migrate-to-langfuse — Claude Code skill: move from LangSmith, Arize
The migrate-to-langfuse skill helps switch to Langfuse from LangSmith, Arize AX, Phoenix, Braintrust, Helicone or Promptfoo. It transfers live tracing, historical spans, datasets, prompts and evaluators using current Langfuse documentation.
MIT 186 installs ★ 344 updated 2026-10-23
npx skills add langfuse/skills --skill migrate-to-langfuse -a claude-codeWhat migrate-to-langfuse does
The skill runs a structured migration: interviews you first (source platform, destination region or self-hosted, what to migrate), builds a plan with scope and timeline, then executes after your confirmation. It starts with a test batch of ~10 traces so you can verify output in the Langfuse UI before the full run.
Uses Langfuse integration guides and each platform's export API. Runs both exporters in parallel during cutover — old one drops only after you approve. Migrates historical traces with deterministic IDs, original timestamps and scores intact; tracks progress in a state file for resumable runs.
Does not cover Langfuse-to-Langfuse scenarios (SDK upgrades, project or region moves, self-hosted to Cloud) — use the main langfuse skill for those. Identifies server-side assets outside the supported list (dashboards, saved views, annotation queues) and flags them so you plan their recreation separately.
What it helps with
- Switch to Langfuse from LangSmith, Arize or another observability platform
- Cut over live tracing without downtime
- Bring historical traces with original timestamps and scores
- Copy datasets and prompts into Langfuse
- Recreate evaluators and rerun experiments on migrated data
Example prompts for the agent
- Migrate my project from LangSmith to Langfuse Cloud with full history
- Plan a move from Promptfoo and set up parallel export
- Test migration of the first 10 traces from Phoenix so I can review them
How to install migrate-to-langfuse
1Skills CLI (Vercel)
npx skills add langfuse/skills --skill migrate-to-langfuse -a claude-code2Manually: put the files in place
~/.claude/skills/migrate-to-langfuse/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/migrate-to-langfuse/
1Skills CLI (Vercel)
npx skills add langfuse/skills --skill migrate-to-langfuse -a codex2Manually: put the files in place
.agents/skills/migrate-to-langfuse/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add langfuse/skills --skill migrate-to-langfuse -a cursor2Manually: put the files in place
.cursor/skills/migrate-to-langfuse/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install langfuse/skills skills/migrate-to-langfuse2Manually: put the files in place
.github/skills/migrate-to-langfuse/SKILL.md1Skills CLI (Vercel)
npx skills add langfuse/skills --skill migrate-to-langfuse -a gemini-cli2Platform CLI
gemini skills install https://github.com/langfuse/skills --path skills/migrate-to-langfuseManually: put the files in place
.devin/skills/migrate-to-langfuse/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add langfuse/skills --skill migrate-to-langfuse -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about migrate-to-langfuse
How do I install migrate-to-langfuse in Claude Code?
Run npx skills add langfuse/skills --skill migrate-to-langfuse -a claude-code, then open ~/.claude/skills/migrate-to-langfuse/SKILL.md.
What does the migrate-to-langfuse skill do?
It automates migration from LangSmith, Arize, Phoenix, Braintrust, Helicone or Promptfoo to Langfuse: switches live tracing, transfers historical traces with scores, copies datasets and prompts, recreates evaluators, starting with a test batch and supporting parallel export during the window before cutover.
Can I use this skill to upgrade Langfuse itself?
No — this handles moves from other platforms only. Upgrading Langfuse versions, migrating between projects or regions, or moving code-based prompts into Langfuse are covered by the main langfuse skill.
