The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill operates using a ReAct loop that autonomously generates and executes Python code based on its analysis of a modeling problem. This pattern of generating and then running scripts is central to the skill's functionality but carries inherent risks associated with executing dynamically created logic.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, which serves as a potential vector for indirect prompt injection. 1. Ingestion points: Problem PDF/text files and raw data files provided by the user, as well as academic content retrieved via web search and fetch tools. 2. Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its core operating instructions and potentially adversarial content. 3. Capability inventory: The skill possesses significant execution capabilities, including running local Python scripts, shell commands for document compilation, and the ability to write and execute its own modeling code. 4. Sanitization: The instructions do not define any sanitization or validation processes for the data ingested from external sources.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands and Python scripts to manage the workspace, generate images, and compile the final PDF report.

mcm-master