The Agent Skills Directory

Installation Script from Vendor Repository: The skill fetches an installation script for the LangSmith CLI tool directly from the vendor's official GitHub repository (langchain-ai/langsmith-cli). This is a standard method for deploying the necessary tooling to interact with the LangSmith platform.
Command Line Interface (CLI) Usage: The skill utilizes the langsmith CLI tool to manage evaluators, including listing, uploading, and deleting them. These operations are essential for maintaining evaluation datasets and project-level quality checks.
Agent Output Processing Surface: The skill defines evaluators that ingest data from agent execution traces (run.outputs) to perform quality assessments (LLM-as-Judge). As these evaluators process potentially untrusted outputs generated by AI agents, they represent an indirect prompt injection surface.
Ingestion points: Evaluator functions like accuracy_evaluator and trajectory_evaluator in SKILL.md access data via run.outputs.
Boundary markers: No specific delimiters or "ignore instructions" directives are present in the provided templates for processing these outputs.
Capability inventory: The evaluators are designed to return scores and comments; they do not possess capabilities for file writing, network exfiltration, or arbitrary command execution.
Sanitization: The skill does not include specific logic for sanitizing or filtering agent outputs before they are processed by the judge model.

langsmith-evaluator