Skip to content

Playwright is a framework for Web Testing and Automation. It allows testing Chromium, Firefox and WebKit with a single API.

chore: add playwright-devops skill for CI failure analysis (#40291)

Pavel Feldmanb86038bd105b
diff --git a/.claude/skills/playwright-devops/SKILL.md b/.claude/skills/playwright-devops/SKILL.md
new file mode 100644
index 0000000000..b4a9deccc6
--- /dev/null
+++ b/.claude/skills/playwright-devops/SKILL.md
@@ -0,0 +1,12 @@
+---
+name: playwright-devops
+description: DevOps workflows for Playwright - CI failure analysis, workflow debugging, and release operations.
+user_invocable: true
+---
+
+# Playwright DevOps
+
+## Guides
+
+- [CI Commit Failure Report](commit-failures.md) — analyze GitHub Actions failures for the last commit on main
+- [fetch-commit-logs.sh](fetch-commit-logs.sh) — script to download failed job logs into `~/tmp/commit-<sha>/`
diff --git a/.claude/skills/playwright-devops/commit-failures.md b/.claude/skills/playwright-devops/commit-failures.md
new file mode 100644
index 0000000000..6fc3e2ffd6
--- /dev/null
+++ b/.claude/skills/playwright-devops/commit-failures.md
@@ -0,0 +1,69 @@
+# CI Health Report
+
+Generate a CI health report for the last commit on the `main` branch of `microsoft/playwright`.
+This is an overall tree health report — not a commit regression analysis. The goal is to show
+the full picture of what's failing, grouping by root cause. If any failures appear to be
+regressions introduced by this specific commit, call that out, but most failures will be
+pre-existing flakes, infrastructure issues, or platform-specific problems.
+
+## Phase 1 — Fetch logs
+
+Run the fetch script to download all failed job logs into a local folder:
+
+```
+bash .claude/skills/playwright-devops/fetch-commit-logs.sh [<sha>]
+```
+
+- If no SHA is provided, it fetches the last commit on `main`.
+- Creates `~/tmp/commit-<short-sha>/` with:
+ - `summary.json` — commit info, failed workflows, and failed job metadata
+ - `<workflow-name>/<job-name>.log` — failed log output for each failed job
+
+**Note:** The script fetches failed jobs from both failed AND in-progress workflows.
+Workflows may still be running while some of their jobs have already failed — these
+must be included in the report. If any workflows are still in progress, note this in
+the report summary.
+
+## Phase 2 — Analyze
+
+1. **Read `summary.json`** to get the commit message and the list of failed workflows/jobs.
+
+2. **Read each `.log` file** and extract failing test names and error messages.
+
+3. **Compile the report leading with a summary**, then detailed tables:
+
+ ```markdown
+ # CI Health Report — <short-sha>
+
+ Commit: `<commit message>`
+
+ ## Summary
+
+ Brief overview: N workflows, N failed jobs, N total test failures.
+ Note if any workflows are still in progress.
+
+ ### Possible regressions (may be related to this commit)
+ - **N failures** in `test/file.spec.ts` across <browsers/platforms> — <brief description>
+
+ ### Pre-existing / flaky
+ - **N failures** in `test/file.spec.ts` — <brief description> (timeouts, infrastructure, platform-specific)
+
+ ### Infrastructure issues
+ - <description of non-test failures>
+
+ ---
+
+ ## Detailed Failures
+
+ ### Workflow: <name> (run <id>)
+
+ #### <job name> (job <id>) -- N failures
+ | Test | Error |
+ |------|-------|
+ | `path/to/test.spec.ts:line` -- test title | error message |
+ ```
+
+ The summary should appear first so readers immediately see what matters. Group related failures
+ (e.g. same test failing across browsers) into single summary bullet points rather than listing each individually.
+
+4. **Save the report** to `ci-failures-<short-sha>.md` in the repo root.
