chore(ai): bump AI workbench to e82e1029 and regenerate scaffolds

CHANGELOG.md

@@ -16,6 +16,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - The onboarding notebook is a single page; its "Next step" button navigates to the organization's setup page on dltHub.
 - The dataset viewer now hides the notebook header, and opens on whichever stack (local or hosted) you're connected to instead of always pointing to production.
 - Refreshed the bundled workspace `uv.lock` (`dlthub-client` 0.28.1, `marimo` 0.23.13; `pandas` 3.0.3 — 3.0.4 was yanked).
+- Refreshed the bundled AI workbench: added a `performance` toolkit for speeding up slow pipelines, and the onboarding agent now explains that sample data loads into a hosted playground destination (not for production).
 
 ## [0.10.1] - 2026-07-01
 

src/create_dlthub_workspace/config.py

@@ -60,7 +60,7 @@
 # To bump: pick a new SHA (the workbench repo has no tags today), update the
 # constant below, run `make generate-ai`, commit the resulting scaffold diff
 # alongside this change.
-WORKBENCH_REF: str | None = "f290a9b3eab9db7ebb7b6e1e56d75856674488e7"
+WORKBENCH_REF: str | None = "e82e1029ca56512631d4e7133f5d3b7186f54965"
 
 
 @dataclass(frozen=True)

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/claude/.claude/rules/init-dlthub-workspace.md