Source text
SKILL.md · MIT
--- name: migrate-to-langfuse description: >- Migrate to Langfuse from another LLM observability/evals platform (LangSmith, Arize AX, Phoenix, Braintrust, Helicone, Promptfoo, ...). A one-off vendor switch: live tracing cutover plus transfer of historical traces, datasets, prompts, and evaluators. Not for anything Langfuse-to-Langfuse — SDK or version upgrades, project/region moves, self-hosted to Cloud — and not for moving in-code prompts into Langfuse; the main langfuse skill covers those. allowed-tools: - WebFetch(domain:langfuse.com) - Bash(curl *langfuse.com/*) --- # Migrate to Langfuse You are running a one-off vendor switch to Langfuse. Interview first, plan second, execute only after confirmation. Never implement from memory — fetch the current Langfuse docs listed below at execution time. ## 1. Interview (always, before any code) Ask only what you cannot detect. Inspect the repo first: dependencies and imports usually reveal the source platform (`langsmith`, `arize`, `phoenix`, `braintrust`, `openinference`, old `langfuse` pins). 1. **Source** — which platform and version? Cloud or self-hosted? What access exists: API keys, a database connection, or only the application repo? 2. **Destination** — Langfuse Cloud (which region) or self-hosted? Verify credentials by presence only (e.g. `[ -n "$LANGFUSE_SECRET_KEY" ]`); never ask for secrets in chat, and never access `.env` contents or key values by any means — no shell commands that echo them, no reading `.env` with file tools, no grepping for values. Detect which variable names are set, nothing more. 3. **Scope** — which of: live tracing cutover, historical traces (including their scores, score configs, and custom model definitions), datasets, prompts, evaluators/LLM judges, experiment code. For history: everything or from a cutoff date? Ask what else exists in the source account (dashboards, alerts, annotation queues, saved views, ...). The repo cannot reveal server-side assets. Anything outside the supported list above is not covered by this skill: name it in the scope table as such, so the user plans to recreate it manually instead of discovering the gap after cutover. 4. **Mode** — plan only (a written runbook) or execute? If the request contradicts the repository (it names a platform that is not present, or asks to upgrade an SDK that is not installed), stop and reconcile with the user first — never reinterpret the request as a different migration and start executing it. Present the resulting plan (scope table plus order of operations) and get the user's confirmation before changing any file. However imperative the request sounds, it does not authorize removing or replacing the existing vendor's instrumentation before the plan is confirmed. ## 2. Defaults to apply - **Live cutover first.** Historical data migration is optional and billed as new ingestion in Langfuse — say so before migrating history. If the source already emits OTel/OpenInference spans, add Langfuse as a second exporter on the existing instrumentation instead of re-instrumenting; the old destination is dropped only when the parallel-run window below closes. - **Datasets and prompts copy. Evaluators and judges are recreated.** Experiment code is rewritten and re-run against the migrated dataset; never import old experiment scores. - **Test on a sample before the full run.** Migrate ~10 traces first, link the destination traces, and wait for the user to confirm they look correct before migrating everything. - **Show progress** (pages/counts) during long runs and make re-runs resumable: track sent source IDs in a state file and skip them, re-sending the same IDs creates duplicate observations (see references/reingest-history.md). - **Parallel-run window.** Keep the old exporter until the user confirms the destination data, then remove it in a separate, explicit step. This binds the implementation, not just the plan: after your edits, the old vendor's exporter/instrumentation must still be present and active. Before showing a diff, re-check that no vendor import, register call, exporter endpoint, or exporter env var was deleted or repointed. Implement parallel export as **one instrumentation with two exporters** (both span processors on the same tracer provider) — never instrument the application twice. Duplicated generations and double-counted costs in the destination are the symptom of a second capture path. ## 3. Route by scenario Fetch and follow; do not reproduce their content: | Scenario | Source of truth | | --- | --- | | Vendor with a published guide (LangSmith, Arize AX, Phoenix, Braintrust, Helicone, Promptfoo, ...) | `https://langfuse.com/resources/engineering/migrate-from-<vendor>.md` — check `https://langfuse.com/llms.txt` for the current list | | Vendor without a published guide, or historical data from any vendor | references/reingest-history.md, plus the vendor's own export API docs | Historical-trace reingestion is never improvised: read references/reingest-history.md in full before writing or running any reingestion code, even when a vendor guide exists. The vendor guide covers extraction; the reference governs reingestion — deterministic IDs, raw OTLP payloads, original timestamps, the cutover-time bound, and the warning that imported history is billed as new ingestion. If the source turns out to be another Langfuse deployment (an older version, a different project, region, or host), stop — that is not a vendor switch. Point the user to the main langfuse skill (its SDK-upgrade and v4-migration references) and the [project-to-project migration cookbook](https://langfuse.com/guides/cookbook/example_data_migration). For the live instrumentation cutover, use the integration guide matching the user's framework or SDK from https://langfuse.com/llms.txt. For deep Langfuse-side work (prompt creation, dataset APIs, experiment code), defer to the main langfuse skill if installed. ## 4. Verify and close - Sample check confirmed by the user in the destination UI before the full run. - Before asking the user to confirm a live-cutover sample, audit it yourself via the Langfuse API or CLI: one trace per request; the root span named after the agent/app (and the trace named after the root); every span type present (generations **and** tool/chain spans); each LLM call captured exactly once; token usage and cost populated. Only then send the user the trace links. - Counts match per migrated data type (source vs destination); report a summary table including failures. - Live traffic arrives in the destination; the old exporter is removed only after the parallel-run window the user chose. - The close-out summary accounts for every scope-table row from the interview, including the ones marked skipped or not covered — nothing is silently dropped.
© langfuse · MIT · langfuse/skills