The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the langsmith package via pip. This package is provided by the official LangChain organization, which is a recognized and trusted source in the AI ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the evaluation of LLM outputs, particularly through patterns like 'LLM-as-a-judge' shown in references/advanced-usage.md. This creates an attack surface where untrusted data could influence the agent's evaluation logic.
Ingestion points: Data ingested through evaluate() and aevaluate() functions, which process production traces or user-defined datasets.
Boundary markers: The example llm_judge_evaluator prompt lacks explicit delimiters or instructions to ignore embedded commands within the Model Response being evaluated.
Capability inventory: The skill utilizes network capabilities to communicate with LangSmith and OpenAI APIs.
Sanitization: While the skill documentation provides an example of using process_inputs for data masking, the primary evaluation examples do not demonstrate sanitization against prompt injection.
[DATA_EXPOSURE]: In references/troubleshooting.md, the documentation suggests setting the environment variable LANGSMITH_VERIFY_SSL to false as a solution for SSL certificate errors. Disabling certificate verification is a security risk that exposes network traffic to potential Man-in-the-Middle (MITM) attacks.
[COMMAND_EXECUTION]: The skill includes instructions for managing files and datasets, including an example in references/advanced-usage.md using the @traceable(dangerously_allow_filesystem=True) decorator. This is a standard SDK feature for attaching local files to traces, but it explicitly acknowledges the capability to interact with the host filesystem.

langsmith-observability