The Agent Skills Directory

[SAFE]: The code samples and instructional text are consistent with legitimate use of the DSPy library for model optimization.- [PROMPT_INJECTION]: The skill demonstrates how to process training datasets to generate and select few-shot demonstrations. This architectural pattern represents a surface for indirect prompt injection if the datasets provided by the user contain adversarial instructions.
Ingestion points: Training examples (trainset) and evaluation examples (devset) in SKILL.md and examples.md.
Boundary markers: None identified; inputs are treated as standard data objects.
Capability inventory: The skill uses dspy.LM to communicate with language models and optimized_qa.save() to write the resulting program state to local storage.
Sanitization: No sanitization or input filtering is demonstrated in the examples; the process assumes trusted input data.

dspy-bootstrap-few-shot