Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →MANDATORY: Activate this skill ANY TIME you need to build the project, run tests, or verify code health in DevTools. You MUST use this skill before executing commands like npm test, npm run build, autoninja, or linters, as it contains critical, repository-specific instructions on how to correctly format these commands, filter test runs, and interpret failures.
.claude/skills/chromedevtools-devtools-verification/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -84% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -73% | 0% |
npm run test -- <FILEPATH> where FILEPATH is a path to a *.test.ts file, relative to the working directory.front_end/panels/timeline/TimelinePanel.ts, you would expect the test file to be defined in front_end/panels/timeline/TimelinePanel.test.ts.npm run test -- front_end/models/trace will run all tests in that directory.autoninja -C out/Default.npm run lint will execute ESLint and StyleLint. It will report any violations and automatically fix them where possible.npm run lint -- <PATH> where PATH is a path to a file or directory. This will also automatically fix violations where possible.Use scripts/tools/get_cl_test_results.py to fetch all tryjob build statuses, per-test failure logs, Blink web test text diffs, and screenshot artifacts (actual_image, expected_image, image_diff) for a CL in one command:
bash# Fetch all build statuses, failure logs, and artifacts for a CL into an output directory: scripts/tools/get_cl_test_results.py --cl <CL_NUMBER> --output-dir <OUTPUT_DIR> # Filter for specific failing tests (e.g. DOMTreeWidget) or specific builders: scripts/tools/get_cl_test_results.py --cl <CL_NUMBER> --test-filter "DOMTreeWidget" --output-dir <OUTPUT_DIR> # Also apply Linux actual_image screenshot artifacts directly to test/goldens/linux/...: scripts/tools/get_cl_test_results.py --cl <CL_NUMBER> --output-dir <OUTPUT_DIR> --apply-goldens
test/goldens/linux/...) produced by Linux trybots (dtf_linux_rel / dtf_linux_dbg). Do not run screenshot tests locally on macOS to verify goldens.actual_image.png, expected_image.png, and image_diff.png directly using your multimodal file viewer (view_file). Never write or run programmatic image comparison scripts.git cl presubmit -u at the end of your code changes.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 24,700 | 2,954 | -88% | 1 | 1 | 0% | 1,999 | 1,117 | -44% | 0 | 0 | — |
case-02 | fail→pass | 16,237 | 4,434 | -73% | 1 | 1 | 0% | 2,574 | 1,343 | -48% | 0 | 0 | — |
case-03 | fail→fail | 19,901 | 15,918 | -20% | 1 | 1 | 0% | 2,684 | 2,293 | -15% | 0 | 0 | — |
case-04 | fail→pass | 17,432 | 3,865 | -78% | 1 | 1 | 0% | 2,626 | 983 | -63% | 0 | 0 | — |
case-05 | fail→pass | 41,226 | 9,193 | -78% | 1 | 1 | 0% | 8,253 | 1,310 | -84% | 0 | 0 | — |
case-06 | fail→pass | 33,963 | 4,567 | -87% | 1 | 1 | 0% | 4,411 | 1,197 | -73% | 0 | 0 | — |
case-07 | fail→pass | 18,941 | 4,864 | -74% | 1 | 1 | 0% | 2,950 | 1,499 | -49% | 0 | 0 | — |
case-08 | fail→pass | 12,756 | 5,106 | -60% | 1 | 1 | 0% | 1,844 | 1,239 | -33% | 0 | 0 | — |
case-09 | fail→fail | 16,282 | 3,107 | -81% | 1 | 1 | 0% | 2,515 | 1,176 | -53% | 0 | 0 | — |
case-10 | fail→pass | 8,620 | 2,549 | -70% | 1 | 1 | 0% | 1,095 | 1,092 | -0% | 0 | 0 | — |
case-11 | pass→pass | 11,376 | 2,680 | -76% | 1 | 1 | 0% | 1,662 | 921 | -45% | 0 | 0 | — |
case-12 | pass→pass | 15,680 | 5,189 | -67% | 1 | 1 | 0% | 2,421 | 1,304 | -46% | 0 | 0 | — |
case-13 | fail→fail | 9,044 | 7,227 | -20% | 1 | 1 | 0% | 1,191 | 1,161 | -3% | 0 | 0 | — |
case-14 | pass→pass | 52,610 | 8,997 | -83% | 1 | 1 | 0% | 2,253 | 1,861 | -17% | 0 | 0 | — |
case-15 | pass→fail | 15,652 | 2,964 | -81% | 1 | 1 | 0% | 2,096 | 996 | -52% | 0 | 0 | — |
case-16 | fail→pass | 14,291 | 4,108 | -71% | 1 | 1 | 0% | 2,052 | 1,115 | -46% | 0 | 0 | — |
case-17 | fail→pass | 15,126 | 2,957 | -80% | 1 | 1 | 0% | 1,967 | 989 | -50% | 0 | 0 | — |
case-18 | fail→pass | 9,792 | 4,410 | -55% | 1 | 1 | 0% | 1,569 | 1,140 | -27% | 0 | 0 | — |
case-19 | fail→fail | 15,593 | 12,644 | -19% | 1 | 1 | 0% | 2,278 | 1,577 | -31% | 0 | 0 | — |
case-20 | pass→pass | 12,978 | 9,562 | -26% | 1 | 1 | 0% | 2,208 | 2,298 | +4% | 0 | 0 | — |
case-21 | pass→pass | 17,338 | 37,991 | +119% | 1 | 1 | 0% | 2,190 | 2,311 | +6% | 0 | 0 | — |
case-22 | pass→fail | 25,461 | 5,678 | -78% | 1 | 1 | 0% | 2,636 | 902 | -66% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 21 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/11/2026 | +46% |
Other measured skills in the registry, with their headline benchmark lift.