The Agent Skills Directory

[COMMAND_EXECUTION]: Executes administrative system commands for security auditing and host hardening.
Evidence: The skill instructions include the execution of ufw, firewall-cmd, nft, pfctl, and /usr/libexec/ApplicationFirewall/socketfilterfw to check and modify host security postures.
Evidence: Utilizes the openclaw cron add command to schedule periodic security audits, creating a form of task-based persistence.
[EXTERNAL_DOWNLOADS]: Communicates with external package registries for version verification.
Evidence: The skill instructions include running npm view openclaw version to check the current published version against the local installation.
[DATA_EXFILTRATION]: Accesses sensitive host metadata to generate security reports.
Evidence: Gathers OS details, privilege levels, listening ports, network exposure, and disk encryption status. The skill includes explicit instructions to redact secrets, tokens, and credentials from all logs and memory files.
[PROMPT_INJECTION]: Contains an indirect prompt injection surface by ingesting and processing output from untrusted system commands.
Ingestion points: Command output from ss, lsof, uname, and openclaw security audit is fed back into the agent's context in SKILL.md (Workflow Step 1 & 2).
Boundary markers: No specific delimiters or markers are defined to isolate command output from agent instructions.
Capability inventory: Shell command execution, persistence via cron, and filesystem write access to the memory/ directory.
Sanitization: Includes instructions to redact sensitive host details from the final reports.

healthcheck