The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires downloading a repository from an external, non-trusted third-party source: https://github.com/sibilleb/AAP-Enterprise-MCP-Server.
[REMOTE_CODE_EXECUTION]: The installation instructions involve cloning an external repository and executing a Python script (uv run eda.py), which facilitates the execution of code from a source outside the control of the skill author or trusted organizations.
[DYNAMIC_EXECUTION]: The skill functions by dynamically executing a Python script (eda.py) located in the cloned external repository to serve as the MCP backend.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from an external Ansible EDA server, including rulebooks and activation details, which represents an attack surface for indirect prompt injection. 1. Ingestion points: Untrusted data enters the agent context via list_activations, get_activation, list_rulebooks, get_rulebook, and list_event_streams. 2. Boundary markers: There are no instruction delimiters or explicit warnings for the agent to ignore instructions embedded within rulebooks or activation metadata. 3. Capability inventory: The skill provides capabilities to create_activation, enable_activation, disable_activation, restart_activation, delete_activation, and create_decision_environment. 4. Sanitization: There is no evidence of sanitization or filtering of the content retrieved from the EDA API before it is presented to the agent.

aap-eda