---

name: fix-tests description: Diagnose and fix failing CI jobs. Can target a specific job (e.g., /fix-tests electron-e2e) or diagnose all failures when called without arguments.

Fix Failing Tests

First: Get PR info using the agentTool wrapper:

./scripts/agents/tooling/agentTool.ts getPrInfo --fields number,headRefName

Extract number as PR_NUMBER for use in subsequent commands.

This skill diagnoses and fixes failing CI jobs. It can target a specific job when called with an argument (e.g., /fix-tests electron-e2e) or diagnose all failures when called without arguments.

Job Names Reference

Job Name	Workflow	Approx Duration
build	build.yml	~15 min
web-e2e	web-e2e.yml	~10 min
electron-e2e	electron-e2e.yml	~10 min
android-instrumented	android-instrumented.yml	~10-15 min
android-maestro-release	android-maestro-release.yml	~20 min
ios-maestro-release	ios-maestro-release.yml	~30 min
website-e2e	website-e2e.yml	~5 min

1. Identify Failing Jobs

Get the failing workflow run for the current commit:

COMMIT=$(git rev-parse HEAD)
./scripts/agents/tooling/agentTool.ts getCiStatus --commit "$COMMIT"

This returns JSON with run_id, status, conclusion, and jobs array with each job's {name, status, conclusion}.

If a specific job name was provided (e.g., /fix-tests electron-e2e), focus on that job. Otherwise, identify all failing jobs from the jobs array where conclusion == "failure".

2. Download Failed Job Logs

Download only the logs for failed jobs (not all artifacts):

# View failed job logs directly (most token-efficient)
gh run view <run-id> --log-failed

# Or download specific artifact for detailed analysis
./scripts/agents/tooling/agentTool.ts downloadArtifact --run-id <run-id> --artifact <artifact-name> --dest /tmp/<folder>

Artifact Names by Job

Job	Artifact Name	Contents
build	coverage-reports	Coverage JSON files
web-e2e	playwright-report	HTML report, screenshots, traces
electron-e2e	(logs only)	Electron test output
android-maestro-release	android-maestro-debug	Screenshots, logcat, UI hierarchy
ios-maestro-release	ios-maestro-debug	Screenshots, logs

3. Categorize the Failure

Analyze the logs to determine the failure type:

Lint/Type Errors (build job)

Symptoms: ESLint errors, TypeScript errors in build job logs

# Look for lint errors
grep -i "error\|warning" /tmp/logs.txt | head -20

# Common patterns:
# - "'foo' is defined but never used"
# - "Type 'X' is not assignable to type 'Y'"
# - "Unexpected any. Specify a different type"

Fix: Address the lint/type error in the source file. Run locally to verify:

pnpm lint
pnpm exec tsc -b

Test Failures

Symptoms: Test assertion failures, timeout errors

# For unit tests (build job)
grep -i "fail\|error\|timeout" /tmp/logs.txt

Fix: Update test assertions or fix the code logic. Run locally:

# Unit tests
pnpm --filter @tearleads/<package> test

# E2E tests
pnpm --filter @tearleads/client test:e2e

Coverage Drop (build job)

Symptoms: "Coverage threshold not met" errors

# Look for coverage errors
grep -i "coverage\|threshold" /tmp/logs.txt

Fix: Add tests to restore coverage. Check which files dropped:

# Run coverage locally
pnpm --filter @tearleads/<package> test:coverage

# Check the coverage report
cat packages/<package>/coverage/coverage-summary.json

Focus on:

• New code that lacks tests
• Branches not covered
• Functions not called

Flaky Tests

Symptoms: Test passes locally but fails intermittently in CI

Common causes:

• Race conditions in async code
• Timing-dependent assertions
• Shared state between tests

Fix options:

1. Add explicit waits/retries
2. Isolate test state
3. Use deterministic mocks
4. Mark as .skip temporarily and create an issue

Infrastructure/Timeout Failures

Symptoms: Network errors, service unavailable, timeout without test output

Fix: Usually transient - rerun CI:

./scripts/agents/tooling/agentTool.ts rerunWorkflow --run-id <run-id>

If persistent, check:

• CI resource limits
• External service dependencies
• Database connection issues

Android Instrumented D8 OOM (known recurring pattern)

Symptoms (from failed logs):

• :capacitor-preferences:mergeExtDexDebugAndroidTest FAILED
• ERROR: D8: java.lang.OutOfMemoryError: Java heap space
• DexArchiveMergerException during connectedAndroidTest

Interpretation: This is typically a CI memory-pressure failure during dex merging, not an app-logic regression.

