The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions direct users to download code from a third-party GitHub repository (AstroPilot-AI/Denario) and a Docker image from an external account (pablovd/denario).
[REMOTE_CODE_EXECUTION]: The get_results() method dynamically executes code generated by AI agents to perform data analysis and create visualizations, which presents a risk of executing unvetted or malicious logic.
[COMMAND_EXECUTION]: The skill provides a CLI command denario run to launch a local web server (Streamlit) for the application interface.
[PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection via the set_data_description input. Maliciously crafted data descriptions could influence the agents to generate and execute dangerous code during the analysis phase.
Ingestion points: The set_data_description method in references/research_pipeline.md accepts arbitrary text data.
Boundary markers: No explicit boundary markers or safety instructions are provided to separate user data from agent instructions.
Capability inventory: The execution agent has the capability to run Python code via get_results().
Sanitization: There is no evidence of input validation or sanitization before the data description is processed by the agents.

denario