level diagnostic and configuration commands to inspect network listening ports (ss, lsof), firewall settings (ufw, firewall-cmd, nft), and OS metadata. It also facilitates persistence by scheduling periodic security audits using the openclaw cron command-line utility, which requires explicit user consent.
[DATA_EXFILTRATION]: It harvests sensitive system metadata, including disk encryption status, backup states, and network exposure levels. While these are used for hardening, the skill includes safeguards to redact secrets and requires permission for accessing sensitive files.
[EXTERNAL_DOWNLOADS]: The skill queries the public NPM registry to verify the current version status of the OpenClaw software.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interprets raw output from system utilities and openclaw audit tools as context. A compromised host could potentially manipulate these outputs to influence the agent's logic.
Ingestion points: Shell command outputs from ss, lsof, ufw, and openclaw security audit (SKILL.md).
Boundary markers: The skill lacks explicit delimiters when interpolating command results into the prompt, relying instead on user review of generated plans.
Capability inventory: The agent has capabilities to modify firewall rules, SSH configurations, and system crontabs based on these inputs.
Sanitization: Directives are included to redact tokens and secrets from persistent memory files.

healthcheck