[kittimzhe]

What changed

RuffChecker.fix and RuffFormatter.fix ran ruff with expect=dagger.ReturnType.ANY, awaited only combined_output(), and built a Changeset without ever inspecting the exit code. When ruff itself fails — a broken ruff.toml (unknown option), a usage error, a runtime error — the function returned an empty Changeset, which reads as a successful no-op ("everything is already clean"), and ruff's error text never reached the terminal in dagger call. The same input makes lint() fail loudly (it uses the default expect=SUCCESS), so the module disagreed with itself.

Fix

After combined_output(), read exit_code() and raise RuntimeError when it is not 0 or 1:

• 0 — clean, 1 — violations remain after --fix — are normal outcomes and still return a Changeset exactly as before (an over-strict check would break the "fix what's fixable" flow).
• 2+ — ruff's error code (configuration / usage / runtime) — now fails the function with the exit code in the message.

lint() is untouched.

Verification (Dagger engine, all six runs against the same source dirs)

#	Scenario	Before	After
1	ruff.toml with an unknown option → check fix	empty Changeset (no changes to apply)	✗ FunctionError: ruff check --fix failed with exit code 2
2	same → format fix	empty Changeset	✗ same, ruff format
3	fixable violations (F401 unused import) → check fix	Changeset with the fix	✓ unchanged
4	fixable + unfixable (F401 + F821, ruff exits 1) → check fix	Changeset	✓ still returns the Changeset (no false failure)
5	well-formed source → format fix	Changeset	✓ unchanged
6	broken config → lint	loud failure (TOML parse error ...)	✓ unchanged

ruff check --config ruff.toml passes on both changed files.

Fixes #99.

---

Prepared and engine-verified with an AI coding agent; verification matrix above was run with dagger v0.21.10 on darwin/arm64, not inferred from reading the code.

[danielgafni]

thanks!