upstream-twenty/.github/workflows/ci-e2e-main.yaml

name: CI E2E Main

on:
  push:
    branches: [main]

  pull_request:
    types: [labeled, synchronize, opened, reopened]

  # Manual QA Scout run against an already-merged PR (executes on main).
  workflow_dispatch:
    inputs:
      pr_number:
        description: "Merged PR number for the QA Scout (defaults to the PR of the latest main commit)"
        required: false

permissions:
  contents: read

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}

jobs:
  e2e-test:
    if: >
      github.event_name == 'push' ||
      github.event_name == 'workflow_dispatch' ||
      (github.event_name == 'pull_request' &&
       contains(github.event.pull_request.labels.*.name, 'qa-scout')) ||
      (github.event_name == 'pull_request' &&
       contains(github.event.pull_request.labels.*.name, 'run-merge-queue'))
    runs-on: ubuntu-latest-8-cores
    # 60 so build + suite plus the QA Scout's 15-minute step cap cannot hit
    # the job ceiling: a job timeout overrides step continue-on-error and
    # would redden the status check.
    timeout-minutes: 60
    outputs:
      qa-scout-run: ${{ steps.qa-scout-pr.outputs.run }}
      qa-scout-pr-number: ${{ steps.qa-scout-pr.outputs.pr_number }}
    env:
      NODE_OPTIONS: "--max-old-space-size=10240"
    services:
      postgres:
        image: postgres:18
        env:
          POSTGRES_USER: postgres
          POSTGRES_PASSWORD: postgres
        ports:
          - 5432:5432
        options: >-
          --health-cmd pg_isready
          --health-interval 10s
          --health-timeout 5s
          --health-retries 5
      redis:
        image: redis
        ports:
          - 6379:6379
    steps:
      # Audit only: records every outbound endpoint this job touches in the
      # StepSecurity insights, blocks nothing, changes no behavior. This is
      # the observation phase for the QA Scout containment follow-up: the
      # agent step's allowlist scopes what it does, not where the runner can
      # reach, so once the job's endpoint set is known and stable from a few
      # weeks of audit data, egress-policy block with that allowlist makes
      # real-time exfiltration impossible rather than merely unscoped.
      - name: Harden runner (egress audit)
        # Fail-open while auditing: an observer must never redden the status
        # check. The eventual block-mode flip drops this so it fails closed.
        continue-on-error: true
        uses: step-security/harden-runner@05e31511f85b41b11d1cf0ef85d0992719546e2c # v2.21.0
        with:
          egress-policy: audit

      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
        with:
          fetch-depth: 10

      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: lts/*

      - name: Install dependencies
        uses: ./.github/actions/yarn-install

      - name: Restore Nx build cache
        uses: actions/cache/restore@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0 (restore)
        with:
          key: v4-e2e-build-${{ github.ref_name }}-${{ github.sha }}
          restore-keys: |
            v4-e2e-build-${{ github.ref_name }}-
            v4-e2e-build-main-
          path: |
            .nx
            node_modules/.cache
            packages/*/node_modules/.cache

      - name: Build twenty-shared
        run: npx nx build twenty-shared

      - name: Install Playwright Browsers
        run: npx nx setup twenty-e2e-testing

      - name: Setup environment files
        run: |
          cp packages/twenty-front/.env.example packages/twenty-front/.env
          npx nx reset:env:e2e-testing-server twenty-server

      - name: Build frontend
        run: NODE_ENV=production npx nx build twenty-front

      - name: Build server
        run: npx nx build twenty-server

      - name: Save Nx build cache
        if: always()
        uses: actions/cache/save@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0 (save)
        with:
          key: v4-e2e-build-${{ github.ref_name }}-${{ github.sha }}
          path: |
            .nx
            node_modules/.cache
            packages/*/node_modules/.cache

      - name: Create and setup database
        run: |
          PGPASSWORD=postgres psql -h localhost -p 5432 -U postgres -d postgres -c 'CREATE DATABASE "default";'
          PGPASSWORD=postgres psql -h localhost -p 5432 -U postgres -d postgres -c 'CREATE DATABASE "test";'
          npx nx run twenty-server:database:reset

      # Cookie sessions are credentialed, so the front only authenticates from
      # the API's own origin, workspace subdomains included.
      - name: Serve the frontend from the server
        run: cp -r packages/twenty-front/build packages/twenty-server/dist/front

