tracer-cloud--opensre/.github/workflows/e2e-openclaw.yml

# OpenClaw end-to-end CI.
#
# Triggers, in order of intent:
#   1. PRs that touch OpenClaw-relevant paths            (auto, primary signal)
#   2. Pushes to main that touch the same paths          (post-merge regression catch)
#   3. Daily schedule at 06:00 UTC                       (catches environment drift —
#                                                        OpenClaw version bumps, MCP SDK
#                                                        breaks, npm install regressions)
#   4. Adding the `ci:openclaw` label to a PR            (manual re-run on matching PRs)
#   5. Workflow_dispatch from the Actions UI             (manual force-run on any ref,
#                                                        no PR needed)
#
# The path filter at the trigger level is the primary gate, so non-OpenClaw PRs
# pay no CI cost. The label is only useful as a re-trigger on PRs that already
# matched paths; for force-running against an unrelated branch, use the manual
# dispatch button in the Actions UI.

name: CI (OpenClaw E2E)

on:
  pull_request:
    branches: [main]
    types: [opened, synchronize, reopened, labeled, unlabeled]
    paths:
      - "integrations/openclaw.py"
      - "tools/OpenClawMCPTool/**"
      - "tests/e2e/openclaw/**"
      - "tests/utils/alert_factory/**"
      - ".github/workflows/e2e-openclaw.yml"
      - "docs/openclaw.mdx"
      - ".tool-versions"
  push:
    branches: [main]
    paths:
      - "integrations/openclaw.py"
      - "tools/OpenClawMCPTool/**"
      - "tests/e2e/openclaw/**"
      - "tests/utils/alert_factory/**"
      - ".github/workflows/e2e-openclaw.yml"
      - "docs/openclaw.mdx"
      - ".tool-versions"
  schedule:
    - cron: "0 6 * * *"
  workflow_dispatch:

permissions:
  contents: read

env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"

concurrency:
  group: ci-openclaw-${{ github.ref }}
  cancel-in-progress: true

jobs:
  # Quality gates (lint / format-check / mypy) are run by the main
  # ``ci.yml`` workflow for every PR. We don't duplicate them here —
  # this workflow's sole job is the OpenClaw e2e suite. If you reach
  # this workflow, your PR has already cleared the quality gates.
  openclaw-test:
    if: github.repository == 'Tracer-Cloud/opensre'
    name: openclaw test
    runs-on: ubuntu-latest
    timeout-minutes: 25

    # No ``env:`` block by design. CI runs the boot + use-case
    # scenarios only — the full-RCA sub-tests call an LLM, cost API
    # credits, and take ~25s each, so we deliberately don't wire any
    # LLM secret in. They auto-skip via
    # ``LLM_CREDENTIAL_SKIP_REASON`` when no key is present in env.
    # Contributors with ANTHROPIC_API_KEY / OPENAI_API_KEY /
    # GEMINI_API_KEY set locally still run them via
    # ``make test-openclaw``. Lang Smith env vars stay scoped to the
    # local-RCA path for the same reason — no value tracing the
    # use-case-only CI runs.

    steps:
      - uses: actions/checkout@v5

      - name: Set up Node (OpenClaw requires >=22.12; version pinned in .tool-versions)
        uses: actions/setup-node@v4
        with:
          node-version-file: ".tool-versions"

      - name: Cache npm downloads
        uses: actions/cache@v4
        with:
          path: ~/.npm
          key: ${{ runner.os }}-openclaw-npm-${{ hashFiles('.tool-versions') }}
          restore-keys: |
            ${{ runner.os }}-openclaw-npm-

      - name: Install OpenClaw CLI
        # ``continue-on-error: true`` so an npm-registry hiccup, a package-name
        # rename on OpenClaw's side, or a network blip does not kill the whole
        # workflow. The "Probe OpenClaw availability" step below catches the
        # failure and downgrades the run to "tests skipped" with a clear
        # explanation rather than crashing.
        continue-on-error: true
        id: install_openclaw
        run: npm install -g openclaw --prefer-offline --no-audit

      - name: Probe OpenClaw availability
        # Whether or not ``npm install`` succeeded, check if ``openclaw`` is
        # actually runnable. Failure modes we surface explicitly:
        #   - CLI not installed (npm step failed silently)
        #   - CLI installed but Node version mismatch (older runner image)
        #   - CLI installed but the bundled JS errored on first run
        # ``openclaw_runnable`` env var feeds the next step's gating.
        run: |
          set +e
          if ! command -v openclaw >/dev/null 2>&1; then
            echo "::warning::openclaw CLI is not on PATH (npm install may have failed); the e2e suite will be skipped this run."
            echo "openclaw_runnable=false" >> "$GITHUB_ENV"
            exit 0
          fi
          version_output="$(openclaw --version 2>&1)"
          if [ $? -ne 0 ]; then
            echo "::warning::openclaw is on PATH but errors when run:"
            echo "$version_output"
            echo "::warning::Skipping e2e suite this run."
            echo "openclaw_runnable=false" >> "$GITHUB_ENV"
            exit 0
          fi
          echo "✓ OpenClaw verified: $version_output"
          echo "openclaw_runnable=true" >> "$GITHUB_ENV"

      - name: Note LLM credential policy
        # CI deliberately does not wire any LLM key, so the full-RCA
        # sub-tests skip every run. This step is informational — it
        # makes the policy obvious in the run log instead of leaving a
        # silent skip that a reviewer has to dig for. Run RCA locally
        # with ``make test-openclaw`` if you have a key configured.
        run: |
          echo "ℹ︎ CI policy: full-RCA sub-tests are skipped (LLM credentials intentionally not wired)."
          echo "  Boot + use-case sub-tests still run and gate the merge."

      - name: Install uv
        uses: astral-sh/setup-uv@v8.1.0
        with:
          enable-cache: true
          cache-dependency-glob: |
            pyproject.toml
            uv.lock

      - name: Set up Python
        uses: actions/setup-python@v6
        with:
          python-version: "3.13"

      - name: Install dependencies
        run: uv sync --frozen --extra dev

      - name: Run OpenClaw E2E suite
        # Skips with a clear message if the OpenClaw probe failed earlier,
        # so the workflow finishes green with a documented "couldn't run"
        # status rather than a confusing red on the install step.
        run: |
          if [ "$openclaw_runnable" != "true" ]; then
            echo "::warning::Skipping pytest invocation — openclaw CLI unavailable in this environment. See earlier warnings."
            exit 0
          fi
          uv run python -m pytest -m e2e -v tests/e2e/openclaw/

      - name: Upload gateway logs on failure
        if: failure()
        uses: actions/upload-artifact@v4
        with:
          name: openclaw-gateway-logs
          path: /tmp/openclaw-e2e-logs/
          if-no-files-found: ignore
          retention-days: 7

      - name: Summary
        # Always runs, regardless of earlier step outcomes. The probe
        # step above downgrades a missing-CLI / wrong-Node environment
        # to "tests skipped" rather than failing the workflow, so this
        # summary is the one place a reviewer sees why fewer tests ran
        # than expected. Real pytest failures still fail the job.
        if: always()
        run: |
          echo "### OpenClaw E2E run summary"
          echo "  OpenClaw runnable: ${openclaw_runnable:-unknown}"
          echo "  Full-RCA sub-tests: skipped by CI policy (no LLM key wired)"
          echo "  Run RCA locally with: make test-openclaw"