The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Ansible playbooks, roles, and inventory data, which presents a surface for indirect prompt injection via malicious task definitions or variables.
Ingestion points: The skill is intended to review and debug user-provided automation code (playbooks, roles) and dynamic inventory data as specified in SKILL.md.
Boundary markers: Includes a dedicated "AI Self-Check" and "Production Checklist" in SKILL.md that directs the agent to validate generated or reviewed content against a set of security and idiomatic rules.
Capability inventory: The reference files (e.g., references/playbook-patterns.md) document extensive capabilities including shell command execution (ansible.builtin.shell), network requests (ansible.builtin.uri), and file system manipulation.
Sanitization: The skill provides strong guidance on sanitizing sensitive output using no_log: true and protecting secrets with ansible-vault (documented in references/vault-and-secrets.md).
[EXTERNAL_DOWNLOADS]: The skill references several external resources and dependencies common to the Ansible ecosystem.
Evidence: Recommends the use of ansible.posix, community.general, and ansible-lockdown collections for compliance and hardening, and provides instructions for fetching images from official registries like docker.io and ghcr.io for testing.
[COMMAND_EXECUTION]: Provides guidance and templates for executing shell commands when specialized Ansible modules are unavailable.
Evidence: references/playbook-patterns.md includes examples of ansible.builtin.command and ansible.builtin.shell with explicit instructions on ensuring idempotency via creates, removes, and changed_when attributes.

ansible