resolve-checks（解决检查）

Resolve Checks

Systematically resolve all failing CI checks and address PR review feedback by running tests locally first, fixing issues, incorporating valid feedback, and verifying CI passes.

When to Use

• When CI checks are failing on your PR
• When you have unresolved PR review comments
• Before attempting to merge a PR
• When you want to proactively verify all checks pass
• After making changes and before pushing
• After receiving code review feedback

Core Principle

Run tests locally first, don't wait for CI. You have access to the full test suite locally. Catching failures locally is faster than waiting for CI round-trips.

Reference Documentation

For detailed information on test types, setup files, utilities, and common failure patterns, see test-reference.md.

Process

1. Identify the PR

# Get current branch
git branch --show-current

Use GitHub MCP to find the PR:

mcp__github__list_pull_requests with state: "open" and head: "<branch-name>"

2. Run Local Test Suite

Run all tests locally before checking CI status:

cd platform/flowglad-next

# Step 1: Type checking and linting (catches most issues)
bun run check

# Step 2: Backend tests (unit + db combined)
bun run test:backend

# Step 3: Frontend tests
bun run test:frontend

# Step 4: RLS tests (run serially)
bun run test:rls

# Step 5: Integration tests (end-to-end with real APIs)
bun run test:integration

# Step 6: Behavior tests (if credentials available)
bun run test:behavior

Run these sequentially. Fix failures at each step before proceeding to the next.

Note: test:backend combines unit and db tests for convenience. You can also run them separately with test:unit and test:db.

3. Fix Local Failures

When tests fail locally:

1. Read the error output carefully - Understand what's actually failing
2. Reproduce the specific failure - Run the single failing test file: bun test path/to/failing.test.ts
3. Fix the issue - Make the necessary code changes
4. Re-run the specific test - Verify your fix works
5. Re-run the full suite - Ensure no regressions

Common Failure Patterns

See test-reference.md for detailed failure patterns and fixes.

4. Check CI Status

After local tests pass, check CI:

mcp__github__get_pull_request_status with owner, repo, and pull_number

Review each check's status. For failing checks:

1. Get the failure details from GitHub
2. Compare with local results - Did it pass locally?
3. If CI-only failure, investigate environment differences:

- Missing environment variables - Different Node/Bun versions - Timing/race conditions - External service availability

5. Fix CI-Specific Failures

For failures that only occur in CI:

1. Check the CI logs - Look for the actual error message
2. Check environment differences: # Compare local vs CI environment bun --version node --version
3. Check for parallelism issues - Some tests may not be parallel-safe (see RLS tests)
4. Investigate flaky tests - See "Handling Flaky Tests" section below

6. Handling Flaky Tests

CRITICAL: Do not simply re-run CI hoping tests will pass. Flaky tests indicate real problems that must be diagnosed and fixed.

When a test passes locally but fails in CI (or fails intermittently):

Step 1: Identify the Flakiness Pattern

Run the failing test multiple times locally:

# Run 10 times to check for intermittent failures
for i in {1..10}; do bun test path/to/flaky.test.ts && echo "Pass $i" || echo "FAIL $i"; done

Step 2: Diagnose the Root Cause

Common causes of flaky tests:

Step 3: Fix the Test (Not Just Re-run)

Always fix the underlying issue:

// BAD: Non-deterministic - order not guaranteed
const results = await db.query.users.findMany()
expect(results).toEqual([user1, user2])

// GOOD: Sort before comparing
const results = await db.query.users.findMany()
expect(results.sort((a, b) => a.id.localeCompare(b.id))).toEqual([user1, user2].sort((a, b) => a.id.localeCompare(b.id)))

// BAD: Race condition - side effect may not be complete
triggerAsyncOperation()
expect(sideEffect).toBe(true)

// GOOD: Wait for the operation
await triggerAsyncOperation()
expect(sideEffect).toBe(true)

// BAD: Timing-dependent
await sleep(100) // Hope this is enough time
expect(result).toBeDefined()

// GOOD: Poll for condition
await waitFor(() => expect(result).toBeDefined())

Step 4: Verify the Fix

After fixing:

1. Run the test 10+ times locally to confirm it's stable
2. Push and verify it passes in CI
3. If it still fails in CI, there's an environment difference to investigate

7. Push Fixes and Verify

After fixing issues:

# Stage and commit fixes
git add -A
git commit -m "fix: resolve failing checks

- [describe what was fixed]

Co-Authored-By: Claude <noreply@anthropic.com>"

# Push to trigger CI
git push

Then wait for CI to complete and verify all checks pass:

mcp__github__get_pull_request_status with owner, repo, and pull_number

8. Iterate Until Green

Repeat steps 4-7 until all checks pass. Common iteration scenarios:

• New failures appear - Your fix may have caused regressions
• Flaky test still fails - Revisit "Handling Flaky Tests" section, dig deeper into root cause
• CI timeout - Tests may be too slow, need optimization

Remember: The goal is a stable, passing test suite - not a lucky CI run. Every fix should address the root cause.

9. Address PR Review Feedback

After CI checks pass, review and address any PR comments left by reviewers.

Step 1: Fetch PR Comments

Get all review comments on the PR:

mcp__github__get_pull_request_comments with owner, repo, and pull_number

Also get the formal reviews:

mcp__github__get_pull_request_reviews with owner, repo, and pull_number

Step 2: Categorize Each Comment

For each comment, determine if it is:

Step 3: Incorporate Valid Feedback

For each valid comment:

1. Read the specific file and line mentioned in the comment
2. Understand the concern - What issue is the reviewer pointing out?
3. Implement the fix - Make the necessary code changes
4. Reply to the comment - Briefly explain what was changed

mcp__github__add_issue_comment with owner, repo, issue_number, and body

Or reply directly to the review comment thread.

Step 4: Resolve Addressed Comments

After incorporating feedback or providing clarification:

For comments you addressed: The comment should be resolved to indicate the feedback was incorporated. If GitHub MCP supports resolving comments, use that. Otherwise, reply with "Done" or "Fixed in [commit hash]" to signal completion.

For invalid comments: Reply with a clear, respectful explanation of why the current implementation is correct or intentional. Include:

• What the code actually does
• Why it's designed this way
• Any relevant context the reviewer may have missed

Example reply for invalid feedback:

This is intentional - the `userId` here refers to the authenticated user making the request, not the target user. The authorization check happens in the middleware at line 45, so by this point we know the user has permission.

Step 5: Handle Review Requests

If the PR has "Changes Requested" status:

1. Address all blocking comments from that review
2. Re-request review from the reviewer once changes are made: gh pr edit <PR_NUMBER> --add-reviewer <USERNAME>

Common Review Feedback Patterns

10. Final Verification

After addressing all feedback:

1. Re-run local tests - Ensure your fixes didn't break anything
2. Push changes - Trigger CI again
3. Verify CI passes - All checks should be green
4. Verify comments resolved - No unaddressed blocking feedback remains

Quick Reference

For detailed test types, setup files, utilities, failure patterns, and parallel safety rules, see test-reference.md.

Output

Report the final status:

If all checks pass and feedback addressed:

• Confirm all local tests pass
• Confirm all CI checks are green
• List any notable fixes made
• Summarize PR feedback that was incorporated
• Note any comments that were resolved with explanations (not code changes)

If checks still fail:

• List which checks are still failing
• Describe what was attempted
• Explain what remains to be investigated

If feedback remains unresolved:

• List comments that still need discussion
• Explain any disagreements with reviewer feedback
• Suggest next steps (e.g., "discuss with reviewer" or "waiting for clarification")