Response order:

1. Rerun the failed android-instrumented workflow run first.
2. If downstream CI Gate is red only because of that failed run, rerun CI Gate after the rerun passes.
3. If the same OOM reproduces repeatedly (3x), treat as infra/config debt and escalate with an issue instead of guessing at product-code changes.

4. Job-Specific Fixes

build Job

The build job runs lint, type-check, build, and tests with coverage.

# Run the full build pipeline locally
pnpm lint && pnpm exec tsc -b && pnpm build && pnpm test:coverage

Common issues:

• Lint errors: Fix in source, run pnpm lint
• Type errors: Fix types, run pnpm exec tsc -b
• Build errors: Check imports, run pnpm build
• Coverage drop: Add tests, run pnpm --filter @tearleads/<pkg> test:coverage

web-e2e / electron-e2e Jobs

# Download Playwright report
./scripts/agents/tooling/agentTool.ts downloadArtifact --run-id <run-id> --artifact playwright-report --dest /tmp/playwright
open /tmp/playwright/index.html

# Run locally
pnpm --filter @tearleads/client test:e2e

Common issues:

• Selector changed: Update the selector in the test
• Timing issue: Add await page.waitFor* assertions
• API response changed: Update expected values

android-maestro-release / ios-maestro-release Jobs

# Download debug artifacts
./scripts/agents/tooling/agentTool.ts downloadArtifact --run-id <run-id> --artifact android-maestro-debug --dest /tmp/maestro

# View screenshots
open /tmp/maestro/*.png

# Check logcat (Android)
grep -i "console\|error\|capacitor" /tmp/maestro/logcat.txt

# View UI hierarchy
cat /tmp/maestro/ui-hierarchy.xml

Common issues:

• Element not found: Update selector or add wait
• WebView NAF="true": Add androidWebViewHierarchy: devtools to test
• Timing issue: Use waitForAnimationToEnd or increase retryTapIfNoChange

5. Apply Fix and Verify

1. Make the fix in the source code

2. Verify locally (run the relevant test command)

3. Commit and push:

   git add -A
   git commit -S -m "fix(tests): <description of fix>" >/dev/null
   git push >/dev/null
   

4. Report which job was fixed and what changed

6. Common Issues Reference

Maestro: WebView Not Accessible (NAF="true")

If Maestro can't find elements in a WebView:

appId: com.tearleads.app
androidWebViewHierarchy: devtools
- launchApp:
    clearState: true
- assertVisible: "My Text"

Android: JavaScript Errors / Blank Screen

If the app shows a blank white screen after launch, check logcat for JavaScript errors:

# Check for JS errors in Maestro debug output
grep -i "Capacitor/Console\|TypeError\|Error" /tmp/maestro/logcat.txt

# Common error: "Object.hasOwn is not a function"
# This occurs on WebView < Chrome 93 (system images ship with WebView 91)

Root cause: Android system images (even API 35) ship with WebView 91 by default. WebView 91 doesn't support ES2022 features like Object.hasOwn.

Fix: Add polyfills for missing features in packages/client/src/main.tsx:

// Before any imports
if (!Object.hasOwn) {
  Object.defineProperty(Object, 'hasOwn', {
    value: function (obj: object, prop: PropertyKey) {
      return Object.prototype.hasOwnProperty.call(obj, prop);
    },
    configurable: true,
    writable: true
  });
}

Check WebView version on emulator:

adb shell "dumpsys webviewupdate" | grep "Current WebView"

Maestro: Use evalScript for Complex Interactions

- evalScript: document.querySelector('[data-testid="my-button"]').click()

Playwright: Flaky Selector

// Instead of:
await page.click('.dynamic-class');

// Use:
await page.getByTestId('button-submit').click();
await page.getByRole('button', { name: 'Submit' }).click();

Coverage: Adding Tests for New Code

1. Identify uncovered lines in coverage report
2. Write tests that exercise those code paths
3. Focus on branch coverage (if/else, switch cases)
4. Run pnpm --filter @tearleads/<pkg> test:coverage to verify

Token Efficiency

• Use --log-failed instead of downloading all artifacts when possible
• Use --jq filtering to minimize JSON response size
• Only download artifacts for the specific failing job
• Suppress git output: git commit ... >/dev/null && git push >/dev/null