Add Claude AI code review workflow

.github/claude-review-guide.md

@@ -0,0 +1,82 @@
+# Claude Review Guide for OCS-CI
+
+This guide complements `.claude/CLAUDE.md` — it defines **how to review** code, not how to write it.
+
+## Review Scope
+
+- Only review code added or modified by the PR — never flag existing code
+- If the codebase already uses a pattern, don't flag it in new code that follows the same pattern
+- Read docstrings and comments to understand the author's intent before flagging design decisions
+- Ask "is this a realistic failure scenario?" — not "could this theoretically fail?"
+
+## Severity Calibration
+
+Map findings to severity levels based on real impact, not theoretical risk.
+
+### BLOCKER
+
+These violate critical project rules and must always be flagged:
+
+- `except Exception` or bare `except:` — must use specific exceptions
+- UI locators defined outside `ocs_ci/ocs/ui/views.py` (unless the locator contains a variable)
+- Fixtures using `yield` — must use `request.addfinalizer()` instead
+- Tier marker at class level when `parametrize` marks have different tiers
+- Test classes that create resources without a `teardown()` method
+- Version-specific CSS selectors (`.pf-c-` instead of `[class*='c-']`)
+
+### HIGH
+
+- Missing type hints on new public method parameters or return values
+- Missing docstrings on new public methods
+- Magic numbers (unnamed timeout values like `time.sleep(2)`)
+- Use of `@pytest.mark.usefixtures` instead of passing fixtures as parameters
+- Global variables used for sharing state between tests
+
+### MEDIUM
+
+- Missing `log_step()` in UI test methods (non-UI tests use `logger.info()`)
+- Missing fallback selectors for UI elements known to be flaky
+- Inconsistent return types (method returns different types based on a flag)
+
+### LOW
+
+- Missing exception chaining (`from e`)
+- Minor formatting issues not caught by black/flake8
+
+## File-Type Guidance
+
+### conftest.py
+
+- Focus on fixture teardown: are resources cleaned up via `request.addfinalizer()`?
+- Check for `yield` usage — this is a BLOCKER
+- Verify no `request.node.cls` usage for setting/reading class attributes
+- Ignore docstrings on fixtures (per project convention)
+
+### Page Objects (`ocs_ci/ocs/ui/page_objects/`)
+
+- Verify locators are in `views.py`, not inline
+- Check navigation pattern: destination pages should NOT navigate to themselves
+- Look for lazy imports to prevent circular dependencies
+- Verify selectors are version-agnostic (`[class*='c-']` not `.pf-c-`)
+
+### Test Files (`tests/`)
+
+- Verify `teardown()` exists if the class creates resources
+- Check tier markers are only in `parametrize` marks, not at class level
+- Verify `log_step()` usage in UI tests
+- Check that factories are used for resource creation (automatic cleanup)
+- Verify `create_unique_resource_name()` for naming resources
+
+### views.py (`ocs_ci/ocs/ui/views.py`)
+
+- Verify selectors follow the locator priority order: `data-test` > `id` > `aria-label` > text+ancestor > class > index
+- Check for version-agnostic patterns
+- Verify fallback selector lists for unreliable elements
+
+## What NOT to Flag
+
+- Style issues caught by black or flake8 (formatting, import ordering, line length)
+- Existing codebase patterns used consistently in new code
+- Missing docstrings on fixtures (project convention exempts them)
+- Import ordering or grouping
+- Use of `time.sleep()` when the timeout value IS a named constant

.github/workflows/claude-review.yml

@@ -0,0 +1,53 @@
+name: Claude Code Review
+
+on:
+  pull_request:
+    types: [labeled]
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+
+jobs:
+  review:
+    if: >
+      (github.event_name == 'pull_request' && github.event.label.name == 'claude-review') ||
+      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '@claude')) ||
+      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '@claude'))
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    permissions:
+      contents: read
+      pull-requests: write
+      issues: write
+
+    steps:
+      - name: Run Claude Review
+        uses: <ORG>/claude-review-action@v1
+        with:
+          anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
+          model: claude-sonnet-4-6
+          review-authority: comment-only
+          review-guide-path: .github/claude-review-guide.md
+          max-diff-lines: '4000'
+          max-diff-bytes: '120000'
+          max-files: '40'
+          context-intro: >
+            You are reviewing a PR in ocs-ci — a Python/pytest testing framework
+            for Red Hat OpenShift Container Storage (OCS/ODF). The codebase uses
+            Selenium for UI tests, pytest fixtures for resource management, and
+            has strict conventions for locator placement, exception handling, and
+            test teardown. Read the project CLAUDE.md for full conventions.
+          critical-rules: |
+            1. Never use `except Exception` or bare `except:` — always use specific exceptions
+            2. UI locators MUST be centralized in `ocs_ci/ocs/ui/views.py` (unless the locator contains a variable)
+            3. Fixtures MUST NOT use `yield` — use `request.addfinalizer()` instead
+            4. No tier marker at class level when parametrize marks have different tiers
+            5. Test classes that create resources MUST have a `teardown()` method
+            6. Version-agnostic CSS selectors — use `[class*='c-']` not `.pf-c-`
+          spot-check-hint: 'Files matching CRITICAL RULES patterns (views.py, conftest.py, page objects, test files)'
+          extra-prompt: >
+            This project has a comprehensive CLAUDE.md that you auto-read. The CLAUDE.md
+            contains interactive development instructions (e.g., "wait for approval",
+            "present changes step by step") — IGNORE those for automated review. Focus only
+            on the code quality rules, patterns, and checklists documented there.