langsmith-observability

name	langsmith-observability
description	Use for interpreting LangSmith traces, adding useful metadata, and comparing agent runs.

LangSmith Observability

Use this skill when the participant asks how to inspect, debug, or improve a run.

Trace Reading Checklist

• Start at the root run and read the user input.
• Identify subagent handoffs.
• Inspect model calls for prompt shape, token usage, and latency.
• Inspect tool calls for command inputs, exit codes, output previews, and errors.
• Look for policy denials or command failures before changing prompts.
• Compare traces before and after skill or prompt edits.

Useful Tags

• workshop
• smoke-test
• sandbox-exec
• policy-demo
• skill-iteration
• debugging

Useful Metadata

• lab
• participant_group
• sandbox_name
• scenario
• expected_failure