@@ -61,8 +61,9 @@ deploy / schedule pipelines on the dltHub platform      → dlthub-platform
 guided end-to-end tour, ingest to dashboard (uses the real toolkits) → quick-start | dlthub --non-interactive ai toolkit install quick-start          | quick-start
 test/try dlthub end-to-end — minimal pipeline + educational test deploy, NOT production → one-shot       | dlthub --non-interactive ai toolkit install one-shot               | deploy-run-sample-pipeline
 build and deploy a minimal custom REST API pipeline after uvx dlthub-init setup → dlthub-init-skills | dlthub --non-interactive ai toolkit install dlthub-init-skills     | deploy-minimal-ingestion-pipeline
+optimize / speed up a slow or memory-heavy pipeline — parallelism, workers, batching → performance | dlthub --non-interactive ai toolkit install performance            | optimize-performance
 ```
 * `one-shot` vs `rest-api-pipeline`: one-shot is for **testing / trying dlthub / onboarding / a quick demo** — a minimal single-endpoint, row-limited pipeline on local DuckDB plus an educational test deploy. Educational examples only, NOT production-grade. For a **real or production** REST pipeline (auth, incremental, multiple endpoints, production deploy), use `rest-api-pipeline`. `quick-start` is the guided tour that walks the real toolkits end-to-end.
-* After installing, run `uv run dlthub ai status` to confirm, then continue **in the same session** — load the new toolkit's entry skill + workflow rule via `toolkit_info` (or read the installed files) and proceed. No restart needed (toolkits reuse the already-running `dlt-workspace-mcp`); don't lose the user's context.
+* After installing, confirm success from the install output (run `uv run dlthub ai status` only if the output is unclear or the MCP server hasn't been verified this session), then continue **in the same session** — load the new toolkit's entry skill + workflow rule via `toolkit_info` (or read the installed files) and proceed. No restart needed (toolkits reuse the already-running `dlt-workspace-mcp`); don't lose the user's context.
 * The `dlthub-router` skill wraps this flow and is the fallback for needs not covered above (it uses live `list_toolkits` to discover newer toolkits).
 * DO NOT start data engineering work if no workflow toolkit is installed.

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/claude/.claude/skills/deploy-run-sample-pipeline/SKILL.md

@@ -44,7 +44,7 @@ Summarize which jobs were created or updated.
 
 **Run:**
 
-Print to the user: `Running your pipeline and preparing your data in the background.`
+Print to the user: `Running your pipeline and preparing your data in the background. Your data is loading into a dltHub-hosted playground destination — zero setup and no credentials needed, but it's limited to playground/onboarding use, not production.`
 
 Run both commands at the same time — start `serve_headless.py` in the background, then run the pipeline in the foreground:
 
@@ -97,7 +97,7 @@ Print to the user: `- [x] Opening dltHub dataset browser`
 
 After Step 4 completes, immediately print to the user:
 
-> "Onboarding complete! When you're ready to continue, ask me: `Help me get started building and running a data pipeline on dltHub`"
+> "Onboarding complete! Your data lives in the dltHub-hosted playground destination — when you build your own pipeline, you'll choose a real destination of your own. When you're ready to continue, ask me: `Help me get started building and running a data pipeline on dltHub`"
 
 ## When the user says "Help me get started building and running a data pipeline on dltHub"
 

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/claude/.claude/skills/dlthub-router/SKILL.md

@@ -27,13 +27,16 @@ Use this **only** when the index has no matching row (an unfamiliar need, or you
 
 Match intent to the best toolkit, then install as in Step 1. Toolkits marked `(installed: <version>)` are already available.
 
-## Step 3: Confirm & enable MCP
+## Step 3: Verify install (only when needed)
 
-```
-uv run dlthub ai status
-```
-1. You should see the new toolkit and its entry skill.
-2. If you see any **WARNING** about the MCP server (e.g. cannot be started), **fix it** using the error message.
+**Skip this step** when the install output already confirms success and the new toolkit's entry skill is available in this session — that is all the confirmation you need. (MCP health was already checked at session start via `dlthub ai status`.)
+
+Run `uv run dlthub ai status` only if:
+- the install output was ambiguous or reported an error,
+- the entry skill doesn't appear to be available, or
+- the `dlt-workspace-mcp` server hasn't been verified this session (no session-start status check and no successful MCP call yet).
+
+If status shows a **WARNING** about the MCP server (e.g. cannot be started), **fix it** using the error message before handing over.
 
 ## Step 4: Handover (no restart needed)
 

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/claude/.dlt/.toolkits

@@ -11,7 +11,7 @@ init:
   agent: claude
   files:
     .claude/skills/dlthub-router/SKILL.md:
-      sha3_256: b0e91d5f6926a091cfb1cbfcfedf653f982d95ff1c14585b558f1b2e7ae3fc85
+      sha3_256: af56355d874845d1e58f9e5dcefbc929b9cdd2ce92162cf16814a66016c58396
     .claude/skills/improve-skills/SKILL.md:
       sha3_256: c52b222d82d101eaacbbf19d271d3f24b20d8e2bc70d49919d56144d448cd5c4
     .claude/skills/setup-secrets/SKILL.md:
@@ -21,7 +21,7 @@ init:
     .claudeignore:
       sha3_256: 1aa4dc9286dd219dcb55399f388c5ee1f940ab5e399a42b644d488cba57a9ec0
     .claude/rules/init-dlthub-workspace.md:
