Give your coding agent workflows to investigate LangSmith traces, build evaluation datasets, and assess LLM application quality.
LangSmith Skills equips coding agents to work with the traces, datasets, and evaluators used to understand LLM applications. Its workflows connect debugging with evaluation: inspect what an application did, turn examples into tests, and investigate results. It also covers ongoing production evaluation and custom views inside LangSmith. In Claude Code, the agent can select relevant skills automatically, or you can choose one from the plugin’s slash-command menu. The project is in early development, so APIs and skill content may change.
Ask your agent to find traces by project, time, errors, or latency, then inspect the execution tree to see the model and tool calls behind a response. The tracing skill also covers adding instrumentation to Python and JavaScript applications.
Build a dataset from exported traces and create evaluators for expected answers, tool sequences, or other quality criteria. The workflows cover inspecting actual output structures before writing checks, then testing evaluators locally or attaching them to a dataset.
Investigate an existing experiment by comparing low-scoring examples with their reference outputs, evaluator comments, and run evidence. The diagnosis workflow separates likely application failures from evaluator problems and proposes targeted fixes without changing the application or evaluation setup.
Design and test an evaluator for production runs or conversation threads, then attach a rule with deliberate filters, sampling, and spending limits. Inspect scores and execution logs to distinguish evaluator defects from application quality problems.
On LangSmith’s Enterprise plan, build a React and TypeScript Custom App that reads LangSmith data inside the workspace. The workflow covers local development, checking calculations against real data, and replicating an app into another workspace.
Create evaluation datasets from traces, upload examples, and manage existing datasets.
Create and test code or model-based evaluators for datasets and production runs.
Design and calibrate production evaluators with scoped rules, sampling, and spending limits.
Add application tracing, investigate execution trees, and export traces for analysis.
Build, verify, and replicate Custom Apps inside LangSmith; requires the Enterprise plan.
Investigate low-scoring experiment examples and propose fixes supported by run evidence.
For CLI-based tasks, install the LangSmith CLI and configure LangSmith authentication. Model-based evaluations also need access to the chosen model provider. These are requirements for using the workflows. The repository also lists Deep Agents CLI and Goose as supported agents.
Install the complete LangSmith Skills plugin through its marketplace.
In an interactive Claude Code session, add the LangSmith Skills marketplace.
/plugin marketplace add langchain-ai/langsmith-skillsIn the same session, open the plugin’s installation panel. Review the plugin and choose the installation scope.
/plugin install langsmith-skills@langsmith-skillsRead the installation summary. If it says the plugin is active, continue to the check below. If it requests a reload, closing the panel normally applies it; follow any remaining reload instructions.
Confirm availability by typing / in Claude Code and looking for the skills under the langsmith-skills plugin name.
Install the repository’s skills directly with Skills CLI. Use a terminal with npm / npx available.
Installation scope
Run this in your project’s terminal to install the repository’s skills for the current project.
npx skills add langchain-ai/langsmith-skills --skill '*' --yesOpen Claude Code in that project. If its skills directory was created during an existing session, reload the skills in that session.
/reload-skillsIn Claude Code, list available skills and confirm that the LangSmith skills appear.
/skillsRun this in your terminal to install the skills globally and target Claude Code explicitly.
npx skills add langchain-ai/langsmith-skills --agent claude-code --skill '*' --yes --globalOpen Claude Code. If its skills directory was created during an existing session, reload the skills in that session.
/reload-skillsIn Claude Code, list available skills and confirm that the LangSmith skills appear.
/skillsThe repository’s Bash installer copies the skills into Claude Code’s skills directory. It also attempts to install the LangSmith CLI if it is missing; a CLI installation failure does not stop skill installation.
In your terminal, clone the repository.
git clone https://github.com/langchain-ai/langsmith-skills.gitEnter the cloned repository in the same terminal.
cd langsmith-skillsInstallation scope
From the cloned repository, run the installer for your project. Replace ~/my-project with your project’s path, then confirm the installation when prompted.
./install.sh ~/my-projectOpen Claude Code in the target project. If its skills directory was created during an existing session, reload the skills in that session.
/reload-skillsIn Claude Code, list available skills and confirm that the LangSmith skills appear.
/skillsFrom the cloned repository, run the global installer and confirm the installation when prompted.
./install.sh --globalOpen Claude Code. If its skills directory was created during an existing session, reload the skills in that session.
/reload-skillsIn Claude Code, list available skills and confirm that the LangSmith skills appear.
/skillsInstall the skills directly using the repository’s Skills CLI instructions. Use a terminal with npm / npx available.
Installation scope
Run this in the terminal for the project you use in Cursor.
npx skills add langchain-ai/langsmith-skills --skill '*' --yesReopen the project in Cursor so it discovers the installed skills at startup. In Agent chat, type / and search for a LangSmith skill to confirm it is available.
Run this in your terminal to install the skills globally.
npx skills add langchain-ai/langsmith-skills --skill '*' --yes --globalRestart Cursor so it discovers the installed skills at startup. In Agent chat, type / and search for a LangSmith skill to confirm it is available.
Install the skills directly using the repository’s Skills CLI instructions. Use a terminal with npm / npx available.
Installation scope
Run this in the terminal for your Windsurf project.
npx skills add langchain-ai/langsmith-skills --skill '*' --yesIn the Agents window, open Customize → Skills and confirm that the LangSmith skills appear among the loaded skills.
Run this in your terminal to install the skills globally.
npx skills add langchain-ai/langsmith-skills --skill '*' --yes --globalIn the Agents window, open Customize → Skills and confirm that the LangSmith skills appear among the loaded skills.