skills/lebsral/dspy-programming-not-prompting-lms-skills/dspy-program-of-thought/Gen Agent Trust Hub
dspy-program-of-thought
Warn
Audited by Gen Agent Trust Hub on Mar 17, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill provides instructions and worked examples for using the dspy.ProgramOfThought module to generate and execute Python code based on language model reasoning. This pattern allows for arbitrary logic execution driven by AI model outputs.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data into a process that generates executable code. Ingestion points: Input fields such as sales_data, question, and problem defined in SKILL.md and examples.md. Boundary markers: None are implemented or recommended in the examples to separate data from instructions. Capability inventory: Execution of Python code through the ProgramOfThought module. Sanitization: The skill relies on external library guarantees of a restricted sandbox environment without providing or demonstrating input validation or escaping mechanisms.
Audit Metadata