test(e2e): add ExTester end-to-end test for the Deepnote notebook flow#430
Merged
Conversation
Adds a black-box ExTester (vscode-extension-tester) E2E suite that drives the full Deepnote happy path through the real VS Code UI: open a workspace folder and a one-notebook `.deepnote` file, create a Deepnote environment, select it for the notebook (kernel connects, venv + deepnote-toolkit provisioned), run the cell, and assert the rendered output contains "hello world". Beyond the test itself, this wires up reproducible setup and the fixes required to make it pass in a headless/sandboxed VS Code instance: - enable-proposed-api.js allow-lists the extension's proposed APIs in the test VS Code's product.json (the extension does not activate otherwise) - open the containing folder as a workspace (the serializer otherwise blocks on a "no workspace folder" snapshot warning) and open the notebook via Quick Open, since ExTester's `code -r` reuse-window silently no-ops in the sandbox - accept the simple folder dialog via its OK button (Enter navigates into dirs) - run via the toolbar "Run All" button and re-issue until output renders (deepnote.runallcells is gated behind context keys unset under automation) - idempotent environment creation with a stable name (reuses the venv) - exclude e2e artifacts from the VSIX in .vscodeignore Verified passing locally on headless Linux (Xvfb, VS Code 1.111.0), including a clean-state run that provisions the venv from scratch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #430 +/- ##
===========================
===========================
🚀 New features to boost your workflow:
|
Verified the extension does not need a proposed-API grant: although it declares Jupyter's `enabledApiProposals`, the core flow — activation, the notebook serializer, kernel execution, and output rendering — runs on stable VS Code APIs (output goes through `controller.createNotebookCellExecution` with a `replaceCells` fallback; the one genuinely-proposed call is guarded). On a plain stable VS Code the proposals are simply ungranted (a non-fatal log) and nothing breaks — which is how the published Marketplace / Open VSX extension runs for end users. So the E2E suite no longer patches the test VS Code's `product.json`: - remove `test/e2e/enable-proposed-api.js` - remove the `setup:e2e:proposed-api` script and drop it from `setup:e2e` - update the plan doc accordingly (§0, §6.x, §8, §9) Verified: `npm run test:e2e` passes with no Deepnote entry in the test VS Code's `product.json` allow-list, including a clean-state run that builds the venv from scratch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
Mirror the unit-test convention (compile-tsc / compile-tsc-watch then test:unittests): compile the test sources explicitly via `compile-e2e` (or `compile-e2e-watch` while iterating) instead of auto-compiling through a `pretest:e2e` npm lifecycle hook. `test:e2e` now just runs the already-compiled suite under out/e2e. Updates the plan doc to match. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
vscode-extension-tester's transitive deps tripped two CI gates: - Check Licenses: allow WTFPL (@azu/style-format) and Artistic-2.0 (binaryextensions/editions/istextorbinary/textextensions/version-range), and exclude the proprietary Microsoft-licensed @vscode/vsce-sign* binaries via --excludePackagesStartingWith. - Package Lock Drift: regenerate package-lock.json so `npm install` is a no-op; typed-rest-client's requires.qs had the resolved "6.15.2" pinned instead of its declared range "^6.9.1", which npm install kept rewriting. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
Remove specs/e2e-extester-testing-plan.md from version control while keeping the working-tree copy on disk (now untracked). .gitignore is intentionally left unchanged. The E2E suite under test/e2e/ does not depend on this doc. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
Move the reusable ExTester interaction helpers out of the single large test file into focused, reusable modules under test/e2e/helpers/: - constants timeouts + the output-iframe selector - fixtures copyFixtureToTempDir - notifications dismissAllNotifications, waitForNotification - quickInput tryOpenInputBox, clickDialogOkButton - workspace openFolderViaDialog, openWorkspaceFile - notebook clickRunAll, readRenderedOutput, runAndAwaitOutput - deepnoteEnvironment createEnvironment, selectEnvironmentForNotebook - index barrel re-export Helpers take the WebDriver from VSBrowser.instance.driver and receive the notebook name as a parameter instead of closing over suite state, so future suites can reuse them. helloWorld.e2e.test.ts shrinks from 439 to 87 lines and is now just the suite wiring. Behaviour is unchanged — verified passing. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
