[Exeloo]

What does this PR do?

Add coverage and improve tests

How do you test this PR?

[Tchips46]

Review: tests and coverage (9a299b7)

Line numbers are at this PR's head. Most of the new specs are sound, and splitting the long git e2e test into independent ones is a good change. The findings below are about the CI wiring, the e2e fixtures, and some production code that ships inside this test PR.

Should fix

1. Account sync status hides errors, and saved() can return before anything was pushed (packages/settings/src/account-sync/account-sync-store.ts:112, apps/editor/src/lib/editor/follow-saved.ts:15)

• write() now sets the status to pending unless it is syncing. When the status was error (for example "too large to sync") or conflict, any edit replaces it with "Changes waiting to sync", and the error text, which only shows while the status is error, disappears.
• followSaved counts offline and error as saved. In e2e, a failed or offline push makes saved() return at once, which is the stale-account leak between tests that this helper exists to prevent.
• Fix: switch to pending only from idle. Treat offline and error as not saved, so saved() times out loudly.

2. A PR that only moves the engine or CLI pin never runs e2e (.github/workflows/e2e.yml:9-19)

• scripts/** is not in the paths list, but scripts/pinned-refs.json now picks the engine and CLI that e2e builds, and scripts/engine-bridge.mjs drives the workspace setup.
• Example: a PR that bumps only the pin is merged with no e2e run, and a bad pin first fails in the nightly run on main.
• Fix: add 'scripts/**' to paths.

3. Game cleanup failures are swallowed, and the scene spec now cleans up while the game plays (apps/editor/e2e/test.ts:35-47, apps/editor/e2e/scene.spec.ts)

• Every step of the new stopsTheGame fixture ends in .catch(() => undefined). When Stop hangs, the test passes, and the next test fails with the "server exits at once" cascade that CLAUDE.md warns about, which hides the real cause.
• scene.spec.ts dropped its own stop hook. Playwright runs afterEach hooks before fixture teardown, so the spec now restores the breakout sources, resets the layout and calls saved() while the game is still playing.
• Fix: drop the .catch on the wait for the Play button. Keep the scene spec's own stop, or stop the game first in its afterEach.

4. The keymap test asserts on its own mock (plugins/command-palette/test/keymap/keymap-rows.spec.ts:13-31)

• The mock of @nanoforge-dev/editor-sdk/ui rewrites bindingsConflict and findConflicts as a.key === b.key. The real code (packages/ui/src/workbench/keybinding/resolve-keymap.ts:89) also checks when overlap and chord prefixes (Mod+K against Mod+K S).
• The conflict assertions pass whatever the real functions return, so a regression in conflict detection ships with green tests.
• Fix: have the mock factory re-export the real pure functions from resolve-keymap.ts and keystroke.ts.

5. One turbo cache key per commit is shared by four jobs, and the cache only grows (.github/actions/prepare/action.yml, "Cache turbo")

• Lint, typecheck, unit and e2e all save under turbo-<ref_name>-<sha>, and only the first job to finish wins: the other jobs' task results are lost and rerun next time.
• Each run restores the previous .turbo/cache and saves it again with more in it. Near GitHub's 10 GB limit this evicts the engine, Playwright and pnpm caches.
• Fix: add the job name to the key, and clear .turbo/cache before saving (or drop the cross-branch restore fallback).

6. Production changes are hidden under a test: commit

• This PR also changes production code: the diff editor fix (plugins/code-editor/src/monaco/dispose-diff-editor.ts), the SettingsService.pending/hold counters with holds in LayoutController and WorkbenchService, data-nf-saved on every page, the sync status change from point 1, and about 50 removed exports.
• cliff.toml files ^test commits under "Testing", so the Monaco fix never shows as a fix in the changelog. CLAUDE.md also asks for core changes first, in their own commits.
• Fix: split out fix(code-editor), feat(settings) / feat(ui) and refactor commits.

Nice to have

• check-build (apps/editor/e2e/check-build.ts:6-13, 57) counts tooling sources as editor sources, but the command it prints (--filter=@nanoforge-dev/editor...) does not rebuild tooling/example-plugin. After an edit there, the check keeps failing or passes with a stale dev plugin. Fix: add --filter=./tooling/* to the printed command.
• Duplicates and leftover sleeps: plugins/code-editor/test/document/editor-document.spec.ts:9-27 and type-libraries.spec.ts:8-40 rebuild the fake model that plugins/code-editor/test/fixtures/fake-monaco.ts (added in this PR) already provides. inspectors.spec.ts:31-37 repeats the new stopsTheGame fixture. inspectors.spec.ts:28 and project.spec.ts:37 still sleep 900 ms where saved() now covers it.

Not checked: nothing was run locally. CI is green on this head (e2e: 102 passed, 0 skipped).

[Exeloo]

Thanks for the review. Addressed in 1d26f8e (fix: address the tests review).

Fixed

• 1. write() switches to pending only from idle, so error and conflict stay shown. followSaved counts only idle and conflict as saved, so saved() times out on error or offline. A new spec writes while in error. Note: the error text was only hidden until the next push, which sets error again; the saved() part was the real leak.
• 2. scripts/** added to the e2e paths.
• 3. A shared stopGame helper: only the visibility check may fail quietly, and the Stop click and the wait for Play throw. The scene spec stops the game first in its afterEach. The duplicate hook in inspectors.spec.ts is gone, and its 900 ms sleep is now saved().
• 5. The turbo cache key now includes the job and the month (<os>-turbo-<job>-<YYYY-MM>-<ref>-<sha>). Restores stay within the same job and month, so the cache starts over each month.
• check-build: the printed commands add --filter='./tooling/*'. It is quoted because zsh refuses the glob unquoted.
• Duplicates: editor-document.spec.ts uses the shared fakeModel.

Not changed

• 4. The real bindingsConflict/findConflicts are covered in packages/ui/test/workbench/keybinding/keybindings.spec.ts (when overlap and the Ctrl+K / Ctrl+K S prefix), so a regression there fails CI. The plugin spec tests the plugin's own wiring, and plugins only import the SDK, so mocking /ui is the convention (9 plugin specs do it).
• 6. Splitting the commit would mean rewriting a pushed branch that two other PRs of the stack are built on. These PRs are squash-merged, so the merged commit's title and changelog entry can be set at merge time.
• type-libraries.spec.ts: not a duplicate: it records the addExtraLib calls, and the shared fake does nothing with them.
• project.spec.ts:37: that sleep waits for the Project screen's file writes, which saved() doesn't track, so it stays.

Checked at the top of the stack: lint, typecheck and unit tests pass, and e2e passed 109/109 (0 skipped) against the CI-pinned engine and CLI.