This is correct — emplace() returns a reference to the emplaced value, then .clone() is called on it to make a copy for throwFatalException. The original was kj::cp(permanentException.emplace(...)) which is equivalent.

I've now reviewed the entire diff thoroughly. This is a mechanical change across 42 files:

1. Replace kj::cp(exception) with exception.clone() (the bulk of the change)
2. Replace exception.map([](const kj::Exception& e) { return kj::cp(e); }) with exception.map([](const kj::Exception& e) { return e.clone(); })
3. Fix ErrorWorkerInterface constructor to properly kj::mv() instead of implicitly copying (a pre-existing minor bug that now surfaces because copy ctor is hidden)
4. Change AbortSignal::abortException parameter from by-value to const& (since it no longer needs to be copied, and reason.clone() is used inside)
5. Change QueueEvent::setCompletionStatus to use kj::mv() and return const& from getter
6. Update deps (capnp-cpp and workerd-cxx) hashes
7. Remove NOLINT comment in test.h macro (no longer using copy constructor)

All transformations are correct and consistent. The dependency bump, the ErrorWorkerInterface fix, and the abortException signature change are the only non-mechanical parts, and they're all correct. There are no backward compatibility concerns, no security issues, and no logic bugs.

LGTM

github run

Merging this PR will improve performance by 40.18%

⚡ 1 improved benchmark
✅ 71 untouched benchmarks
⏩ 129 skipped benchmarks¹

Performance Changes

_{Comparing maizatskyi/2026-04-03-exception-clone (fe2f924) with main (22c1ca0)²}

The generated output of @cloudflare/workers-types matches the snapshot in types/generated-snapshot 🎉