Missing newline at end of file

The file has no trailing newline. This will cause ruff format --check (and thus make format-check) to fail in CI.

Context Used: Code style and conventions (source)

Formatting issues — ruff will reject

Two ruff-enforced style violations in this function: a missing space after the comma in the monkeypatch.setattr call and a trailing-whitespace-only blank line inside fake_read_text. Running make format-check will fail on both.

Context Used: Code style and conventions (source)

Broad Path class patch may bleed into unrelated code

monkeypatch.setattr(Path, "exists", lambda self: False) replaces the method on the class itself, so every Path instance created anywhere during the test call-graph (e.g. inside imports, pytest internals, or future changes to _check_memory_health) will be affected. A narrower patch targeting only the specific path object — or patching at the module level with monkeypatch.setattr("app.remote.server.Path", ...) — limits the blast radius and makes intent clearer. The same applies to the read_text patches in the other two tests.

Quick formatting one. Two ruff issues here:

1. I001 — the import block is sorted alphabetically by name; _check_memory_health should be next to _check_disk_health, _imds_get, _imds_token, _lifespan (the underscore-prefixed group), not after investigate_stream.
2. Trailing comma needed (_check_memory_health,) per ruff's format rules.

Both are auto-fixable with make format && ruff check tests/ --fix. Worth running both before pushing, there are 7 ruff errors in this file as-is and CI will reject them.

Reinforcing greptile's "broad Path class patch" concern with a concrete fix. The issue is that remote_server.Path is the same object as pathlib.Path, Python imports are references, not copies, so monkeypatch.setattr(remote_server.Path, "exists", ...) mutates the global Path class for the test's duration. Anything that creates a Path during your test (pytest internals, plugins, transitive imports inside _check_memory_health's call graph) sees the fake.

A cleaner pattern that scopes to just the module under test:

class _FakeMeminfoPath:
    def __init__(self, *_args, **_kwargs): ...
    def exists(self) -> bool: return False

monkeypatch.setattr("app.remote.server.Path", _FakeMeminfoPath)

This replaces the Path name in the app.remote.server namespace only. Anywhere else still gets the real Path. Bonus: the named class lets you name the fake methods clearly per-test (_FakeMissingPath, _FakeIncompletePath, _FakeOsErrorPath) and you don't need lambdas, which sidesteps the ARG005 lint issue entirely.

The lambda self: ... and lambda self, **kwargs: ... patterns are technically correct (you need a self slot when the lambda is bound as a method), but ruff's ARG005 flags 5 unused-argument errors across these tests. Two clean ways to silence them without disabling the rule:

1. Switch to the named-class approach from the comment above, replaces Path wholesale, no lambdas needed.
2. If you'd rather keep the patch shape, prefix unused params with underscore: lambda _self: ... and lambda _self, **_kwargs: .... ruff treats _-prefixed names as intentionally unused.

Also on this line specifically: it's 107 chars, longer than the project's 100-char limit, so ruff format will want to split it across multiple lines anyway. make format will do that automatically.