New .github/workflows/e2e.yml triggers via workflow_run after the CD workflow, downloads the extension VSIX that CD already built and uploaded, installs it into the test VS Code (extest install-vsix), and runs the ExTester suite against it — so the extension is built once (in CD) rather than twice. Adds a test:e2e:prebuilt script (extest run-tests) that runs the suite against an already-installed extension without repackaging; the workflow uses it after install-vsix. Locally verified: install-vsix + test:e2e:prebuilt passes against a prebuilt VSIX. workflow_run workflows fire from the default branch, so this takes effect once on main; a workflow_dispatch trigger is included for manual runs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
Replace the workflow_run-on-CD trigger with push/pull_request to main plus workflow_dispatch, matching the CI workflow. Package the extension here via `npm run package` (with the Tailwind native-module workaround and a global vsce install, same as CD) instead of resolving and downloading the VSIX artifact from the triggering CD run. This makes the E2E suite self-contained and PR-gating, dropping the CD-coupling steps and the actions:read permission they required. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014SefBqWeg8vf7jQ6vDhGU3
The single E2E test spends ~4m30s of its ~5m provisioning the Deepnote environment, which pip-installs the toolkit dependency tree (deepnote-toolkit[server], ipykernel, python-lsp-server[all], deepnote-cli) into a fresh venv — cold from PyPI on every run. Cache ~/.cache/pip so those wheels are reused across runs. The installs already use pip's cache (nothing passes --no-cache-dir). The key busts when the toolkit version / install set changes; the restore-keys prefix keeps the cache warm across unrelated changes since pip's cache is additive. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014SefBqWeg8vf7jQ6vDhGU3
ensureControllerSelectedForNotebook passed the NotebookDocument as the notebook.selectKernel `notebookEditor` argument, which can fail to bind the controller to the notebook — leaving it without an executable kernel so the first cell-execution requests are silently dropped until VS Code happens to settle. The placeholder path already resolves a real NotebookEditor via findNotebookEditor (with a comment noting it is required); apply the same here, falling back to the document. This removes a 40-106s stall observed before the first cell runs after selecting an environment (kernel now binds in ~0.5s), benefiting both real users and the E2E suite. Also harden the E2E run loop: drop RUN_ALL_REISSUE_INTERVAL 25s -> 5s (and extract OUTPUT_POLL_INTERVAL) so that if a run is ever still dropped, recovery is fast instead of up to a full 25s per miss. Measured locally (built VSIX + xvfb + ms-python), the hello-world E2E test body dropped from ~225s to ~88-107s across 5 runs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014SefBqWeg8vf7jQ6vDhGU3
createEnvironment unconditionally drove the packages and description input boxes after confirming the name. When the environment name already exists, `deepnote.environments.create` short-circuits after the name prompt with an "already exists" notification and opens no further inputs, so those `InputBox.create()` calls timed out and threw before the helper reached its `/already exists/` success check — breaking the documented idempotent retry path (Mocha `retries: 1` plus the workflow retry loop) on any rerun with a leftover `E2E Hello Env`. Guard the optional prompts behind `tryOpenInputBox` so they are driven only when the packages box actually appears; otherwise fall through to the existing notification check, which matches the sticky "already exists" toast and reuses the environment. Adds OPTIONAL_PROMPT_TIMEOUT (5s, matching the previous implicit InputBox.create default). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014SefBqWeg8vf7jQ6vDhGU3
jamesbhobbs
reviewed
Jun 26, 2026
Co-authored-by: James Hobbs <15235276+jamesbhobbs@users.noreply.github.com>
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Add shared logging helpers for best-effort E2E paths and use them in all catch handlers across the suite and helpers so transient UI failures remain visible in test output.
Remove the logCaughtError/catchAndLog helpers and log directly where errors are caught.