      # Logs are teed to files so the QA Scout can diff its own log window; the
      # console output is unchanged.
      - name: Start server
        run: |
          mkdir -p /tmp/qa-scout
          npx nx run twenty-server:start:ci 2>&1 | tee /tmp/qa-scout/server.log &
          echo "Waiting for server to be ready..."
          timeout 120 bash -c 'until curl -sf http://localhost:3000/healthz > /dev/null; do sleep 2; done'

      - name: Start worker
        run: |
          npx nx run twenty-server:worker 2>&1 | tee /tmp/qa-scout/worker.log &
          echo "Worker started"

      - name: Run Playwright tests
        env:
          FRONTEND_BASE_URL: http://localhost:3000
        run: npx nx test twenty-e2e-testing

      - name: Upload Playwright results
        if: always()
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: playwright-results
          path: |
            packages/twenty-e2e-testing/run_results/
            packages/twenty-e2e-testing/test-results/
          retention-days: 7

      # ── QA Scout ─────────────────────────────────────────────────────────
      # Post-merge browser QA agent (.claude/skills/qa-scout/SKILL.md): scopes
      # scenarios from the merged PR's diff, drives the already-running app,
      # watches server/worker logs, and writes a verdict. Shadow-safe: every
      # step is continue-on-error so the Scout can never redden main CI. Runs
      # after the deterministic suite even when it failed (a broken main is
      # exactly when a loud report helps), but not on merge-queue PR runs.
      - name: QA Scout / Prepare merged PR context
        id: qa-scout-pr
        if: >-
          ${{ !cancelled() && (
            github.event_name == 'push' ||
            github.event_name == 'workflow_dispatch' ||
            (github.event_name == 'pull_request' &&
             contains(github.event.pull_request.labels.*.name, 'qa-scout') &&
             github.event.pull_request.head.repo.full_name == github.repository)
          ) }}
        continue-on-error: true
        env:
          GH_TOKEN: ${{ github.token }}
          INPUT_PR_NUMBER: ${{ github.event.inputs.pr_number }}
          EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
        run: |
          set -euo pipefail
          mkdir -p /tmp/qa-scout/context /tmp/qa-scout/output /tmp/qa-scout/browser
          MODE="post-merge"
          if [ "${GITHUB_EVENT_NAME}" = "pull_request" ]; then
            # Label-triggered pre-merge validation: the checkout IS the PR
            # head, so the Scout QAs the PR itself before it merges.
            MODE="pre-merge"
            PR_NUMBER="${EVENT_PR_NUMBER:-}"
          else
            PR_NUMBER="${INPUT_PR_NUMBER:-}"
            if [ -z "$PR_NUMBER" ]; then
              PR_NUMBER=$(gh api "repos/${GITHUB_REPOSITORY}/commits/${GITHUB_SHA}/pulls" --jq '.[0].number // empty' || true)
            fi
          fi
          if ! [[ "$PR_NUMBER" =~ ^[0-9]+$ ]]; then
            echo "No target PR resolved; skipping QA Scout."
            echo "run=false" >> "$GITHUB_OUTPUT"
            exit 0
          fi
          gh pr view "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" \
            --json number,title,body,author,url,mergedAt > /tmp/qa-scout/context/pr.json
          # The post-merge Scout QAs main, so an unmerged PR (reachable via
          # manual dispatch) would get a verdict about code main does not
          # contain. Pre-merge label runs check out the PR head, so the guard
          # does not apply there.
          if [ "$MODE" = "post-merge" ] && [ -z "$(jq -r '.mergedAt // empty' /tmp/qa-scout/context/pr.json)" ]; then
            echo "PR #${PR_NUMBER} is not merged; skipping QA Scout."
            echo "run=false" >> "$GITHUB_OUTPUT"
            exit 0
          fi
          printf '%s\n' "$MODE" > /tmp/qa-scout/context/mode
          TITLE=$(jq -r '.title' /tmp/qa-scout/context/pr.json)
          # Mirrors mustBeQa in the eng app: [no-qa]/[noqa] escape hatch, i18n
          # translation PRs are auto-generated and never QA candidates.
          LOWER=$(printf '%s' "$TITLE" | tr '[:upper:]' '[:lower:]')
          case "$LOWER" in
            *"[no-qa]"* | *"[noqa]"*)
              echo "Skipping QA Scout: [no-qa] in PR title."
              echo "run=false" >> "$GITHUB_OUTPUT"
              exit 0
              ;;
            i18n*translation*)
              echo "Skipping QA Scout: i18n translation PR."
              echo "run=false" >> "$GITHUB_OUTPUT"
              exit 0
              ;;
          esac
          # gh rejects --slurp together with --jq, so aggregate the per-page
          # jq output with a local jq instead.
          gh api --paginate "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/files?per_page=100" \
            --jq '.[] | {path: .filename, additions, deletions, status}' \
            | jq -s '.' > /tmp/qa-scout/context/files.json
          gh pr diff "$PR_NUMBER" --repo "$GITHUB_REPOSITORY" > /tmp/qa-scout/full.diff
          DIFF_BYTES=$(wc -c < /tmp/qa-scout/full.diff)
          head -c 2000000 /tmp/qa-scout/full.diff > /tmp/qa-scout/context/pr.diff
          if [ "$DIFF_BYTES" -gt 2000000 ]; then
            printf '\n[pr.diff truncated: 2000000 of %s bytes shown; files.json is complete]\n' "$DIFF_BYTES" >> /tmp/qa-scout/context/pr.diff
          fi
          rm /tmp/qa-scout/full.diff
          cat > /tmp/qa-scout/mcp.json <<'EOF'
          {
            "mcpServers": {
              "playwright": {
                "command": "npx",
                "args": [
                  "-y",
                  "@playwright/mcp@0.0.79",
                  "--headless",
                  "--browser", "chrome",
                  "--isolated",
                  "--output-dir", "/tmp/qa-scout/browser",
                  "--allowed-origins", "http://localhost:3000;http://127.0.0.1:3000;http://app.localhost:3000;http://apple.localhost:3000"
                ]
              }
            }
          }
          EOF
          echo "run=true" >> "$GITHUB_OUTPUT"
          echo "pr_number=$PR_NUMBER" >> "$GITHUB_OUTPUT"

