|
| 1 | +name: evidence-demo |
| 2 | + |
| 3 | +# Throwaway: produces evidence packs as Actions artifacts on GitHub-hosted |
| 4 | +# runners, so the hosted viewer can open them from a link. Not part of any PR. |
| 5 | +on: |
| 6 | + push: |
| 7 | + branches: [evidence-demo] |
| 8 | + workflow_dispatch: |
| 9 | + |
| 10 | +env: |
| 11 | + E2E_TELEMETRY_DISABLED: '1' |
| 12 | + |
| 13 | +permissions: |
| 14 | + contents: read |
| 15 | + actions: read |
| 16 | + |
| 17 | +jobs: |
| 18 | + testbed: |
| 19 | + runs-on: ubuntu-latest |
| 20 | + timeout-minutes: 30 |
| 21 | + steps: |
| 22 | + - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 |
| 23 | + - uses: pnpm/action-setup@9fd676a19091d4595eefd76e4bd31c97133911f1 # v4.2.0 |
| 24 | + - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0 |
| 25 | + with: |
| 26 | + node-version: 26 |
| 27 | + cache: pnpm |
| 28 | + - run: pnpm install --frozen-lockfile |
| 29 | + - run: pnpm run build |
| 30 | + - name: Install browsers |
| 31 | + run: pnpm --filter @e2e-dev/web exec playwright install chromium --with-deps |
| 32 | + - name: Testbed |
| 33 | + run: pnpm --filter @e2e-dev/testbed test |
| 34 | + - name: Upload evidence |
| 35 | + id: evidence |
| 36 | + if: ${{ !cancelled() }} |
| 37 | + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 |
| 38 | + with: |
| 39 | + name: evidence-testbed |
| 40 | + path: apps/testbed/.e2e/evidence/*.evidence |
| 41 | + if-no-files-found: error |
| 42 | + retention-days: 30 |
| 43 | + - name: Link the evidence |
| 44 | + if: ${{ !cancelled() && steps.evidence.outputs.artifact-url != '' }} |
| 45 | + env: |
| 46 | + ARTIFACT_URL: ${{ steps.evidence.outputs.artifact-url }} |
| 47 | + run: | |
| 48 | + link="https://evidence.lambdatest.com/?pack=$(node -e 'process.stdout.write(encodeURIComponent(process.env.ARTIFACT_URL))')" |
| 49 | + echo "[Open the testbed run in the evidence viewer]($link)" >> "$GITHUB_STEP_SUMMARY" |
| 50 | + echo "EVIDENCE_LINK $link" |
| 51 | +
|
| 52 | + web-benchmark-agent: |
| 53 | + runs-on: ubuntu-latest |
| 54 | + timeout-minutes: 30 |
| 55 | + steps: |
| 56 | + - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 |
| 57 | + - uses: pnpm/action-setup@9fd676a19091d4595eefd76e4bd31c97133911f1 # v4.2.0 |
| 58 | + - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0 |
| 59 | + with: |
| 60 | + node-version: 26 |
| 61 | + cache: pnpm |
| 62 | + - run: pnpm install --frozen-lockfile |
| 63 | + - run: pnpm run build |
| 64 | + - name: Install browsers |
| 65 | + run: pnpm --filter @e2e-dev/web exec playwright install chromium --with-deps |
| 66 | + # Replays the committed recordings; with no model key a step that needs one is recorded as broken. |
| 67 | + - name: Agentic suite from recordings |
| 68 | + continue-on-error: true |
| 69 | + run: pnpm --filter @e2e-dev/web-benchmark run test:agent --strict-cache |
| 70 | + - name: Upload evidence |
| 71 | + id: evidence |
| 72 | + if: ${{ !cancelled() }} |
| 73 | + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 |
| 74 | + with: |
| 75 | + name: evidence-web-benchmark-agent |
| 76 | + path: apps/web-benchmark/.e2e/evidence/*.evidence |
| 77 | + if-no-files-found: error |
| 78 | + retention-days: 30 |
| 79 | + - name: Link the evidence |
| 80 | + if: ${{ !cancelled() && steps.evidence.outputs.artifact-url != '' }} |
| 81 | + env: |
| 82 | + ARTIFACT_URL: ${{ steps.evidence.outputs.artifact-url }} |
| 83 | + run: | |
| 84 | + link="https://evidence.lambdatest.com/?pack=$(node -e 'process.stdout.write(encodeURIComponent(process.env.ARTIFACT_URL))')" |
| 85 | + echo "[Open the web benchmark agent run in the evidence viewer]($link)" >> "$GITHUB_STEP_SUMMARY" |
| 86 | + echo "EVIDENCE_LINK $link" |
0 commit comments