-      sha3_256: 75f470e9587a3d29589d655a135cc882a340b362478e0b11b1d67fa5a76331cd
+      sha3_256: bd0d85b8bf6b0e0ce454982cfa1f63baddc67cbd1f9d05254a4b6185b48643c0
   mcp_servers:
   - dlt-workspace-mcp
   name: init
@@ -41,6 +41,6 @@ one-shot:
   agent: claude
   files:
     .claude/skills/deploy-run-sample-pipeline/SKILL.md:
-      sha3_256: 0d318282b0c0e848e43c99978f4e95be29154b6f19a06883186f9518cae5f1d3
+      sha3_256: 91eb1893d0eff52a11af80459d54663428254a9ad17bda38940154d2600a056e
     .claude/rules/one-shot-workflow.md:
       sha3_256: 35dd0e85bc37f1702998d362bddc596c4a6b1d16d2350ec990d37c61af228f5a

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/codex/.agents/skills/deploy-run-sample-pipeline/SKILL.md

@@ -44,7 +44,7 @@ Summarize which jobs were created or updated.
 
 **Run:**
 
-Print to the user: `Running your pipeline and preparing your data in the background.`
+Print to the user: `Running your pipeline and preparing your data in the background. Your data is loading into a dltHub-hosted playground destination — zero setup and no credentials needed, but it's limited to playground/onboarding use, not production.`
 
 Run both commands at the same time — start `serve_headless.py` in the background, then run the pipeline in the foreground:
 
@@ -97,7 +97,7 @@ Print to the user: `- [x] Opening dltHub dataset browser`
 
 After Step 4 completes, immediately print to the user:
 
-> "Onboarding complete! When you're ready to continue, ask me: `Help me get started building and running a data pipeline on dltHub`"
+> "Onboarding complete! Your data lives in the dltHub-hosted playground destination — when you build your own pipeline, you'll choose a real destination of your own. When you're ready to continue, ask me: `Help me get started building and running a data pipeline on dltHub`"
 
 ## When the user says "Help me get started building and running a data pipeline on dltHub"
 

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/codex/.agents/skills/dlthub-router/SKILL.md

@@ -39,13 +39,16 @@ Use this **only** when the index has no matching row (an unfamiliar need, or you
 
 Match intent to the best toolkit, then install as in Step 1. Toolkits marked `(installed: <version>)` are already available.
 
-## Step 3: Confirm & enable MCP
+## Step 3: Verify install (only when needed)
 
-```
-uv run dlthub ai status
-```
-1. You should see the new toolkit and its entry skill.
-2. If you see any **WARNING** about the MCP server (e.g. cannot be started), **fix it** using the error message.
+**Skip this step** when the install output already confirms success and the new toolkit's entry skill is available in this session — that is all the confirmation you need. (MCP health was already checked at session start via `dlthub ai status`.)
+
+Run `uv run dlthub ai status` only if:
+- the install output was ambiguous or reported an error,
+- the entry skill doesn't appear to be available, or
+- the `dlt-workspace-mcp` server hasn't been verified this session (no session-start status check and no successful MCP call yet).
+
+If status shows a **WARNING** about the MCP server (e.g. cannot be started), **fix it** using the error message before handing over.
 
 ## Step 4: Handover (no restart needed)
 

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/codex/.agents/skills/init-dlthub-workspace/SKILL.md

@@ -65,8 +65,9 @@ deploy / schedule pipelines on the dltHub platform      → dlthub-platform
 guided end-to-end tour, ingest to dashboard (uses the real toolkits) → quick-start | dlthub --non-interactive ai toolkit install quick-start          | quick-start
 test/try dlthub end-to-end — minimal pipeline + educational test deploy, NOT production → one-shot       | dlthub --non-interactive ai toolkit install one-shot               | deploy-run-sample-pipeline
 build and deploy a minimal custom REST API pipeline after uvx dlthub-init setup → dlthub-init-skills | dlthub --non-interactive ai toolkit install dlthub-init-skills     | deploy-minimal-ingestion-pipeline
+optimize / speed up a slow or memory-heavy pipeline — parallelism, workers, batching → performance | dlthub --non-interactive ai toolkit install performance            | optimize-performance
 ```
 * `one-shot` vs `rest-api-pipeline`: one-shot is for **testing / trying dlthub / onboarding / a quick demo** — a minimal single-endpoint, row-limited pipeline on local DuckDB plus an educational test deploy. Educational examples only, NOT production-grade. For a **real or production** REST pipeline (auth, incremental, multiple endpoints, production deploy), use `rest-api-pipeline`. `quick-start` is the guided tour that walks the real toolkits end-to-end.
-* After installing, run `uv run dlthub ai status` to confirm, then continue **in the same session** — load the new toolkit's entry skill + workflow rule via `toolkit_info` (or read the installed files) and proceed. No restart needed (toolkits reuse the already-running `dlt-workspace-mcp`); don't lose the user's context.
+* After installing, confirm success from the install output (run `uv run dlthub ai status` only if the output is unclear or the MCP server hasn't been verified this session), then continue **in the same session** — load the new toolkit's entry skill + workflow rule via `toolkit_info` (or read the installed files) and proceed. No restart needed (toolkits reuse the already-running `dlt-workspace-mcp`); don't lose the user's context.
 * The `dlthub-router` skill wraps this flow and is the fallback for needs not covered above (it uses live `list_toolkits` to discover newer toolkits).
 * DO NOT start data engineering work if no workflow toolkit is installed.

