dspy-fundamentals
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill documents dspy.ProgramOfThought, which generates and executes Python code at runtime. Evidence in SKILL.md and reference.md.
- [DYNAMIC_EXECUTION]: The skill describes using cloudpickle for full program serialization. Deserializing untrusted data with cloudpickle can lead to arbitrary code execution. Found in SKILL.md and reference.md.
- [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for LLM pipelines that ingest external data, creating a surface for indirect prompt injection. Ingestion points: The forward method in example_qa.py accepts untrusted input. Boundary markers: Examples do not demonstrate explicit delimiters. Capability inventory: Predictors include code generation and tool use. Sanitization: The skill recommends Pydantic models for output validation.
Audit Metadata