Contributor
|
@tkislan should we have CodeRabbit do a final pass? |
Contributor
Author
|
@coderabbitai full review |
Contributor
✅ Action performedFull review finished. |
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (1)
🚧 Files skipped from review as they are similar to previous changes (1)
📝 WalkthroughWalkthroughAdds E2E support for the Deepnote extension with new helper modules, a Deepnote fixture, a hello-world UI test, and supporting npm, TypeScript, Mocha, settings, ignore, and GitHub Actions updates. The workflow installs dependencies, builds the extension and E2E sources, provisions the VS Code test runtime, runs the suite under Xvfb, and uploads screenshots on failure. A notebook kernel selection change now passes a Estimated code review effort: 4 (Complex) | ~60 minutes Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (5 passed)
Comment |
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (4)
test/e2e/helpers/fixtures.ts (1)
17-24: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueTemp dirs are never cleaned up. Each call leaks a
deepnote-e2e-*dir underos.tmpdir(). Negligible in CI, noisy locally. Consider returning/registering a cleanup (e.g.fs.rmSync(tempDir, { recursive: true, force: true })) in the suite'safter.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@test/e2e/helpers/fixtures.ts` around lines 17 - 24, The copyFixtureToTempDir helper leaks temporary directories because it creates a deepnote-e2e-* folder without any cleanup hook. Update copyFixtureToTempDir to either return cleanup info alongside tempDir/filePath or register the created tempDir for teardown, and make sure the e2e suite’s after/afterEach removes it with fs.rmSync using recursive and force options. Keep the fix centered around copyFixtureToTempDir and the suite teardown so every temporary fixture directory is cleaned up.test/e2e/helpers/deepnoteEnvironment.ts (1)
34-34: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winName the interpreter-pick timeout.
tryOpenInputBox(5_000)uses a raw literal while line 69 usesOPTIONAL_PROMPT_TIMEOUT. Add a sibling constant (e.g.INTERPRETER_PROMPT_TIMEOUT) for consistency.As per coding guidelines: "Extract magic numbers (retry counts, delays, timeouts) as named constants near the top of the module".
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@test/e2e/helpers/deepnoteEnvironment.ts` at line 34, The interpreter-pick flow in deepnoteEnvironment still uses a raw timeout literal, which should be extracted into a named constant for consistency with the existing prompt timeout. Add a top-of-module constant alongside OPTIONAL_PROMPT_TIMEOUT (for example, INTERPRETER_PROMPT_TIMEOUT) and use it in tryOpenInputBox within the interpreter selection logic so the timeout is clearly named and easy to maintain.Source: Coding guidelines
test/e2e/helpers/notebook.ts (1)
49-49: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winExtract the
5_000frame-switch timeout as a named constant.
switchToFrame(5_000)is a raw literal while every other timeout in this suite is a named constant inconstants.ts. Hoist it (e.g.OUTPUT_FRAME_SWITCH_TIMEOUT).As per coding guidelines: "Extract magic numbers (retry counts, delays, timeouts) as named constants near the top of the module".
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@test/e2e/helpers/notebook.ts` at line 49, The notebook helper uses a raw frame-switch timeout literal in the webView.switchToFrame call, while the rest of the suite keeps timeouts in constants. Extract the 5_000 value into a named constant near the top of test/e2e/helpers/notebook.ts (for example, a dedicated frame-switch timeout constant alongside the other module constants), then update switchToFrame to use that constant.Source: Coding guidelines
test/e2e/suite/helloWorld.e2e.test.ts (1)
39-40: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winExtract the suite timeout constant.
Line 40 hardcodes another timeout value inline. Please lift it into a named module constant with the other test constants so future tuning stays centralized. As per coding guidelines, extract magic numbers (retry counts, delays, timeouts) as named constants near the top of the module.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@test/e2e/suite/helloWorld.e2e.test.ts` around lines 39 - 40, The suite timeout in helloWorld.e2e.test.ts is still hardcoded inline, so move the value used by the `this.timeout(...)` call into a named module-level constant alongside the other test constants near the top of the file. Update the `describe`/suite setup to reference that constant instead of the literal so timeout tuning stays centralized and consistent with the coding guidelines.Source: Coding guidelines
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.github/workflows/e2e.yml:
- Around line 30-31: Update the Checkout step in the e2e workflow to disable
credential persistence. In the actions/checkout usage, set persist-credentials
to false so the GitHub token is not left in local git config before the later
npm and packaging commands run.
- Around line 44-53: The fallback dependency installs in the e2e workflow are
unpinned, so they can pull newer releases than the lockfile intends. Update the
Install dependencies and Install vsce steps to use exact versions for the native
packages and `@vscode/vsce`, or move those dependencies into package.json and
invoke them via npm ci / npm exec. Keep the fixes localized around the
lightningcss, `@tailwindcss/oxide`, and vsce install commands.