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/codex/.dlt/.toolkits

@@ -11,7 +11,7 @@ init:
   agent: codex
   files:
     .agents/skills/dlthub-router/SKILL.md:
-      sha3_256: 5eab0df88b50966b41dfe1fca3e0e3f92c496627a59a0414beb6409bab0b2ac9
+      sha3_256: 5d383df737cd33c30a5418af43f5b470d5bcc51aa25b9e346b8f4c3698a6e27f
     .agents/skills/improve-skills/SKILL.md:
       sha3_256: c52b222d82d101eaacbbf19d271d3f24b20d8e2bc70d49919d56144d448cd5c4
     .agents/skills/setup-secrets/SKILL.md:
@@ -21,7 +21,7 @@ init:
     .codexignore:
       sha3_256: 1aa4dc9286dd219dcb55399f388c5ee1f940ab5e399a42b644d488cba57a9ec0
     .agents/skills/init-dlthub-workspace/SKILL.md:
-      sha3_256: 049543af4bf8621ef57dc708aa600a76e6181773f995b60d916a23ef0531c445
+      sha3_256: 0641e3c3fc78e924050bf6dda703fc2072d1f52f73c500f66bdc964b0df96a3c
   mcp_servers:
   - dlt-workspace-mcp
   name: init
@@ -41,6 +41,6 @@ one-shot:
   agent: codex
   files:
     .agents/skills/deploy-run-sample-pipeline/SKILL.md:
-      sha3_256: 0d318282b0c0e848e43c99978f4e95be29154b6f19a06883186f9518cae5f1d3
+      sha3_256: 91eb1893d0eff52a11af80459d54663428254a9ad17bda38940154d2600a056e
     .agents/skills/one-shot-workflow/SKILL.md:
       sha3_256: 0f985a1242bd241020f1ff4b8e56ccc157004314f164e483186b238bd0860a9d

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/codex/AGENTS.md

@@ -8,7 +8,7 @@ they are essential for ANY work in this project
 CRITICAL: never ask for credentials in chat. Always let the user edit secrets directly and do not attempt to read them.
 
 ## toolkits — match intent → install → open the entry skill (no discovery round-trip needed)
-Workflow toolkits are installed on demand. This index is authoritative for shipped toolkits: match the user's intent, run the install command, confirm with `dlthub ai status`, then hand over to the entry skill. No discovery call needed for these.
+Workflow toolkits are installed on demand. This index is authoritative for shipped toolkits: match the user's intent, run the install command, confirm from its output (`dlthub ai status` only if unclear), then hand over to the entry skill. No discovery call needed for these.
 <!-- This shipped index can drift from the live catalog on a user's machine until runtime refresh lands; tracked in dlt-hub/dlthub-ai-workbench-internal#71. -->
 
 ```
@@ -23,6 +23,7 @@ deploy / schedule pipelines on the dltHub platform      → dlthub-platform
 guided end-to-end tour, ingest to dashboard (uses the real toolkits) → quick-start | dlthub --non-interactive ai toolkit install quick-start          | quick-start
 test/try dlthub end-to-end — minimal pipeline + educational test deploy, NOT production → one-shot       | dlthub --non-interactive ai toolkit install one-shot               | deploy-run-sample-pipeline
 build and deploy a minimal custom REST API pipeline after uvx dlthub-init setup → dlthub-init-skills | dlthub --non-interactive ai toolkit install dlthub-init-skills     | deploy-minimal-ingestion-pipeline
+optimize / speed up a slow or memory-heavy pipeline — parallelism, workers, batching → performance | dlthub --non-interactive ai toolkit install performance            | optimize-performance
 ```
 * `one-shot` vs `rest-api-pipeline`: one-shot is for **testing / trying dlthub / onboarding / a quick demo** — a minimal single-endpoint, row-limited pipeline on local DuckDB plus an educational test deploy. Educational examples only, NOT production-grade. For a **real or production** REST pipeline (auth, incremental, multiple endpoints, production deploy), use `rest-api-pipeline`. `quick-start` is the guided tour that walks the real toolkits end-to-end.
 * Use the `dlthub-router` skill for needs not covered above — it uses live `list_toolkits` to discover newer toolkits.

