The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents the use of the dspy.ProgramOfThought module, which is designed to generate and execute Python code at runtime to perform computations. Additionally, code examples for building tools (e.g., a math calculator in references/examples.md) utilize the eval() function to process mathematical expressions.
[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building complex AI systems, such as Retrieval-Augmented Generation (RAG) and ReAct agents, that process external context and user input. The documentation identifies an attack surface where malicious instructions embedded in ingested data could influence agent behavior, particularly when combined with capabilities like tool execution and code generation.
Ingestion points: Untrusted data enters the context through dspy.InputField() in signatures like SearchQA and ResearchAgent, and via the dspy.Retrieve module.
Boundary markers: The provided examples do not explicitly show the use of delimiters or 'ignore' instructions for external data, relying on internal library prompting logic.
Capability inventory: The skill facilitates tool execution via dspy.ReAct, code execution via dspy.ProgramOfThought, and mathematical evaluation via eval().
Sanitization: While one example uses an empty __builtins__ dictionary for eval(), other examples show direct evaluation of strings.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions to install the DSPy framework from official sources, including the Python Package Index (PyPI) and Stanford NLP's public GitHub repository.
[COMMAND_EXECUTION]: The SKILL.md file contains standard shell commands for package installation using pip.

dspy