In `@src/notebooks/deepnote/deepnoteKernelAutoSelector.node.ts`:
- Around line 740-747: The notebook kernel selection path is still falling back
to a NotebookDocument when findNotebookEditor() returns nothing, but
notebook.selectKernel only works with a NotebookEditor. Update
deepnoteKernelAutoSelector.node.ts in the code path around selectKernel to
return early when findNotebookEditor(notebook) yields no editor, and only pass
the notebookEditor into commands.executeCommand; align this behavior with
selectPlaceholderController() so we do not send an invalid payload.
---
Nitpick comments:
In `@test/e2e/helpers/deepnoteEnvironment.ts`:
- Line 34: The interpreter-pick flow in deepnoteEnvironment still uses a raw
timeout literal, which should be extracted into a named constant for consistency
with the existing prompt timeout. Add a top-of-module constant alongside
OPTIONAL_PROMPT_TIMEOUT (for example, INTERPRETER_PROMPT_TIMEOUT) and use it in
tryOpenInputBox within the interpreter selection logic so the timeout is clearly
named and easy to maintain.
In `@test/e2e/helpers/fixtures.ts`:
- Around line 17-24: The copyFixtureToTempDir helper leaks temporary directories
because it creates a deepnote-e2e-* folder without any cleanup hook. Update
copyFixtureToTempDir to either return cleanup info alongside tempDir/filePath or
register the created tempDir for teardown, and make sure the e2e suite’s
after/afterEach removes it with fs.rmSync using recursive and force options.
Keep the fix centered around copyFixtureToTempDir and the suite teardown so
every temporary fixture directory is cleaned up.
In `@test/e2e/helpers/notebook.ts`:
- Line 49: The notebook helper uses a raw frame-switch timeout literal in the
webView.switchToFrame call, while the rest of the suite keeps timeouts in
constants. Extract the 5_000 value into a named constant near the top of
test/e2e/helpers/notebook.ts (for example, a dedicated frame-switch timeout
constant alongside the other module constants), then update switchToFrame to use
that constant.
In `@test/e2e/suite/helloWorld.e2e.test.ts`:
- Around line 39-40: The suite timeout in helloWorld.e2e.test.ts is still
hardcoded inline, so move the value used by the `this.timeout(...)` call into a
named module-level constant alongside the other test constants near the top of
the file. Update the `describe`/suite setup to reference that constant instead
of the literal so timeout tuning stays centralized and consistent with the
coding guidelines.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: f4de13a4-70fc-412b-b8d6-ac5991e9f604
⛔ Files ignored due to path filters (1)
package-lock.jsonis excluded by!**/package-lock.json
📒 Files selected for processing (18)
.github/workflows/e2e.yml.gitignore.vscodeignorepackage.jsonsrc/notebooks/deepnote/deepnoteKernelAutoSelector.node.tstest/e2e/.mocharc.jstest/e2e/fixtures/hello-world.deepnotetest/e2e/helpers/constants.tstest/e2e/helpers/deepnoteEnvironment.tstest/e2e/helpers/fixtures.tstest/e2e/helpers/index.tstest/e2e/helpers/notebook.tstest/e2e/helpers/notifications.tstest/e2e/helpers/quickInput.tstest/e2e/helpers/workspace.tstest/e2e/settings.jsontest/e2e/suite/helloWorld.e2e.test.tstest/e2e/tsconfig.json
dinohamzic
reviewed
Jun 30, 2026
dinohamzic
left a comment
dinohamzic
left a comment
Contributor
There was a problem hiding this comment.
Some AI findings:
-
src/notebooks/deepnote/deepnoteKernelAutoSelector.node.ts: The fix still falls back to passing
NotebookDocumentintonotebook.selectKernelwhenfindNotebookEditor()misses. That is the exact path the comment says can leave the notebook without an executable kernel, and the caller still reports “Kernel ready”. I’d align this withselectPlaceholderController: warn/return or throw when no editor is found, and only callnotebook.selectKernelwith a realNotebookEditor. -
test/e2e/helpers/notebook.ts: The E2E test can mask the kernel-selection regression because it re-clicks “Run All” every 5s and ignores click failures until eventual output appears. Since the production bug is “first execution requests are silently dropped”, this can still pass if a later click succeeds. For this regression, the test should issue one run after environment selection, or have a separate stricter assertion for first-run behavior.