diff --git a/.claude/skills/playwright-devops/fetch-commit-logs.sh b/.claude/skills/playwright-devops/fetch-commit-logs.sh
new file mode 100755
index 0000000000..fb5bca862f
--- /dev/null
+++ b/.claude/skills/playwright-devops/fetch-commit-logs.sh
@@ -0,0 +1,107 @@
+#!/bin/bash
+set -euo pipefail
+
+# Usage: fetch-commit-logs.sh [<sha>]
+# Fetches CI failure logs for a commit into ~/tmp/commit-<short-sha>/
+
+REPO="microsoft/playwright"
+REF="${1:-main}"
+
+# Resolve commit
+COMMIT_JSON=$(gh api "repos/$REPO/commits/$REF" --jq '{sha: .sha, message: .commit.message}')
+SHA=$(echo "$COMMIT_JSON" | jq -r '.sha')
+SHORT_SHA="${SHA:0:7}"
+MESSAGE=$(echo "$COMMIT_JSON" | jq -r '.message' | head -1)
+
+OUTDIR="$HOME/tmp/commit-$SHORT_SHA"
+mkdir -p "$OUTDIR"
+
+echo "Commit: $SHORT_SHA — $MESSAGE"
+echo "Output: $OUTDIR"
+
+# Get all workflow runs for this commit
+RUNS_JSON=$(gh api "repos/$REPO/actions/runs?head_sha=$SHA&per_page=50" \
+ --jq '[.workflow_runs[] | {id, name, conclusion, workflow_id}]')
+
+# Filter to failed workflows
+FAILED_RUNS=$(echo "$RUNS_JSON" | jq -c '[.[] | select(.conclusion == "failure" or .conclusion == null)]')
+FAILED_COUNT=$(echo "$FAILED_RUNS" | jq 'length')
+
+if [ "$FAILED_COUNT" -eq 0 ]; then
+ echo "No failed workflows."
+ echo '{"sha":"'"$SHA"'","short_sha":"'"$SHORT_SHA"'","message":"'"$MESSAGE"'","workflows":[]}' | jq . > "$OUTDIR/summary.json"
+ echo "$OUTDIR"
+ exit 0
+fi
+
+echo "Found $FAILED_COUNT failed workflow(s). Fetching failed jobs..."
+
+# Build summary and collect jobs to fetch
+SUMMARY='{"sha":"'"$SHA"'","short_sha":"'"$SHORT_SHA"'","message":'"$(echo "$MESSAGE" | jq -Rs .)"',"workflows":[]}'
+
+sanitize() {
+ echo "$1" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9._-]/-/g' | sed 's/--*/-/g' | sed 's/^-//;s/-$//'
+}
+
+PIDS=()
+
+for i in $(seq 0 $((FAILED_COUNT - 1))); do
+ RUN=$(echo "$FAILED_RUNS" | jq -c ".[$i]")
+ RUN_ID=$(echo "$RUN" | jq -r '.id')
+ RUN_NAME=$(echo "$RUN" | jq -r '.name')
+ WORKFLOW_DIR=$(sanitize "$RUN_NAME")
+
+ # Get failed jobs for this run
+ JOBS_JSON=$(gh run view "$RUN_ID" --json jobs \
+ --jq '[.jobs[] | select(.conclusion == "failure") | {name: .name, id: .databaseId}]')
+ JOB_COUNT=$(echo "$JOBS_JSON" | jq 'length')
+
+ if [ "$JOB_COUNT" -eq 0 ]; then
+ continue
+ fi
+
+ mkdir -p "$OUTDIR/$WORKFLOW_DIR"
+
+ # Add to summary
+ WORKFLOW_ENTRY=$(jq -n \
+ --arg name "$RUN_NAME" \
+ --argjson id "$RUN_ID" \
+ --argjson jobs "$JOBS_JSON" \
+ '{name: $name, id: $id, failed_jobs: $jobs}')
+ SUMMARY=$(echo "$SUMMARY" | jq --argjson w "$WORKFLOW_ENTRY" '.workflows += [$w]')
+
+ # Fetch logs in parallel
+ for j in $(seq 0 $((JOB_COUNT - 1))); do
+ JOB=$(echo "$JOBS_JSON" | jq -c ".[$j]")
+ JOB_ID=$(echo "$JOB" | jq -r '.id')
+ JOB_NAME=$(echo "$JOB" | jq -r '.name')
+ JOB_FILE=$(sanitize "$JOB_NAME")
+ LOG_PATH="$OUTDIR/$WORKFLOW_DIR/$JOB_FILE.log"
+
+ (
+ echo "# $JOB_NAME" > "$LOG_PATH"
+ echo "" >> "$LOG_PATH"
+ gh run view --job "$JOB_ID" --log-failed 2>&1 | sed 's/\x1b\[[0-9;]*[a-zA-Z]//g' | sed $'s/^[^\t]*\t[^\t]*\t\xef\xbb\xbf\{0,1\}[0-9T:.Z-]* //' >> "$LOG_PATH" || true
+ # If only header (e.g. workflow still in progress), fetch via jobs API
+ if [ "$(wc -l < "$LOG_PATH")" -le 3 ]; then
+ echo "# $JOB_NAME" > "$LOG_PATH"
+ echo "" >> "$LOG_PATH"
+ gh api "repos/$REPO/actions/jobs/$JOB_ID/logs" 2>&1 | sed 's/\x1b\[[0-9;]*[a-zA-Z]//g' | sed $'s/\xef\xbb\xbf//g' | sed 's/^[0-9T:.Z-]* //' >> "$LOG_PATH" || true
+ fi
+ echo " Fetched: $WORKFLOW_DIR/$JOB_FILE.log"
+ ) &
+ PIDS+=($!)
+ done
+done
+
+# Wait for all parallel fetches
+for pid in "${PIDS[@]}"; do
+ wait "$pid" 2>/dev/null || true
+done
+
+# Write summary
+echo "$SUMMARY" | jq . > "$OUTDIR/summary.json"
+
+echo ""
+echo "Done. Logs saved to: $OUTDIR"
+echo "$OUTDIR"