The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a 'Parse-Validate-Retry' pattern (Instructor Pattern) for handling LLM output that fails schema validation. This creates an ingestion surface for potentially malicious data produced by the model. • Ingestion points: The skill instructs re-prompting the model by injecting original prompts, failed outputs, and validation errors (SKILL.md). • Boundary markers: The instructions do not explicitly mandate the use of delimiters or 'ignore embedded instructions' warnings for the re-injected data. • Capability inventory: The pattern describes an iterative generation loop which can be leveraged by sophisticated injections. • Sanitization: No sanitization or filtering of the failed model output is mentioned before its re-interpolation into the system prompt.
[EXTERNAL_DOWNLOADS]: The skill provides references and links to official documentation and source code repositories for established libraries such as Zod, Pydantic, Instructor, Outlines, and BAML. These references target well-known open-source projects and official developer documentation sites.

structured-outputs