The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for interpolating untrusted text directly into LLM prompts without explicit security boundaries or delimiters.
Ingestion points: The skill ingests untrusted text via variables interpolated into messages in client.messages.create calls in SKILL.md and references/examples.md.
Boundary markers: Prompt examples lack explicit delimiters like XML tags or clear instructions to ignore instructions embedded in the source text.
Capability inventory: The skill is limited to LLM API calls and does not demonstrate dangerous local capabilities such as file system writes or subprocess execution.
Sanitization: While Pydantic provides structural validation of the output, it does not sanitize the input text for potential prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the instructor library and its dependencies (pydantic, openai, anthropic) from official package registries. These are widely used, well-known, and established libraries in the AI development ecosystem.
[DYNAMIC_EXECUTION]: The skill utilizes Pydantic features such as create_model and various validators to dynamically define data structures at runtime. This is a standard and expected use of the library to ensure type safety and output verification.

instructor