@qestro/self-healing

@qestro/self-healing

AI-powered self-healing selectors for Playwright tests. Auto-fix broken tests when your UI changes.

When your UI ships, selectors break. @qestro/self-healing inspects a failed test run and proposes ranked fixes — new selectors, longer waits, assertion rewrites, API schema migrations — with a confidence score you can auto-apply above a threshold you set.

Zero runtime dependencies. Works with any test framework that can map its result shape into TestResult.

Install

npm install @qestro/self-healing
# Playwright is an optional peer (only needed for the reporter example)
npm install --save-dev playwright

60-second example

import { SelfHealingEngine } from '@qestro/self-healing';

const engine = new SelfHealingEngine();
const result = await engine.analyzeAndHeal(testId, failedTestResult);

if (result.healed) {
  // confidence >= 0.85 by default — apply `result.appliedFix.suggestedValue`
  applyFix(result.appliedFix);
} else {
  // inspect `result.suggestions` for ranked alternatives
  console.log(result.suggestions);
}

What it fixes

Failure type	Healer	Typical fix
selector_changed	SelectorHealer	Swap stale selector for [data-testid] / ARIA / text
timing_issue	TimingHealer	Add waitForSelector, bump timeout, retry on flake
assertion_drift	AssertionHealer	Update expected value, swap .toBe → .toContain
api_schema_change	APIHealer	Patch field paths, error shape, status range checks

Each healer returns multiple ranked suggestions (confidence: 0..1). The engine picks the best one and auto-applies it if its confidence clears the threshold.

Options

new SelfHealingEngine({
  autoApplyThreshold: 0.9,        // default 0.85
  logger: console,                // anything with info/warn/error
  onLowConfidence: async ({ testId, suggestion, confidence }) => {
    // Route below-threshold cases to a review queue, Slack, GitHub PR, etc.
  },
});

Playwright reporter

Drop the self-healing engine into your Playwright config:

// playwright.config.ts
export default {
  reporter: [['@qestro/self-healing/dist/examples/playwright-integration.js']],
};

See examples/playwright-integration.ts for the full reporter — maps Playwright's TestResult into the engine's input shape and forwards low-confidence cases to a notifier.

Public API

import {
  SelfHealingEngine,
  SelectorHealer,
  TimingHealer,
  AssertionHealer,
  APIHealer,
} from '@qestro/self-healing';

import type {
  TestResult,
  AssertionResult,
  HealingResult,
  HealingSuggestion,
  FailureAnalysis,
  FailureType,
  SelfHealingEngineOptions,
  Logger,
  LowConfidenceNotifier,
} from '@qestro/self-healing';

You can also use the healers independently:

import { SelectorHealer } from '@qestro/self-healing';

const suggestions = new SelectorHealer().heal(myFailedTestResult);

TestResult shape

interface TestResult {
  id: string;
  testId: string;
  status: 'passed' | 'failed' | 'pending' | 'skipped';
  startTime: Date;
  endTime: Date;
  duration: number;
  errors: string[];            // stringified error messages
  assertions: AssertionResult[];
  screenshots?: string[];
  logs?: string[];
}

Map your test runner's output into this shape once and you're done.

Confidence scoring

Confidence	Meaning
>= 0.90	Safe to auto-apply in CI without human review
0.80–0.89	Auto-apply with PR comment / changelog entry
0.60–0.79	Send to human review queue (default onLowConfidence)
< 0.60	Surface as informational suggestion only

Tune autoApplyThreshold to match your team's risk tolerance.

Made by Qestro

This package is one of the healers that powers qestro.app — the copilot for testing AI vibe coding.

Qestro is a managed platform that adds team collaboration, CI/CD integration, cross-browser runs, mobile testing, visual regression, and test generation from plain English. Paste a URL, describe what to test in plain English, and get production-ready test cases across browser, mobile, and API.

Try Qestro free at qestro.app.

Contributing

Found a healing strategy that should ship in the OSS core? PRs welcome at github.com/finsavvyai/qestro.

License

MIT. See LICENSE.