feat(aria): treat frame elements like iframes in aria snapshots

[dgozman]

Summary

• <frame> elements now appear as iframe nodes in aria snapshots, and their content is stitched into ai snapshots
• child frameset documents are stitched recursively
• selector generation treats frames like iframes, preferring frame[name=...]/frame[title=...]

References #41784.

[github-actions]

Test results for "MCP"

3 failed
❌ [chromium] › mcp/annotate.spec.ts:110 › should abort annotation when last screenshot is removed @mcp-macos-latest-chromium
❌ [firefox] › mcp/cli-killall.spec.ts:42 › kill-all kills filtered dashboard pid @mcp-ubuntu-latest-firefox
❌ [webkit] › mcp/annotate.spec.ts:110 › should abort annotation when last screenshot is removed @mcp-macos-latest-webkit

7757 passed, 1249 skipped

---

Merge workflow run.

[github-actions]

Test results for "tests 1"

3 flaky

⚠️ [chromium-library] › library/video.spec.ts:736 › screencast › should work with video+trace `@realtime-time-library-chromium-linux`
⚠️ [firefox-library] › library/inspector/cli-codegen-3.spec.ts:255 › cli codegen › should generate frame locators (4) `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-page] › page/page-emulate-media.spec.ts:144 › should keep reduced motion and color emulation after reload `@firefox-ubuntu-22.04-node20`

49956 passed, 1158 skipped

---

Merge workflow run.

[github-actions]

Hi, I'm the Playwright bot and I took a first look at the CI failures here.

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

The only real failures are 3 MCP tests, and each one flips verdict across ~250 runs on SHAs that have nothing to do with this PR. This change only touches aria-snapshot and selector-generation code, which the MCP annotate and CLI flows don't exercise.

Details

Overall: No failure traces back to this diff. The PR edits ariaSnapshot.ts, roleUtils.ts, selectorGenerator.ts, page.ts and two aria-snapshot specs — none of that reaches the failing MCP tests.

Pre-existing flake / infra

• [chromium] & [webkit] › mcp/annotate.spec.ts:110 › should abort annotation when last screenshot is removed — flaky across the results DB, failing on many unrelated runs: PRs devops(test-results-db): CLI to compact CI test results into DuckDB #41673 (webkit), feat(webkit): roll to r2327 #41684 (chrome), devops: use node lts/* for non-test workflows #41716 (msedge), fix(reporter): expose setup/teardown projects to preprocessSuite #41731 (chromium), feat(skills): add component-testing agent skill #41738 (webkit), chore(recorder): move signals from the action onto the action context #41755 (webkit), and push commits bddd05f5/a553b746/ba4a0936 (firefox). Bimodal on every browser (e.g. firefox 11 failed / 232 passed of 243 runs; webkit 2/250; chromium 3/248).
• [firefox] › mcp/cli-killall.spec.ts:42 › kill-all kills filtered dashboard pid — same story: fails on PRs fix(mcp): preserve directory for additional tab videos #41680 and fix(test runner): handle fixture option overrides #41821 (chromium) and push commits 696225c6/8c6ae3b5/4650ba4b/91565f0d/4582949c across chromium, firefox and webkit. Bimodal everywhere (firefox 4/243, chromium 8/248, webkit 3/250).

Both are MCP tests; the PR doesn't change any MCP code path, so nothing here is caused by it.

_{Triaged by the Playwright bot - agent run}