The Agent Skills Directory

[COMMAND_EXECUTION]: Provides instructions for installing the outlines library and associated backends via pip (e.g., pip install outlines, pip install vllm).
[EXTERNAL_DOWNLOADS]: References downloading model weights and configurations from well-known repositories on Hugging Face (e.g., microsoft/Phi-3-mini-4k-instruct, meta-llama/Llama-3.1-8B-Instruct) and calls to the OpenAI API.
[DYNAMIC_EXECUTION]: Demonstrates how to generate structured content like Python functions and SQL queries. Mentions the trust_remote_code flag in model configuration, which is a standard feature for loading specific model architectures from remote repositories.
[INDIRECT_PROMPT_INJECTION]: The skill documents patterns for ingesting untrusted text data into structured generators, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Untrusted text variables used in extraction examples throughout SKILL.md and references/examples.md.
Boundary markers: Often uses contextual labels (e.g., "Extract from:") to delimit input text.
Capability inventory: Demonstrates code generation (Python/SQL) and model backend orchestration.
Sanitization: Examples do not show explicit input sanitization, relying on the library's grammar-based constraints for output formatting.

outlines