-
.github/workflows/e2e.yml:
actions/checkoutleaves the PR token persisted in git config, then the workflow runs PR-controlled npm/build commands. Setpersist-credentials: falsefor this pull-request workflow. -
.github/workflows/e2e.yml: The fallback
npm install ...commands and globalnpm install -g @vscode/vscebypass the lockfile, so the E2E VSIX may be built with unreviewed newer tooling/packages. Use exact versions or make them normal dev dependencies invoked throughnpm ci/npm exec.
Resolve the review comments from CodeRabbit and Dino on PR #430. Major: - ci: set persist-credentials: false on actions/checkout in e2e.yml so the GITHUB_TOKEN is not left in git config for PR-controlled build steps to read. - deps: declare @vscode/vsce as a devDependency and the lightningcss / @tailwindcss/oxide linux-x64 binaries as optionalDependencies, then drop the unpinned `npm install` fallbacks and the global vsce install from both e2e.yml and cd.yml. The native binaries are os/cpu constrained, so optionalDependencies (installed on linux, skipped elsewhere) keeps cross-platform installs working while removing the lockfile-bypassing fallback. - deepnote: in ensureControllerSelectedForNotebook, return early with a warning when findNotebookEditor() yields no editor instead of passing a NotebookDocument to notebook.selectKernel, matching selectPlaceholderController. Passing the document can leave the controller unbound and silently drop the first execution. Test guard (Dino): - e2e: replace the re-click-every-5s runAndAwaitOutput loop with a single-run runOnceAndAwaitOutput so a dropped first execution fails the test instead of being masked by a later re-run. Make clickRunAll retry the find+click atomically to survive the StaleElementReferenceError toolbar race without re-running the notebook. Drop the now-unused RUN_ALL_REISSUE_INTERVAL/OUTPUT_TIMEOUT constants and add FIRST_RUN_OUTPUT_TIMEOUT. Nitpicks: - e2e: clean up the throwaway fixture temp dir in the suite's after() hook. - e2e: extract INTERPRETER_PROMPT_TIMEOUT, OUTPUT_FRAME_SWITCH_TIMEOUT and SUITE_TIMEOUT named constants (consistent _000 formatting). Verified: full project + e2e type-check, formatter, and the ExTester e2e suite running under Xvfb (1 passing). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019VuoSZXF8YHsN81Lra45JW
dinohamzic
approved these changes
Jul 1, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a black-box ExTester (
vscode-extension-tester) E2E test that drives the full Deepnote happy path through the real VS Code UI: open a workspace folder + a one-notebook.deepnotefile → create a Deepnote environment → select it for the notebook (kernel connects, venv +deepnote-toolkitprovisioned) → Run All → assert the rendered output containshello world.What's included
test/e2e/— the suite (helloWorld.e2e.test.ts), config (tsconfig.json,.mocharc.js,settings.json), and the hello-world fixturepackage.json—vscode-extension-testerdevDep +compile-e2e/setup:e2e*/test:e2escripts.gitignore/.vscodeignore— ignore ExTester artifacts and keep them out of the packaged VSIXspecs/e2e-extester-testing-plan.md— self-contained plan/reference, with a "§0 Implementation reality" section documenting the fixes belowKey fixes needed to pass headlessly
code -rreuse-window silently no-ops in the sandboxdeepnote.runallcellsis gated behind context keys that are unset under automation)No proposed-API grant needed
The extension declares Jupyter's
enabledApiProposals, but the core flow (activation, the notebook serializer, kernel execution, output rendering) runs on stable VS Code APIs, so the suite runs on a plain stable VS Code with noproduct.jsonallow-listing — the way the published Marketplace / Open VSX extension runs for end users. (An earlier revision patchedproduct.json; that was verified unnecessary and removed.)How to run
Verification
Verified passing locally on headless Linux (Xvfb, VS Code 1.111.0), with no proposed-API grant in
product.json, including a clean-state run that provisions the venv from scratch:🤖 Generated with Claude Code
https://claude.ai/code/session_01LQk7n13UfmeDo2ujy8X4LX
Summary by CodeRabbit