Skip to content
Kovar

Self-Healing

When you pass --heal, Kovar runs the generated test immediately after recording. If the test fails, it sends the failure output back to the AI, which rewrites the test code to fix the issue. This loop repeats up to --heal-attempts times (default: 3).

Usage

Section titled “Usage”
Terminal window
kovar record https://your-app.com --heal

Allow up to 5 healing attempts:

Terminal window
kovar record https://your-app.com --heal --heal-attempts 5

What It Fixes

Section titled “What It Fixes”

Typical fixes the self-healing loop catches:

  • Incorrect or flaky locators (element not found, wrong role name)
  • Missing await or missing waits for navigation/network idle
  • Assertion mismatches (wrong expected text or URL pattern)

How It Works

Section titled “How It Works”
  1. After recording, Kovar generates test files as usual.
  2. It runs the generated test with npx playwright test.
  3. If the test fails, the failure output (error message, stack trace) is sent back to the AI.
  4. The AI rewrites the failing code to fix the issue.
  5. Steps 2-4 repeat until the test passes or the max attempts are reached.

Cost

Section titled “Cost”

Each healing attempt makes an additional LLM call, so costs scale with the number of retries. A typical session with 1-2 healing attempts costs ~$0.10-0.30 total (including the initial generation).

When Healing Fails

Section titled “When Healing Fails”

If the test still fails after all attempts, Kovar prints the last error and exits. You can:

  • Review the error output and fix manually.
  • Run --heal again with a higher --heal-attempts value.
  • Check Codebase Awareness to improve locator confidence.
Section titled “Related”