      # The Scout invokes the Claude Code CLI directly rather than through
      # claude-code-action: the action rejects push events outright
      # ("Unsupported event type: push", still true at its latest version),
      # which silently killed the post-merge mode this Scout exists for, and
      # its GitHub bootstrap was the only reason a GitHub token entered this
      # step at all. Direct invocation runs on every trigger and leaves the
      # agent holding no GitHub credential whatsoever; the PR comment is
      # posted by the qa-scout-comment job from the uploaded verdict, so
      # untrusted page/log content can never reach a privileged credential.
      # The Bash allowlist scopes the agent to its task; it is not the
      # containment boundary (psql alone is arbitrarily powerful in this
      # disposable env). Containment is the absence of credentials beyond
      # inference, the scrub step, and Actions secret masking on the token.
      - name: QA Scout / Run
        id: qa-scout-run
        if: ${{ !cancelled() && steps.qa-scout-pr.outputs.run == 'true' }}
        continue-on-error: true
        timeout-minutes: 15
        env:
          CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
          PGPASSWORD: postgres
        run: |
          set -euo pipefail
          npm install -g @anthropic-ai/claude-code@2.1.247
          # stream-json teed into the artifact tree, ahead of the scrub,
          # keeps per-turn detail and permission denials so allowlist tuning
          # reads real data. stderr and the exit code are kept too: a run that
          # never reaches a result line (bad flag, exhausted quota) explains
          # itself only there, and the summary step reads both back.
          set +e
          claude -p "Read .claude/skills/qa-scout/SKILL.md and follow it exactly. This is the CI invocation: every path and credential in the skill's Inputs table applies verbatim." \
            --max-turns 150 \
            --model opus \
            --add-dir /tmp/qa-scout \
            --strict-mcp-config \
            --mcp-config /tmp/qa-scout/mcp.json \
            --allowedTools "Read,Grep,Glob,Write,TodoWrite,Bash(cat *),Bash(tail *),Bash(head *),Bash(wc *),Bash(ls *),Bash(grep *),Bash(jq *),Bash(echo *),Bash(diff *),Bash(sort *),Bash(uniq *),Bash(cut *),Bash(tr *),Bash(mkdir *),Bash(sleep *),Bash(date *),Bash(psql *),mcp__playwright" \
            --output-format stream-json --verbose \
            2> >(tee /tmp/qa-scout/agent-stderr.log >&2) \
            | tee /tmp/qa-scout/execution.json
          AGENT_EXIT=${PIPESTATUS[0]}
          set -e
          printf '%s' "$AGENT_EXIT" > /tmp/qa-scout/agent-exit.txt
          exit "$AGENT_EXIT"

