GitHub can show an unchecked box in the Run workflow form while a conditional step still runs. Look at the expression the workflow evaluates. GitHub documents two ways to read manual inputs, and only the inputs context preserves a checkbox value as a Boolean.

Find the expression that gates the step

Open the run that behaved unexpectedly and identify the exact job or step. Then open the workflow revision used by that run and look for its if expression. A check against github.event.inputs.run_extra can evaluate the nonempty string false as truthy. This can be especially confusing when the workflow’s form declares run_extra as type boolean.

Use the typed input

Declare run_extra under workflow_dispatch inputs as a Boolean and use inputs.run_extra in the step’s if condition. For example, give the step the condition ${{ inputs.run_extra }}. Keep your ordinary tests outside this condition. The checkbox should only gate the extra work it describes, such as an optional longer test suite.

Run a two-case test

Test the change using a harmless step that writes a diagnostic line rather than deploying anything. Run once with the box unchecked and verify the step is skipped in the run log. Run a second time with the box checked and verify that it executes. Both runs should still finish the ordinary test steps. For a manual trigger, GitHub requires the workflow file to be present on the default branch.

Check other triggers separately

If the same workflow also runs on a push or schedule, define what should happen when there is no manual form. Verify those paths using their own runs. Write down the expected behavior for unchecked, checked and automatic runs so future edits cannot quietly change the condition. The example follows GitHub’s documented input behavior.

Official reference