fix(chromium): restore automation launch flag

[Devin Rousso (dcrousso)]

removing --enable-automation lets managed Edge hand off the browser process during SAML login

restore the default flag while preserving MCP launch behavior through an internal launch setting

regression from #40190

fixes#42379

[Yury Semikhatsky (yury-s)]

You can use ignoreDefaultArgs option instead.

[Yury Semikhatsky (yury-s)]

It was removed in this change after some back and forth changes. The idea was that when browser is controlled via remote debugger, the same features are already turned on. --enable-automation is a big hammer and we haven't heard any complaints after removing the flag. Let's narrow down what is not working in their scenario and see if we can fix it differently.

[Devin Rousso (dcrousso)]

unless im misreading the code, that change always set --enable-automation unless assistantMode: true, which was only for MCP, meaning that --enable-automation was always on for all other cases

furthermore, it was unconditionally added and has been unconditionally set since b20e87d 7yrs ago

[Devin Rousso (dcrousso)]

as the original issue demonstrates, Chromium also uses --enable-automation to disable process handoff behavior https://issues.chromium.org/issues/371112535

[Pavel Feldman (pavelfeldman)]

This likely has implications we don't want to deal with.

[github-actions]

Test results for "tests 1"

3 flaky

⚠️ [chromium-library] › library/video.spec.ts:736 › screencast › should work with video+trace `@chromium-ubuntu-22.04-node22`
⚠️ [firefox-page] › page/page-emulate-media.spec.ts:144 › should keep reduced motion and color emulation after reload `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-page] › page/page-event-request.spec.ts:181 › should return response body when Cross-Origin-Opener-Policy is set `@firefox-ubuntu-22.04-node20`

51299 passed, 1237 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

2 failed
❌ [chrome] › mcp/annotate.spec.ts:137 › should abort MCP annotation when last screenshot is removed @mcp-windows-latest-chrome
❌ [firefox] › mcp/cli-devtools.spec.ts:173 › response-body returns just the body @mcp-windows-latest-firefox

8257 passed, 1361 skipped

---

Merge workflow run.

[github-actions]

Hi, I'm the Playwright bot and I took a look at the failing CI on this PR.

🟢 The two failures are pre-existing flakes — this PR is clear

Both are Windows-only MCP flakes that fail on unrelated SHAs, and neither sits on a code path this PR changes. The diff adds --enable-automation back to the Chromium switches and makes the MCP launcher ignore it — for MCP-launched Chrome the net set of flags is unchanged (flag added to defaults, then ignored), and for Firefox the flag is a no-op. So nothing here touches annotation UI or devtools response-body capture.

Details

Latest MCP run had 2 failed; latest "tests 1" run was flaky-only. Both failures reproduce on other PRs in the test-results DB.

Pre-existing flake / infra

• [chrome] › mcp/annotate.spec.ts:137 › should abort MCP annotation when last screenshot is removed@mcp-windows-latest-chrome — fails as a 15s UI timeout (expect(locator('.annotate-sidebar-thumb')).toHaveCount(1), received undefined). Across the DB this test is green 741/743 on chrome and also flaked once on unrelated fix(role): map input[type=search] to searchbox unless list is a datalist #41913 (plus 3 times on msedge); it passes on chromium, firefox, and webkit everywhere. The MCP launch-flag change is a no-op for Chrome's effective args, so it can't explain a sidebar-thumbnail timeout.

• [firefox] › mcp/cli-devtools.spec.ts:173 › response-body returns just the body@mcp-windows-latest-firefox — fails as expect(received).not.toBeNull() (body came back null). This one is a well-worn windows-latest-firefox flake: 6 prior failures on unrelated runs — fix(mcp): launch the Chrome profile that has the extension installed #41939, feat(chromium): roll to r1235 #41974, fix(cli): rewrite CLI command in installed skills #42137, chore(recorder): render highlights via the shared element highlight #42253, and two pushes to main — and green on every other browser/OS. --enable-automation isn't a Firefox flag, so the diff doesn't reach this path.

_{Triaged by the Playwright bot - agent run}