The Agent Skills Directory

Data Ingestion Surface: The dataset generation scripts ingest trace data from external JSONL files located at user-specified paths. Evidence Chain: (1) Ingestion points: load_traces_from_dir and load_traces_from_file in generate_datasets.py, and loadTracesFromDir and loadTracesFromFile in generate_datasets.ts. (2) Boundary markers: Absent. (3) Capability inventory: Network upload via export_to_langsmith and file write via export_to_file in generate_datasets.py, and equivalent operations in generate_datasets.ts. (4) Sanitization: Absent. This process involves extracting inputs and outputs from these files to construct datasets. As the content is processed without explicit sanitization or boundary markers, it represents a surface where data could influence the resulting evaluation datasets.
Service Connectivity: The skill interacts with the LangSmith service to manage and upload datasets. This involves the use of an API key and the transmission of extracted trace data to the platform, which is consistent with the skill's purpose as a developer tool for evaluation management.
File System Interaction: The skill performs routine file operations, including reading from source directories and writing output files to local paths. It also includes logic for overwriting existing files or datasets, which is controlled by user-provided flags.

langsmith-dataset