The agent development loop with LangSmith + Claude Code / Deepagents

LangSmith provides a system of record for traces, which can include long agent trajectories. Code agents like Claude Code or Deep Agents CLI can make use of traces to improve agent performance via prompt optimization, improve tool design, or better context engineering approaches. Here, we showcase langsmith-fetch, a simple CLI that pairs well with Claude Code, Deepagents, or other code agents. It give code agents easy access to traces, and can be added to either as a Skill.

LangSmith Fetch -- 
https://github.com/langchain-ai/langsmith-fetch

Chapters
0:00 - Debugging feedback loops with LangSmith and code agents
0:48 - Setting up LangSmith Fetch: Installation and basic usage
1:24 - Claude Code integration: Running agents and fetching traces
2:36 - Creating a Claude Code skill for LangSmith Fetch
3:28 - Testing the skill in a new Claude Code session
3:51 - Deep Agents CLI setup: Configuring dual tracing projects
5:00 - Running an agent with Deep Agents and viewing separate traces
6:06 - Using LangSmith Fetch within Deep Agents CLI
6:54 - Creating skills for Deep Agents
7:36 - Recap: Why connecting LangSmith with code agents matters