The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill recommends installation via git clone https://github.com/AstroPilot-AI/Denario.git and docker pull pablovd/denario:latest. Neither 'AstroPilot-AI' nor 'pablovd' are on the trusted organizations or repositories list, creating a potential supply chain risk.
[COMMAND_EXECUTION] (MEDIUM): The core workflow includes den.get_results(), which is documented to 'execute computational experiments' and 'run the methodology'. This implies the execution of dynamically generated code or system-level commands based on user-supplied or AI-generated research descriptions.
[REMOTE_CODE_EXECUTION] (LOW): Documentation for Google Vertex AI setup includes curl https://sdk.cloud.google.com | bash. While this pattern (piping to bash) is high-risk, the source (Google) is trusted, resulting in a downgrade to LOW per [TRUST-SCOPE-RULE].
[PROMPT_INJECTION] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8). \n * Ingestion points: Processes untrusted external data via set_data_description and reads external methodology/results files via set_method and set_results. \n * Boundary markers: Absent; no delimiters or 'ignore instructions' warnings are provided for interpolated data. \n * Capability inventory: High-impact capabilities including computational execution (get_results), LaTeX generation (get_paper), and a web-based GUI (denario run). \n * Sanitization: No evidence of sanitization or validation of input data before it influences agent behavior or code execution.

denario