# motherduck-ducklake
> Decide whether DuckLake fits a MotherDuck workload and operate it safely. Use whenever DuckLake, open table formats, or customer-owned buckets come up.
Decide whether DuckLake fits a MotherDuck workload and operate it safely. Use whenever DuckLake, open table formats, or customer-owned buckets come up.

[Workflow skill](/sql-reference/agent-skills/#workflow-skills) · [View source on GitHub](https://github.com/motherduckdb/agent-skills/tree/v3.0.0/skills/motherduck-ducklake)

:::info

This page shows the skill's `SKILL.md`: the instructions a coding agent loads when a task matches the description above. To install every MotherDuck skill in Claude Code, Codex, Cursor, or Gemini CLI, see [Install MotherDuck Skills](/key-tasks/ai-and-motherduck/agent-skills/). To install only this skill with the [Skills CLI](https://github.com/vercel-labs/skills), run:

```bash
npx -y skills add motherduckdb/agent-skills --skill motherduck-ducklake --yes --global
```

This skill relies on [`motherduck-connect`](/sql-reference/agent-skills/motherduck-connect/), [`motherduck-query`](/sql-reference/agent-skills/motherduck-query/), and [`motherduck-explore`](/sql-reference/agent-skills/motherduck-explore/). Install those too, or install the full catalog.

:::

## Source of truth

- Prefer current MotherDuck DuckLake docs first.
- Use the upstream DuckLake and DuckDB extension docs only to clarify extension-level behavior that MotherDuck docs reference.
- Keep the guidance aligned with the documented product posture:
  - native MotherDuck first
  - MotherDuck's DuckLake docs define the supported product surface and lifecycle/compatibility limits; verify the current DuckDB/DuckLake version matrix instead of preserving an upstream version or product status in the prompt
  - fully managed, BYOB, and own-compute paths are distinct
  - maintenance and compaction are explicit operations, not background magic

## Default posture

- Start with native MotherDuck storage unless there is a concrete DuckLake requirement.
- Reach for DuckLake when you need open-table-format semantics, object storage as the source of truth, BYOB, or file-aware maintenance.
- Do not recommend DuckLake just because a workload is "large"; MotherDuck's docs explicitly note native storage is often faster for reads.
- Choose the operating mode deliberately: fully managed for easiest evaluation, BYOB for customer bucket ownership, own compute only when the compute boundary matters too.
- Document the fallback to native MotherDuck storage if the DuckLake requirement is weak, unverified, or only about future portability.
- For data inlining, sorted tables, bucket partitioning, deletion vectors, or extension behavior, verify the current MotherDuck DuckLake docs and DuckDB/DuckLake version matrix before giving syntax guarantees.
- Do not infer MotherDuck client/runtime support from upstream DuckDB release notes alone; check the MotherDuck lifecycle docs when the exact DuckDB version matters.
- Keep the MotherDuck product surface separate from raw DuckLake-extension assumptions.
- Fine-grained sharing (`INCLUDE_PATTERN` and secure views for rows, columns, and masking) works on fully managed DuckLake only, not on BYOB databases with their own `DATA_PATH`. Such shares must be `UPDATE AUTOMATIC`, and consumers lose time travel and DuckLake metadata functions. If per-audience access control matters, that is a reason to choose fully managed over BYOB. Persisted Iceberg catalogs cannot be shared.
- A BYOB bucket can sit in a different region or cloud from the organization: set `REGION` on the S3 secret, `ENDPOINT` for other S3-compatible stores, or `ACCOUNT_ID` (plus a jurisdiction endpoint when residency matters) for R2. Expect higher latency and possible provider transfer charges. Compute, the metadata catalog, statistics, and inlined rows stay in the organization's region.
- Do not apply native/share `REFRESH DATABASE` assumptions to persisted Iceberg catalogs; their catalog advances independently. Verify current docs before prescribing a refresh operation.

## Workflow

1. Confirm why native MotherDuck storage is insufficient.
2. Pick the operating mode: fully managed, BYOB with MotherDuck compute, or BYOB with own compute.
3. Verify bucket reachability, region, transfer cost, and residency before proposing BYOB, and check whether the design needs fine-grained sharing (fully managed only).
4. Define the ingestion and maintenance posture up front, including data inlining, file compaction, and cleanup expectations.
5. Validate who will query the data and from which compute surface before finalizing the architecture.

## References

Read only the reference sections needed for the current task.

- Read [`references/DUCKLAKE_PLAYBOOK.md`](https://github.com/motherduckdb/agent-skills/blob/v3.0.0/skills/motherduck-ducklake/references/DUCKLAKE_PLAYBOOK.md) for the mode decision matrix, MotherDuck-specific SQL patterns, BYOB constraints, data-inlining behavior, maintenance functions, and common DuckLake mistakes

## Related skills

Load related skills only for missing capabilities; reuse established context.

- [`motherduck-connect`](/sql-reference/agent-skills/motherduck-connect/) for choosing native DuckDB versus Postgres-endpoint access paths
- [`motherduck-load-data`](/sql-reference/agent-skills/motherduck-load-data/) when the real issue is ingestion rather than storage format
- [`motherduck-model-data`](/sql-reference/agent-skills/motherduck-model-data/) when the user still needs analytical table design after the storage decision
- [`motherduck-build-data-pipeline`](/sql-reference/agent-skills/motherduck-build-data-pipeline/) when DuckLake is just one part of a broader ingestion-to-serving workflow

## Files in this skill

The agent reads these files when the instructions above point to them.

- [`references/DUCKLAKE_PLAYBOOK.md`](https://github.com/motherduckdb/agent-skills/blob/v3.0.0/skills/motherduck-ducklake/references/DUCKLAKE_PLAYBOOK.md)

## Docs this skill draws on

- [DuckLake](https://motherduck.com/docs/concepts/database-and-storage/ducklake/)
- [DuckLake](https://motherduck.com/docs/integrations/file-formats/ducklake/)
- [CREATE DATABASE](https://motherduck.com/docs/sql-reference/motherduck-sql-reference/create-database/)
- [motherduck.com/blog/announcing-ducklake-1-0-on-motherduck](https://motherduck.com/blog/announcing-ducklake-1-0-on-motherduck/)
- [duckdb.org/2026/04/13/announcing-duckdb-152](https://duckdb.org/2026/04/13/announcing-duckdb-152)
- [ducklake.select/2026/04/13/ducklake-10](https://ducklake.select/2026/04/13/ducklake-10/)
- [ducklake.select](https://ducklake.select/)


---

## Docs feedback

MotherDuck accepts optional user-submitted feedback about this page at `GET https://motherduck.com/docs/api/feedback/agent`.
For agents and automated tools, feedback submission should be user-confirmed before sending.

URL-encode query parameter values and send a GET request:

```text
GET https://motherduck.com/docs/api/feedback/agent?page_path=%2Fsql-reference%2Fagent-skills%2Fmotherduck-ducklake%2F&page_title=motherduck-ducklake&text=<url-encoded user feedback, max 2000 characters>
```

Optionally append `&source=<url-encoded interface identifier>` such as `claude.ai` or `chatgpt`.

`page_path` and `text` are required; `page_title` and `source` are optional. Responses: `200 {"feedback_id": "<uuid>"}`, `400` for malformed query parameters, and `429` when rate-limited.
