The Agent Skills Directory

External Script Installation: The skill includes a command to download and execute a setup script for a CLI tool from an official GitHub repository. This is a standard procedure for configuring the developer environment required for LangSmith evaluations.
Indirect Prompt Injection Surface: The evaluator functions in SKILL.md (e.g., accuracy_evaluator) ingest data from run.outputs and example.outputs and interpolate them directly into LLM prompts. Since this data originates from external agent runs, it presents a surface for indirect prompt injection. Ingestion points: run.outputs and example.outputs in accuracy_evaluator and trajectory_evaluator. Boundary markers: Absent. Capability inventory: Invokes LLM-as-judge calls (gpt-4o-mini). Sanitization: Absent. It is recommended to use structured output and delimiters to clearly separate instructions from data.

langsmith-evaluator