The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The UserManager class in reference/automation.md processes external CSV files to automate system administrative tasks, creating a potential vector for indirect injection.
Ingestion points: External CSV files are ingested via the bulk_create_users function in reference/automation.md.
Boundary markers: Absent. Data from the CSV is directly mapped to command parameters without delimiters.
Capability inventory: Extensive subprocess calls (useradd, userdel, chown, systemctl), file-writes (authorized_keys), and service manipulation across reference/automation.md and reference/infrastructure.md.
Sanitization: Absent. The script does not perform validation or escaping on the input from the CSV before passing it to system shell commands via subprocess.run.
[COMMAND_EXECUTION]: Multiple reference files provide scripts and templates that execute high-privilege system commands via subprocesses.
The SelfHealingAgent in reference/automation.md automatically terminates processes based on memory usage metrics and restarts services.
Baseline configuration scripts in reference/infrastructure.md modify firewall rules using ufw and iptables, harden SSH configurations, and manage system users.
[EXTERNAL_DOWNLOADS]: The skill automates the retrieval of infrastructure binaries from trusted external sources.
The Linux baseline script in reference/infrastructure.md downloads the Prometheus node_exporter binary from its official GitHub releases page.
The Windows baseline script in reference/infrastructure.md downloads the windows_exporter MSI from the Prometheus community GitHub repository.
[PRIVILEGE_ESCALATION]: The provided automation templates and best practice guides frequently rely on elevated privileges (sudo, become: yes in Ansible, and administrative PowerShell commands). While appropriate for IT operations, these powerful capabilities require careful oversight when used in automated contexts.

it-operations