      # The agent process necessarily holds the inference credential, and a
      # prompt-injected agent could copy it into any file it can write. The
      # artifact ships the whole /tmp/qa-scout tree (context, browser output,
      # logs, report), so redact Anthropic token shapes across all of it, in
      # the trusted lane, before anything is published.
      - name: QA Scout / Scrub outputs
        if: ${{ !cancelled() && steps.qa-scout-pr.outputs.run == 'true' }}
        continue-on-error: true
        run: |
          set -euo pipefail
          # NUL-delimited paths and no -I: an agent-chosen filename with
          # spaces or newlines, or a token tucked into a binary-looking file,
          # must not escape redaction.
          if grep -rq 'sk-ant-' /tmp/qa-scout 2>/dev/null; then
            echo "Redacting Anthropic token shapes from Scout files:"
            grep -rl 'sk-ant-' /tmp/qa-scout 2>/dev/null
            grep -rlZ 'sk-ant-' /tmp/qa-scout 2>/dev/null \
              | xargs -0 sed -i -E 's/sk-ant-[A-Za-z0-9_-]+/[REDACTED]/g'
          fi

      # Always leaves a breadcrumb on push/dispatch runs so a broken prepare
      # step can never disable the Scout silently.
      - name: QA Scout / Job summary
        if: ${{ !cancelled() && steps.qa-scout-pr.outcome != 'skipped' }}
        continue-on-error: true
        env:
          SCOUT_RUN: ${{ steps.qa-scout-pr.outputs.run }}
          SCOUT_PREP_OUTCOME: ${{ steps.qa-scout-pr.outcome }}
        run: |
          # Why the run ended, from the evidence the agent leaves behind: the
          # final stream-json result when it produced one, stderr when it died
          # before that. Without this the summary can only say "no report".
          scout_failure_reason() {
            local result_line exit_code
            exit_code=$(cat /tmp/qa-scout/agent-exit.txt 2>/dev/null || echo "unknown")
            echo "**Why it stopped** (exit code \`${exit_code}\`)"
            echo ""
            result_line=$(grep -a '"type":"result"' /tmp/qa-scout/execution.json 2>/dev/null | tail -1 || true)
            if [ -n "$result_line" ]; then
              printf '%s' "$result_line" | jq -r '
                "- outcome: `\(.subtype // "unknown")`" +
                " · \(.num_turns // "?") turns" +
                " · \(((.duration_ms // 0) / 60000 * 10 | round) / 10) min" +
                " · $\(((.total_cost_usd // 0) * 100 | round) / 100)",
                (if ((.errors? // []) | length) > 0 then "- error: " + ((.errors | join("; "))) else empty end)
              ' 2>/dev/null || echo "- the final result line could not be parsed; see the artifact"
            else
              echo "- the agent produced no result line, so it failed before or during startup"
            fi
            if [ -s /tmp/qa-scout/agent-stderr.log ]; then
              echo ""
              echo "<details><summary>stderr (last 10 lines)</summary>"
              echo ""
              echo '```'
              tail -n 10 /tmp/qa-scout/agent-stderr.log
              echo '```'
              echo ""
              echo "</details>"
            fi
          }
          {
            echo "## QA Scout"
            if [ -f /tmp/qa-scout/output/report.md ]; then
              if [ "$(jq -r '.status // "final"' /tmp/qa-scout/output/verdict.json 2>/dev/null || echo final)" != "final" ]; then
                echo "> [!WARNING]"
                echo "> The Scout stopped before finishing; the partial state it left follows."
                echo ""
                scout_failure_reason
                echo ""
              fi
              cat /tmp/qa-scout/output/report.md
            elif [ -f /tmp/qa-scout/output/verdict.json ]; then
              jq -r '"**\(.verdict)**: \(.headline // "no headline recorded")\n\n_Full report missing (agent stopped early)._"' /tmp/qa-scout/output/verdict.json
              echo ""
              scout_failure_reason
            elif [ "$SCOUT_RUN" = "true" ]; then
              echo "The Scout produced no verdict."
              echo ""
              scout_failure_reason
            elif [ "$SCOUT_PREP_OUTCOME" = "failure" ]; then
              echo "Did not run: the prepare step failed; see its logs."
            else
              echo "Did not run: skipped by the prepare step (no merged PR, [no-qa], or i18n translation PR)."
            fi
          } >> "$GITHUB_STEP_SUMMARY"
          # Surfaced on the run page too, so a dead Scout is visible without
          # opening the summary.
          if [ "$SCOUT_RUN" = "true" ] && [ ! -f /tmp/qa-scout/output/report.md ]; then
            echo "::warning title=QA Scout produced no report::$(head -n 1 /tmp/qa-scout/agent-stderr.log 2>/dev/null || echo 'see the job summary for why')"
          fi

