The Agent Skills Directory

[COMMAND_EXECUTION]: The get_results() function performs computational analysis by executing code generated by the agent. This allows the system to run arbitrary Python scripts for data processing, statistical modeling, and visualization directly in the user's environment.
[REMOTE_CODE_EXECUTION]: The research pipeline generates and executes code based on user-provided data descriptions and methodologies. If these inputs contain malicious directives or are sourced from untrusted literature, the agent may generate and run harmful code during the analysis phase.
[EXTERNAL_DOWNLOADS]: The skill directs users to install software from unverified external sources, including a GitHub repository under the 'AstroPilot-AI' organization and a Docker image from a personal account ('pablovd').
[PROMPT_INJECTION]: The system is vulnerable to indirect prompt injection through its data ingestion points. Untrusted information from datasets or literature searches can influence the 'Idea' and 'Method' agents to produce biased or malicious outputs which are then processed by downstream execution tools.
Ingestion points: set_data_description method in SKILL.md and data source inputs in references/examples.md.
Boundary markers: No specific delimiters or safety instructions are used to isolate untrusted data within the prompts.
Capability inventory: The system possesses file-writing and code execution capabilities via the Execution Agent described in references/research_pipeline.md.
Sanitization: The documentation does not specify any sanitization, filtering, or validation steps for external data processed by the multi-agent system.

denario