The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill documentation includes the use of the --become flag and sudo for tasks requiring administrative rights, such as installing system packages (nginx) and managing services. This is standard functionality for configuration management tools.
[EXTERNAL_DOWNLOADS]: Examples are provided for installing third-party roles via ansible-galaxy and cloning repositories using the git module. These are routine operations for deploying software and dependencies.
[DYNAMIC_EXECUTION]: The skill mentions dynamic inventory support, specifically citing the execution of a Python script (inventory.py) to generate host lists at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill describes how to build playbooks and templates using variables and loops. If the agent populates these templates using untrusted data from external sources without proper sanitization, it could lead to the creation of unintended configurations.
Ingestion points: User-defined variables in vars/main.yml and inventory files.
Boundary markers: None explicitly defined in the provided templates.
Capability inventory: File writes (copy, template), package installation (apt, yum, pip), and command execution (shell, command).
Sanitization: Not demonstrated in the templates, though the skill promotes using ansible-vault for encrypting sensitive data.

ansible