      - name: QA Scout / Upload report
        if: ${{ !cancelled() && steps.qa-scout-pr.outputs.run == 'true' }}
        continue-on-error: true
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: qa-scout-report
          path: |
            /tmp/qa-scout/context/
            /tmp/qa-scout/output/
            /tmp/qa-scout/browser/
            /tmp/qa-scout/server.log
            /tmp/qa-scout/worker.log
            /tmp/qa-scout/execution.json
            /tmp/qa-scout/agent-stderr.log
            /tmp/qa-scout/agent-exit.txt
          if-no-files-found: ignore
          retention-days: 14

  ci-e2e-main-status-check:
    if: always() && !cancelled()
    timeout-minutes: 5
    runs-on: ubuntu-latest
    needs: [e2e-test]
    steps:
      - name: Fail job if any needs failed
        if: contains(needs.*.result, 'failure')
        run: exit 1

  notify-main-ci-failure:
    if: always() && github.event_name == 'push' && contains(needs.*.result, 'failure')
    timeout-minutes: 5
    runs-on: ubuntu-latest
    needs: [e2e-test]
    steps:
      - name: Ping engineering on main CI failure
        run: |
          curl -fsS -X POST https://engineering.twenty.com/s/main-ci-failing \
            -H 'Content-Type: application/json' \
            -d '{
              "commit": "${{ github.sha }}",
              "actor": "${{ github.actor }}",
              "run_url": "${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
            }'

  # Posts the QA Scout verdict as a comment on the merged PR. Separate job so
  # pull-requests:write never coexists with the agent step: the agent reads
  # untrusted page/log content, so the write TARGET is bound to the prepare
  # step's trusted pr_number output and the agent-authored artifact can only
  # influence the comment's content, never where it lands.
  # Quiet on PASS; FAIL/INVESTIGATE post (or update) a single marked comment.
  # Kill switch: set repo variable QA_SCOUT_DISABLE_COMMENTS=true.
  qa-scout-comment:
    if: >
      always() && !cancelled() &&
      needs.e2e-test.outputs.qa-scout-run == 'true' &&
      vars.QA_SCOUT_DISABLE_COMMENTS != 'true'
    timeout-minutes: 5
    runs-on: ubuntu-latest
    needs: [e2e-test]
    permissions:
      contents: read
      pull-requests: write
    steps:
      - name: Download QA Scout report
        id: download
        continue-on-error: true
        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
        with:
          name: qa-scout-report
          path: qa-scout-report

