kill_task.py isn't run after a failed test run.
As we can see from the screen shot below (from target-dartino-linux-release-arm build 768), task_kill.py isn't called after a failing test run. This means the next build also fails.

We need to modify the recipe that is run. We need to wrap the two taskkill steps and the annotated_steps step in a with api.step.defer_results(): block.
The recipe that is run is at https://chromium.googlesource.com/chromium/tools/build/+/master/scripts/slave/recipes/client.fletch.recipe_autogen.py
Since this is a generated file, with 20 recipes, all of which need to be changed, and the file has already been modified since being generated, and we don't have the script that generated it, I will modify it using a custom one-off filter.