chore(e2e-tests): force wait for initialization before writing input

[addaleax]

In the past, we've had a number of flaky tests that followed a pattern similar to this:

1. Start a test shell
2. Run a command using executeLine()
3. Make assertions about the command's output

The internal logic of executeLine() determines that it has seen the end of the command's output by looking for the subsequent prompt (similar to how a human being in front of the shell would determine that the command has finished).

However, in the above scheme, there is a race condition that can develop as follows:

1. The test shell process starts and prints part of its initial output
2. The executeLine() command remembers the current output position and sends the input to the process's stdin stream
3. The test shell prints its first prompt
4. The test reads the output from the shell and sees the prompt, assumes that the comamnd has finished
5. The test shell finishes evaluating the input and writes the corresponding output + a new prompt
6. The test ignores the remaining output and instead only treats the text up to the first prompt as the command output
7. An assertion made about the output fails, as the output is not coming from the command it was supposed to come from

In order to avoid the likelihood of introducing such race conditions, this commit adds a guardrail to the input writing methods that ensures that the shell has reached some form of known state that has been awaited, such as seeing the first prompt or other specific output, before it allows writing to the process's input.

[Copilot]

Pull request overview

This PR addresses race conditions in e2e tests by ensuring the test shell completes initialization before accepting input commands. The core issue was that tests could write input before the shell finished starting up, causing the test framework to misinterpret initial prompts as command completion signals.

Key changes:

• Added a guardrail mechanism (_initializationKnownToBeFinished) that prevents writing input until the shell reaches a known stable state
• Introduced requireFinishedInitialization option (defaults to true) for input methods to control this behavior
• Refactored two test cases to explicitly bypass initialization checks where tests intentionally write input before waiting for the shell to start

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
packages/e2e-tests/test/test-shell.ts	Implements initialization tracking flag and adds requireFinishedInitialization parameter to input methods; introduces type definitions for options objects
packages/e2e-tests/test/e2e.spec.ts	Updates two tests to explicitly disable initialization requirement and replaces manual exit/assertion patterns with waitForCleanOutput()
packages/e2e-tests/test/e2e-direct.spec.ts	Adds explicit waitForPrompt() call before executing commands and waits for successful exit to ensure proper synchronization

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The error message should be more actionable. Consider: 'Cannot write input before shell initialization completes. Call waitForPrompt(), waitForLine(), or waitForAnyExit() first, or set requireFinishedInitialization to false.'

Suggested change

	throw new Error('Wait for shell to be initialized before writing input');
	throw new Error(
	"Cannot write input before shell initialization completes. Call waitForPrompt(), waitForLine(), or waitForAnyExit() first, or set requireFinishedInitialization to false."
	);

[addaleax]

Any strong feelings? I imagine anybody getting this error would look up the logic where it's being thrown (i.e. the code), so I'd personally am okay with leaning towards brevity, but I also don't care too deeply

[gagik]

Yeah don't mind either way either.