T-CAWD-ci-lane-scratch-root-e2e
State: planning/proposed · Impact: medium · Complexity: medium
Run every harness-launched e2e suite on each PR that affects it. With HOME pointed at an empty
directory, CI also catches any state path that escapes fleet.
| Location | Role today |
|---|---|
solutions/determined/moon.yml | test-ui, test-ui-fitness, test-ui-polish, test-ui-backup all runInCI: false |
apps/zoo/moon.yml | e2e task, local only. Pumice, agentpants, SDF and Augmented Books have no e2e task yet |
.github/workflows/ci.yml | CI (affected), the moon graph-driven gate. Storybook screenshots run in the separate story-shots job, in a Playwright container |
.github/workflows/release-candidate.yml | Manual dispatch only. Its agent-pants-desktop-e2e job runs the agent-pants suites. No workflow runs solutions/determined/tests/ui |
Proposed
Section titled “Proposed”CI (affected) runs the e2e task of every app migrated in phase 2 (Determined suites, agent-pants,
pumice, zoo, agentpants, SDF, Augmented Books) on Linux with HOME set to an empty temp dir, scoped
by moon’s affected detection. Each task builds its own backend and web build as moon dependencies.
Slow tests, such as the backup schedule test, run nightly.
Approach
Section titled “Approach”- Give each e2e task correct
inputsanddeps(server bin, web build) so moon scopes it, and setrunInCI: true. - Add
bunx playwright install --with-deps chromiumand the empty-HOMEenv to the affected job, once, shared by every suite. - Add a nightly schedule that enables slow tests.
- Settle the PR-gating cost that
T-B2VWraises for agent-pants (about 45 minutes a run) for every suite: path-scoped by moon’s affected detection, slow cases nightly.
| Area | Note |
|---|---|
solutions/determined/moon.yml | Determined and agent-pants e2e tasks |
apps/{pumice,zoo,agentpants}/moon.yml | E2e tasks |
solutions/ontological/desktop/moon.yml, solutions/augmented/apps/augmented-books/moon.yml | E2e tasks |
.github/workflows/ci.yml | Job steps, nightly trigger |
Acceptance criteria
Section titled “Acceptance criteria”- AC-1: Every suite migrated in phase 2 runs green in
CI (affected)on a PR touching its inputs, withHOMEset to an empty temp dir. The PR body lists each suite and its run link. - AC-2: A PR touching none of an app’s inputs does not run that app’s suite.
- AC-3: The nightly run includes the slow schedule test.
Out of scope
Section titled “Out of scope”- Desktop smoke (tauri-driver).
Dependencies
Section titled “Dependencies”- The phase-2 migrations:
T-VEK2,T-GXNH,T-UU4J,T-AAR2,T-KOTJ,T-JPFC,T-TLET. T-B2VWasks the same gating question for agent-pants alone. This task answers it for every suite; close T-B2VW with it.
Discovery context
Section titled “Discovery context”- The harness needs no display, Tauri or real home, so a stock Linux runner can run it.
Depends on
Section titled “Depends on”T-VEK2, T-GXNH, T-UU4J, T-AAR2, T-KOTJ, T-JPFC, T-TLET