Only the two places we already identified. This is clean.

Let me now think through the overall review:

1. Correctness: The refactor correctly replaces a manually maintained list (isValuelessArg for TypeHandler<T> and Arguments<T>) with a concept-based approach (ValueLessParameter) that automatically detects all injected types. The Arguments<T> special case is correctly preserved via isArguments<>().

2. Template instantiation ordering: The comment in the PR correctly explains that RequiredArgCount_ specializations in type-wrapper.h will be visible when resource.h templates are instantiated, because type-wrapper.h includes resource.h (so resource.h is parsed first with only the forward declaration), and by the time the resource.h templates are actually instantiated (deferred until after the whole TU is parsed), the specializations will be visible. This is standard C++ two-phase lookup behavior.

3. Tests: The tests are well-written, covering TypeHandler<T>, Arguments<T>, the new InjectConfiguration case (via FakeConfig), and the ValueLessParameter concept directly. The MockTypeWrapper approach is clean and self-contained.

4. Backward compatibility: No behavioral changes to external APIs — this is purely internal template metaprogramming. The .length property on functions will now correctly account for InjectConfiguration parameters, which was previously a bug (they would have been incorrectly counted as required JS arguments).

5. No compat flag needed: This is a bugfix to an internal mechanism. The incorrect .length was an implementation defect, not an intentional behavior.

6. No security concerns: No sandbox or capability changes.

7. No wire format changes: No .capnp schema changes.

The PR is clean. The design is sound, the test coverage is thorough, and the migration is complete.

LGTM

github run