seed-test-data

로컬 Langfuse 테스트 데이터를 하나의 명령어로 시드: 대규모/분기 관측 트리(v3 및 v4 이벤트), 긴 세션, 목록 성능을 위한 대량 트레이스. 사용…

npx skills add https://github.com/langfuse/langfuse --skill seed-test-data

Seed Test Data

One-shot deterministic test data for local Langfuse. The CLI handles env loading, preflight checks, ClickHouse/Postgres writes, readback verification, and prints UI deep links plus a machine-readable JSON summary (last stdout line).

If anything fails, run doctor first

pnpm run seed -- doctor

Prints PASS/WARN/FAIL per dependency (Postgres, migrations, project, ClickHouse, v4 dev tables, Redis, MinIO, web app) with the exact fix command for every failure. Do not debug Docker/ClickHouse manually before running this.

Need → command

I need...	Command
A very complex observation tree (v3)	`pnpm run seed -- trace-tree --observations 5000 --depth 12 --breadth 500`
The same tree readable in the v4 events UI	add `--v4` (writes `events_full`; `events_core` fills via MV)
Async parents whose subtree outlives their own span (subtree wall-clock duration badge)	add `--async-parents` to `trace-tree` (root + hub end immediately while children keep running)
A realistic agent flow over a timeline (graph view + scrubbable timeline)	`pnpm run seed -- agent-timeline --turns 6 --v4` (LangGraph refine loop planner→retriever→generator→critic→loop, staggered in time; add `--timing-only` for the pure timing fallback)
A trace that is large as a GRAPH (many distinct node names + connections, the trace-graph layout stress)	`pnpm run seed -- agent-graph --v4` (~1,350 distinct connections from 350 observations; `--nodes 120 --steps 100 --parallel 8` crosses the layout ceiling, `--nodes 80 --steps 30 --parallel 4` is small-but-dense)
A dozen SMALL traces, each a different timeline shape (the everyday case, not a stress test)	`pnpm run seed -- timeline-shapes --v4` (12 hand-timed traces of 4-25 observations: rag answer, streamed chat with a TTFT split, 8-way fan-out, retry backoff with widening gaps, a 13-minute wait on a human, one slow tool dwarfing everything, an error cascade with failover, in-flight spans with no end time, zero-duration checkpoints, a ten-level ladder, 24 flat siblings, a three-turn agent loop with think time; `--shape <slug>` for one)
A row carrying every annotation at once, to judge whether the timeline is overloaded	`pnpm run seed -- timeline-annotated --v4` (ONE trace, 12 observations: one/two/four scores — the last collapsing into `+1` — one and twelve comments, a streaming first-token mark, costs and durations spread so the heat map paints some rows, plain rows beside them for contrast, and all three label placements on screen together)
A multi-HOUR agent run (wall clock dominated by waiting, as production ones are)	`pnpm run seed -- agent-timeline --turns 120 --turn-gap-ms 60000 --v4` (~2.5h; the gap is jittered up to 2x, so the work lands at a few percent of the trace instead of packing every span into a few seconds)
A demo-grade, real-looking agent trace (videos, screenshots, docs)	`pnpm run seed -- support-agent --v4 --id-prefix <hex>` (one fixed, fully handcrafted support-copilot refund run: guardrails, parallel context fan-out, 3-turn ReAct loop with real payloads/costs; deterministic — reseed with a FRESH prefix for a clean take; the prefix is the trace id, so a hex prefix reads like production)
A demo-grade, real-looking multi-user SESSION (session timeline videos, screenshots, docs)	`pnpm run seed -- incident-session --id-prefix inc-4471-checkout-latency` (one fixed, handcrafted incident channel: 7 turns, 4 named users, accumulating chat history with reasoning and markdown, tool calls paired to TOOL rows plus two untraced ones, a WARNING and an ERROR tool, a nested sub-agent, a refusal, typed session scores and comments; v4 on by default; the prefix is the session id)
A plain trace with no agentic types (collapsed-by-default graph panel)	add `--plain` to `trace-tree` (SPAN/GENERATION/EVENT only)
A trace with MORE observations than the detail view loads (v4 caps the tree at 10k, `startTime ASC` — the chronological tail is missing)	`pnpm run seed -- trace-tree --observations 12000 --stride-ms 10 --v4` (`--stride-ms` starts each observation index × N ms in, so start times are unique and strictly increasing: observation index < 10000 loads, `-obs-10000` and up fall past the cap. Without it thousands of rows share one millisecond and the boundary is arbitrary. Re-runs that CHANGE timing flags need a fresh `--id-prefix`: `start_time` is an events ORDER BY key, so both versions persist in the same trace)
An extremely DEEP single-chain trace (tree depth = observation count; layout stress)	`pnpm run seed -- deep-chain --v4` (1401 sequential generations, each the sole child of the previous — the mis-parented-instrumentation shape from LFE-10959 that collapses tree/timeline layouts; `--observations N` to change depth)
A super tough session (v3 legacy session view)	`pnpm run seed -- long-session --traces 300 --observations-per-trace 8`
Diverse v4 session shapes (chat / coding-agent / mixed / media) for the session-detail view	`pnpm run seed -- session-shapes --shape all` (the `agent` shape has I/O on AGENT/TOOL with no GENERATION — pre-LFE-10520 the "first generation" default rendered empty cards for it; the current "All observations with I/O" default renders it correctly; v4 on by default)
Many sessions for the sessions TABLE and its filters / search bar (not the detail view)	`pnpm run seed -- session-variety --sessions 120 --days 14` (searchable topic ids, 1-3 userIds and 1-3 tags per session, four environments, session metadata `tier`/`region`/`channel`/`topic`, numeric + categorical + boolean session scores, comments on a quarter; v4 on by default. Widen the time-range picker past the default 1 day.)
Session messages carrying Langfuse media references (inline image, several refs in one message, bare refs in a content array, link-only payload)	`pnpm run seed -- session-shapes --shape media` — uploads the image/audio/pdf fixtures to MinIO and links them to the observation, so the inline chip and the "Media" strip both resolve (LFE-14815, LFE-9577)
Many traces for list/filter performance	`pnpm run seed -- many-traces --count 100000 --days 14`
Long-window v4 traffic with cost/latency/token OUTLIERS (outlier chart strip, LFE-14451)	`pnpm run seed -- outlier-traffic --days 90` (diurnal base load + deterministic spikes + hour-long ×8-latency incidents; root AGENT + GENERATION carrying cost + TOOL per trace; v4 on by default)
Scores with spaces in the name (filter/grammar testing)	`pnpm run seed -- scored-traces --traces 24 --v4`
Custom model definitions reachable from a trace (price editor entry points)	`pnpm run seed -- custom-models --v4` (a tiered model with a condition-gated second tier and one usage type priced at 0, a single-tier model, and a generation whose model matches no definition so its badge opens the create dialog)
Many project-owned evaluators for gallery infinite-scroll testing	`NEXTAUTH_URL=https://pr-<N>.preview.langfuse.com pnpm run seed -- evaluator-gallery --count 200` (uses the default synthetic seed API key and reconciles deterministic evaluator names; also works against local seeded environments when `NEXTAUTH_URL` is omitted)
Lots of scores on every node (dense score badges, tree-row overflow testing)	add `--scores-per-node 12` to `trace-tree` (N distinct scores per observation; try `--depth 2 --breadth 44` for many tall sibling rows)
Extra trace tags, incl. mixed case/accents (tag filter ordering)	add `--tags "Zebra,apple,Ärger"` to `trace-tree` (comma-separated, appended to the scenario's own tags)
Varied human-annotation queues (annotate UI / keyboard testing)	`pnpm run seed -- annotation-queue --core-items 12` (creates a "core types" queue covering every score-field render path + an "edge cases" queue with archived/stale/partial scores and observation/session/deleted/completed items)
Huge/malformed/unicode payloads	`pnpm run seed -- trace-tree --payload-bytes 1000000 --payload-style malformed` (styles: json, text, malformed, unicode, bignum, base64)
Big integers beyond 2^53-1 (number-precision testing)	`pnpm run seed -- trace-tree --observations 1 --payload-style bignum`
Huge base64 data-URI in ChatML IO (multimodal crash shape, LFE-10152)	`pnpm run seed -- trace-tree --observations 30 --payload-bytes 20000000 --payload-style base64 --v4` (one unbroken multi-MB base64 token in trace + root-observation IO; max 50 MB)
See all scenarios and flags	`pnpm run seed -- list --json`
Predict without writing	add `--dry-run`

