The Agent Skills Directory

Command Execution (HIGH): The skill provides examples of using the shell and command modules (e.g., ansible all -m shell -a "df -h | grep /dev"), which allow for the execution of arbitrary commands on target systems managed by the agent.- Privilege Escalation (HIGH): The skill explicitly demonstrates and encourages the use of the become: yes parameter and the --become command-line flag, granting the agent root-level administrative access to target systems.- Indirect Prompt Injection (HIGH): The skill's primary function is to interpret and execute external data files. ● Ingestion points: Reads playbook.yml, inventory files, and requirements.yml. ● Boundary markers: None identified. ● Capability inventory: Full system control through various Ansible modules (apt, service, shell, git, pip). ● Sanitization: None. An attacker providing a malicious playbook or inventory file could achieve Remote Code Execution (RCE) via the agent.- Unverifiable Dependencies (MEDIUM): The skill uses ansible-galaxy to install roles from community repositories (e.g., geerlingguy.nginx) and the git module to clone code from external URLs. These sources do not belong to the predefined trusted list and could lead to the execution of untrusted third-party code.- Dynamic Execution (MEDIUM): The use of Jinja2 templates (.j2) for configuration generation (e.g., template: src=nginx.conf.j2) introduces a risk of template injection if variables used within the templates are sourced from untrusted external inputs.

ansible