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/cursor/.cursor/rules/init-dlthub-workspace.mdc

@@ -65,8 +65,9 @@ deploy / schedule pipelines on the dltHub platform      → dlthub-platform
 guided end-to-end tour, ingest to dashboard (uses the real toolkits) → quick-start | dlthub --non-interactive ai toolkit install quick-start          | quick-start
 test/try dlthub end-to-end — minimal pipeline + educational test deploy, NOT production → one-shot       | dlthub --non-interactive ai toolkit install one-shot               | deploy-run-sample-pipeline
 build and deploy a minimal custom REST API pipeline after uvx dlthub-init setup → dlthub-init-skills | dlthub --non-interactive ai toolkit install dlthub-init-skills     | deploy-minimal-ingestion-pipeline
+optimize / speed up a slow or memory-heavy pipeline — parallelism, workers, batching → performance | dlthub --non-interactive ai toolkit install performance            | optimize-performance
 ```
 * `one-shot` vs `rest-api-pipeline`: one-shot is for **testing / trying dlthub / onboarding / a quick demo** — a minimal single-endpoint, row-limited pipeline on local DuckDB plus an educational test deploy. Educational examples only, NOT production-grade. For a **real or production** REST pipeline (auth, incremental, multiple endpoints, production deploy), use `rest-api-pipeline`. `quick-start` is the guided tour that walks the real toolkits end-to-end.
-* After installing, run `uv run dlthub ai status` to confirm, then continue **in the same session** — load the new toolkit's entry skill + workflow rule via `toolkit_info` (or read the installed files) and proceed. No restart needed (toolkits reuse the already-running `dlt-workspace-mcp`); don't lose the user's context.
+* After installing, confirm success from the install output (run `uv run dlthub ai status` only if the output is unclear or the MCP server hasn't been verified this session), then continue **in the same session** — load the new toolkit's entry skill + workflow rule via `toolkit_info` (or read the installed files) and proceed. No restart needed (toolkits reuse the already-running `dlt-workspace-mcp`); don't lose the user's context.
 * The `dlthub-router` skill wraps this flow and is the fallback for needs not covered above (it uses live `list_toolkits` to discover newer toolkits).
 * DO NOT start data engineering work if no workflow toolkit is installed.

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/cursor/.cursor/skills/deploy-run-sample-pipeline/SKILL.md

@@ -44,7 +44,7 @@ Summarize which jobs were created or updated.
 
 **Run:**
 
-Print to the user: `Running your pipeline and preparing your data in the background.`
+Print to the user: `Running your pipeline and preparing your data in the background. Your data is loading into a dltHub-hosted playground destination — zero setup and no credentials needed, but it's limited to playground/onboarding use, not production.`
 
 Run both commands at the same time — start `serve_headless.py` in the background, then run the pipeline in the foreground:
 
@@ -97,7 +97,7 @@ Print to the user: `- [x] Opening dltHub dataset browser`
 
 After Step 4 completes, immediately print to the user:
 
-> "Onboarding complete! When you're ready to continue, ask me: `Help me get started building and running a data pipeline on dltHub`"
+> "Onboarding complete! Your data lives in the dltHub-hosted playground destination — when you build your own pipeline, you'll choose a real destination of your own. When you're ready to continue, ask me: `Help me get started building and running a data pipeline on dltHub`"
 
 ## When the user says "Help me get started building and running a data pipeline on dltHub"
 

src/create_dlthub_workspace/scaffolds/minimal_workspace/_agents/cursor/.cursor/skills/dlthub-router/SKILL.md

@@ -27,13 +27,16 @@ Use this **only** when the index has no matching row (an unfamiliar need, or you
 
 Match intent to the best toolkit, then install as in Step 1. Toolkits marked `(installed: <version>)` are already available.
 
-## Step 3: Confirm & enable MCP
+## Step 3: Verify install (only when needed)
 
-```
-uv run dlthub ai status
-```
-1. You should see the new toolkit and its entry skill.
-2. If you see any **WARNING** about the MCP server (e.g. cannot be started), **fix it** using the error message.
+**Skip this step** when the install output already confirms success and the new toolkit's entry skill is available in this session — that is all the confirmation you need. (MCP health was already checked at session start via `dlthub ai status`.)
+
+Run `uv run dlthub ai status` only if:
+- the install output was ambiguous or reported an error,
+- the entry skill doesn't appear to be available, or
+- the `dlt-workspace-mcp` server hasn't been verified this session (no session-start status check and no successful MCP call yet).
+
+If status shows a **WARNING** about the MCP server (e.g. cannot be started), **fix it** using the error message before handing over.
 
 ## Step 4: Handover (no restart needed)