The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided training examples to automatically generate few-shot demonstrations, which creates a potential surface for indirect prompt injection if the training data contains adversarial instructions.
Ingestion points: Training data is ingested via the trainset parameter in the optimizer.compile method described in SKILL.md and demonstrated in example.py.
Boundary markers: The provided examples do not implement explicit delimiters or boundary markers to distinguish training data content from the model's instructions.
Capability inventory: The skill is configured with access to the Read, Write, Glob, and Grep tools as specified in the SKILL.md frontmatter.
Sanitization: There is no evidence of content sanitization or validation performed on the training examples before they are used to build prompts for the teacher model.

dspy-bootstrap-fewshot