Lab 8 - Healing

The shop's layout drifts under the tests. A healing listener repairs broken locators while the suite runs, and records every repair as a proposal. Your job is the part no listener can do: decide, heal by heal, whether to trust it.

Module	8 - Self-Healing Tests
Time	12 minutes
Shop preset	drift_and_bug, applied in step 1 and reset in step 6
You need	Lab 0 done. A healing model in .env if you have one (SETUP.md); without one, step 3 shows the other path
You start from	Your repository after the earlier labs, or main

Steps

1. Let the layout drift:

   uv run --no-sync python -m shop preset drift_and_bug

2. Run the suite as it is, without healing, and see what breaks:

   Local shop	Shared instance
   uv run robotcode robot --exclude broken	uv run robotcode -p shared robot --exclude broken

   uv run robotcode results show --failed lists the failures. Their messages are mostly about elements that cannot be found.

3. Run it again with healing.

   • With a healing model, add the heal profile. It attaches the listener to the unchanged suite:

     Local shop	Shared instance
     uv run robotcode -p heal robot --exclude broken	uv run robotcode -p shared -p heal robot --exclude broken

     The report is results/heal/heal_report.html. Open it in your browser. Every heal is a proposal: the old locator, the new one, and whether the step then passed. Your test files are not changed.

   • Without a healing model, use the recorded healing report of the same preset. It lists the same information.

4. Triage every heal. For each one, decide, and write down why:

   • accept: the new locator finds the element the test meant, and the test is right to pass;
   • reject: it finds a different element, or makes a test pass that should not;
   • investigate: you cannot tell from the report.

   Ask of each: Is it the element the step meant? Is the new locator one that will drift again, like a new class name, or does it use the stable contract? docs/conventions.md says what that is.

5. Look at what stayed red. Some tests still fail with healing on. For each: did the heal work, and does the test now fail for another reason? The profile never heals an assertion, so an expected value that does not match stays red. Is that a broken locator, or a defect of the shop?

6. Put the shop back:

   uv run --no-sync python -m shop reset

Stretch

The other way to heal: repair the test itself, with your agent. Apply drift_and_bug again, then:

The test "WEB-006_AC-7 Successful Order" fails under the current shop layout. Use the RobotCode debugger and the robotframework MCP server to find which locator drifted, and repair its keyword onto the stable contract of docs/conventions.md, so that it works in every layout. Show me the change before you make it.

Run the test under drift_and_bug and under clean: it should pass in both. Reset the shop when you are done.

Compare with the reference

When you are done, compare your result with the reference: what the rehearsal produced, why it is a good result, and the answers the debrief covers. Open it after the lab: it gives the answers away.

If your agent fails

Steps 1 to 6 need no agent. For the stretch goal, follow the recorded walkthrough of this lab.