      - name: Post verdict comment
        if: steps.download.outcome == 'success'
        env:
          GH_TOKEN: ${{ github.token }}
          RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
          PR_NUMBER: ${{ needs.e2e-test.outputs.qa-scout-pr-number }}
        run: |
          set -euo pipefail
          if ! [[ "$PR_NUMBER" =~ ^[0-9]+$ ]]; then
            echo "No trusted PR number from the prepare step; nothing to post."
            exit 0
          fi
          VERDICT_FILE=$(find qa-scout-report -name verdict.json | head -1 || true)
          REPORT_FILE=$(find qa-scout-report -name report.md | head -1 || true)
          if [ -z "$VERDICT_FILE" ]; then
            echo "No verdict in the artifact; nothing to post."
            exit 0
          fi
          # A run killed on turns or time can leave the verdict without the
          # report; a synthesized body beats silence on a FAIL.
          if [ -z "$REPORT_FILE" ]; then
            REPORT_FILE=report-fallback.md
            jq -r '"**\(.verdict)**: \(.headline // "no headline recorded")\n\n" + ([.scenarios[]? | "- \(.name): \(.result)"] | join("\n")) + "\n\n_Full report missing (agent stopped early); details in the qa-scout-report artifact._"' \
              "$VERDICT_FILE" > "$REPORT_FILE"
          fi
          VERDICT=$(jq -r '.verdict // empty' "$VERDICT_FILE")
          case "$VERDICT" in
            PASS | INVESTIGATE | FAIL) ;;
            *)
              echo "Invalid or missing verdict '${VERDICT}' in verdict.json; not posting."
              exit 0
              ;;
          esac
          # A checkpoint left by a run that died is not a finding: it says
          # the Scout failed, not that the PR is broken. Only a finished run
          # speaks, unless the partial state already caught something real: a
          # failed scenario, or a top-level FAIL (which is how a boot failure
          # arrives, before any scenario exists to fail).
          STATUS=$(jq -r '.status // "final"' "$VERDICT_FILE" 2>/dev/null || echo final)
          FAILED_COUNT=$(jq '[.scenarios[]? | select(.result == "fail")] | length' "$VERDICT_FILE" 2>/dev/null || echo 0)
          if [ "$STATUS" != "final" ] && [ "${FAILED_COUNT:-0}" -eq 0 ] && [ "$VERDICT" != "FAIL" ]; then
            echo "Scout did not finish (status=${STATUS}) and recorded no failing scenario; nothing to post. See the job summary and artifact."
            exit 0
          fi
          ARTIFACT_PR=$(jq -r '.prNumber // empty' "$VERDICT_FILE")
          if [ "$ARTIFACT_PR" != "$PR_NUMBER" ]; then
            echo "verdict.json prNumber (${ARTIFACT_PR:-none}) differs from the trusted target #${PR_NUMBER}; the trusted target wins."
          fi
          EXISTING=$(gh api --paginate "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments?per_page=100" \
            --jq '.[] | select((.body // "") | startswith("<!-- qa-scout -->")) | .id' \
            | jq -rs 'first // empty')
          if [ "$VERDICT" = "PASS" ]; then
            if [ -z "$EXISTING" ]; then
              echo "Verdict PASS for PR #${PR_NUMBER}; staying quiet (report is in the job summary and artifact)."
              exit 0
            fi
            # A stale FAIL/INVESTIGATE comment misleads forever if a later
            # PASS never speaks: refresh the existing comment, still without
            # creating one.
            {
              echo "<!-- qa-scout -->"
              echo "> [!NOTE]"
              echo "> A newer QA Scout run passed; the earlier finding on this PR no longer stands."
              echo ""
              echo "[QA Scout run](${RUN_URL}) · verdict: PASS · re-run: Actions → CI E2E Main → Run workflow with pr_number=${PR_NUMBER}"
            } > comment.md
            gh api --method PATCH "repos/${GITHUB_REPOSITORY}/issues/comments/${EXISTING}" -F body=@comment.md
            exit 0
          fi
          {
            echo "<!-- qa-scout -->"
            if [ "$STATUS" != "final" ]; then
              echo "> [!WARNING]"
              echo "> The QA Scout stopped before finishing; what follows is partial."
              echo ""
            fi
            # Drops a multi-byte character split by the byte cap so the payload
            # stays valid UTF-8 (iconv -c exits 1 on exactly this case, so
            # python does the lenient decode).
            head -c 60000 "$REPORT_FILE" | python3 -c "import sys; sys.stdout.write(sys.stdin.buffer.read().decode('utf-8', 'ignore'))"
            echo ""
            echo ""
            echo "[QA Scout run](${RUN_URL}) · verdict: ${VERDICT} · re-run: Actions → CI E2E Main → Run workflow with pr_number=${PR_NUMBER}"
          } > comment.md
          # Belt over the scrub step's suspenders: a token shape surviving to
          # here means the scrub was bypassed, so fail loudly instead of
          # publishing.
          if grep -q 'sk-ant-' comment.md; then
            echo "Refusing to post: comment contains an Anthropic token shape."
            exit 1
          fi
          if [ -n "$EXISTING" ]; then
            gh api --method PATCH "repos/${GITHUB_REPOSITORY}/issues/comments/${EXISTING}" -F body=@comment.md
          else
            gh api --method POST "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" -F body=@comment.md
          fi