The Agent Skills Directory

[NO_CODE]: The analyzed skill does not distribute any executable scripts or active binaries, consisting entirely of markdown reference documentation and configuration examples.
[EXTERNAL_DOWNLOADS]: The installation instructions guide users to download software from external unverified repositories, including a GitHub repository under the account 'AstroPilot-AI' and a Docker image from a personal Docker Hub namespace ('pablovd/denario:latest').
[INDIRECT_PROMPT_INJECTION]: The documented multi-agent pipeline possesses an indirect prompt injection attack surface.
Ingestion points: Untrusted external data, text descriptions, and raw methodologies enter the system via set_data_description(), set_method(), and set_results() functions.
Boundary markers: There are no documented boundary delimiters or instructions advising the model to disregard embedded commands within incoming scientific data descriptions.
Capability inventory: The architecture includes an Execution Agent authorized to dynamically generate and run Python analysis code using libraries like pandas, scikit-learn, and matplotlib.
Sanitization: There is no indication of strict input validation or text escaping before interpolation into the multi-agent LLM prompts.

denario