The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The auditing script scripts/agent.py accesses sensitive file paths, including /etc/shadow, /etc/passwd, and /etc/ssh/sshd_config. While the script primarily checks permissions and standard configurations, these files contain authentication metadata and sensitive server settings.
[COMMAND_EXECUTION]: The instruction file SKILL.md contains numerous shell commands intended for system modification. These include editing system configuration files (e.g., /etc/fstab, /etc/modprobe.d/CIS.conf), modifying kernel parameters via sysctl, and reconfiguring the system firewall with ufw.
[DYNAMIC_EXECUTION]: The audit agent in scripts/agent.py implements a run_cmd function that utilizes subprocess.check_output with shell=True for commands containing shell operators. This allows for the dynamic execution of shell commands like modprobe, findmnt, and systemctl based on the script's hardcoded logic.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of security packages, such as chrony, auditd, and openscap-scanner, from official distribution repositories using the apt package manager. It also provides links to official documentation and repositories from trusted organizations such as CIS, NIST, and Ansible Lockdown.
[INDIRECT_PROMPT_INJECTION]: The skill features a results processor in scripts/process.py that ingests external data.
Ingestion points: scripts/process.py parses OpenSCAP XCCDF results from XML files provided as input.
Boundary markers: None identified in the XML parsing logic.
Capability inventory: The skill has the capability to execute shell commands, modify critical system files, and install software packages.
Sanitization: The script uses standard XML iteration but does not explicitly implement safeguards against injection within the XML content or malicious XML structures.

hardening-linux-endpoint-with-cis-benchmark