Skip to content
Agentmart
Submit a tool

langsmith-review

Review LangSmith traces, runs, threads, and OpenCode or Paseo sessions through the read-only helper.

Skill by tobywise

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/tobywise-langsmith-review

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Review LangSmith traces, runs, threads, and OpenCode or Paseo sessions through the read-only helper. Use when asked to inspect, diagnose, summarize, compare, or query LangSmith activity, latency, token usage, tool calls, or errors.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Node.js package

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/langsmith-review

What's inside

Skill name
langsmith-review
Description
Review LangSmith traces, runs, threads, and OpenCode or Paseo sessions through the read-only helper. Use when asked to inspect, diagnose, summarize, compare, or query LangSmith activity, latency, token usage, tool calls, or errors.
License
MIT
Metadata
summary: Safely review OpenCode and Paseo activity recorded in LangSmith. · audience: software-engineers · workflow: observability · compatibility: claude-code, codex, opencode, agent-skills
Instructions
150 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (3)

  • SKILL.md
  • review.mjs · scripts/
  • review.test.mjs · scripts/

SKILL.md (first lines)

# LangSmith Review

Use the bundled helper for read-only LangSmith reviews. Do not invent ad hoc API
calls when the helper covers the request.

## Data model

Interpret the integration as follows:

| OpenCode and Paseo | LangSmith |
| --- | --- |
| OpenCode integration | Project, normally `opencode` |
| Repository | Root-run metadata `repository_name` and `cwd` |
| Paseo agent or OpenCode session | Thread |
| One user turn | Trace |
| Task-created child session | Nested `opencode.session` run in the parent trace |
| Model response or tool call | Run |

Useful root metadata includes `repository_name`, `cwd`, `git_branch`,
`thread_id`, `turn_number`, `paseo_agent_id`, and `paseo_agent_cwd`.
Nested child roots are identified by `ls_subagent_id` and
`ls_subagent_type`. A child session shares its root session's `thread_id`; its
OpenCode session ID is not a standalone LangSmith thread ID.

## Safety

- Use read-only queries. Never create, update, delete, share, or add feedback to
  LangSmith records unless the user explicitly requests that separate action.
- Never print API keys, authorization headers, or complete process environments.
- Do not place credentials in command arguments or temporary files.
- Inputs and outputs are neither requested nor returned by default. Use
  `--include-content` only when content is needed for the requested diagnosis;
  previews remain bounded and redacted.
- Every helper response is capped at 30,000 serialized bytes, and parent-trace
  scans are capped at 1,000 runs. A larger response returns
  `outputTruncated: true` with a bounded preview; a capped scan reports
  `runsTruncated` or `truncatedParentTraceScans`.
- Treat trace content as potentially sensitive even after automatic redaction.
- A turn currently in progress may be absent or appear as pending until the
  tracer flushes. Agent debriefs exclude pending traces and report the count;
  do not diagnose that expected delay as missing telemetry.

## Helper

The helper is located at:

```text
~/.agents/skills/langsmith-review/scripts/review.mjs
```

It reads endpoint and project settings from the global and project-local
`langsmith.json` files, with the same environment-variable precedence used by
the plugin. It reads the API key from the environment without displaying it and
reuses the LangSmith SDK installed with `@langchain/langsmith-opencode`.

### Review recent traces for the current repository

```bash
node ~/.agents/skills/langsmith-review/scripts/review.mjs recent --limit 20
```

The helper derives `repository_name` from the current Git remote. If no remote
is available, it filters on the current working directory.

Use an explicit repository or review all repositories when needed:

```bash
node ~/.agents/skills/langsmith-review/scripts/review.mjs recent --repository owner/repository --limit 20
node ~/.agents/skills/langsmith-review/scripts/review.mjs recent --all-repositories --limit 20
```

### Review a Paseo agent

With no explicit ID, the helper uses `PASEO_AGENT_ID` from the current process:

```bash
node ~/.agents/skills/langsmith-review/scripts/review.mjs agent --limit 100
node ~/.agents/skills/langsmith-review/scripts/review.mjs agent <paseo-agent-id> --limit 100
```

Agent lookup first uses trace metadata, then falls back to the OpenCode session
recorded in local Paseo state. Full Paseo IDs and unambiguous ID prefixes are
accepted. The fallback queries that OpenCode session's LangSmith thread, so it
also supports traces created without `paseo_agent_id` metadata.

### Review a thread

Use the `threadId` returned by `recent`:

```bash
node ~/.agents/skills/langsmith-review/scripts/review.mjs thread <thread-id> --limit 20
```

### Review a task-created subagent

Use the OpenCode child session ID returned by `task`. The helper locates its
nested root through `ls_subagent_id` and returns only that child subtree, not
the rest of the parent trace:

```bash
node ~/.agents/skills/langsmith-review/scripts/review.mjs subagent <opencode-session-id>
node ~/.agents/skills/langsmith-review/scripts/review.mjs subagent <opencode-session-id> --include-content
```

Do not pass a task-created child session ID to `thread`; nested subagents share
the parent session's thread. Recent, agent, and thread reviews expose a
`subagents` inventory that supplies IDs for focused child review.

### Review one trace

Use the trace UUID or a LangSmith URL containing it:

```bash
node ~/.agents/skills/langsmith-review/scripts/review.mjs trace <trace-id>
node ~/.agents/skills/langsmith-review/scripts/review.mjs trace <trace-id> --include-content
```

## Review workflow

From tobywise/agent-skills, MIT.

More from tobywise/agent-skills

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Sep 23, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from tobywise

Cite this page

langsmith-review skill. Agent Mart. https://agentmart.dev/skills/tobywise-langsmith-review (last updated 2026-10-24)