For a v4 experiment with chat messages and nested JSON input/output, run pnpm run seed -- experiment-io. It creates one dataset, one experiment, and three items, then prints the experiment results link. Set NEXTAUTH_URL to your local app URL when using a port other than 3000.

Contract

Last stdout line is a JSON summary: traceIds, sessionIds, counts, verified (ClickHouse readback), links (UI deep links). Use --json to suppress progress logs. Non-zero exit = data did not land; the error includes a fix: line.
Deterministic: same --seed (default 42) and flags → same ids (ids never contain dates), with timestamps anchored to the current UTC day. Re-running within the same day overwrites in place; a later-day re-run updates the same ids with re-anchored timestamps (the previous day's rows persist under their old dates until then). Independent copies come only from --id-prefix.
Default project is the seeded 7a88fb47-b4e2-43b8-a06c-a5ce950dc53a (login demo@langfuse.com / password); override with --project.
Open the printed links in the browser to verify visually. The v4 events-backed UI is the per-user "Fast (Preview)" sidebar toggle, or LANGFUSE_MIGRATION_V4_WRITE_MODE=events_only server-side.

Extending

Add a scenario in packages/shared/scripts/seeder/scenarios/: a plain function using the deterministic Rng (never Math.random), register it in scenarios/index.ts, and update the table in packages/shared/scripts/seeder/AGENTS.md and this skill. Scenario names, flags, and JSON keys are additive-only contracts. Design rationale: packages/shared/scripts/seeder/README.md.