XC4 fails under Node in composability's clean clone at c21e598

[taras]

main went red at c21e598 (#495) on the composability job only. One Node test failed:

not ok 4 - XC4: this runtime's entrypoint relaunches a worker from another directory
location: '/tmp/xmd-verify-clean-04LRd0/packages/test-support/bdd.ts:1:1897'
expect(received).toBe(expected) // Object.is equality
Expected: 0
Received: 1
duration_ms: 32663.08767

Suite totals: 3463 tests, 3462 pass, 1 fail.

• Test: packages/cli/tests/command.test.ts:63
• Runtime: Node, inside deno task verify:clean's clone
• Run: https://github.com/taras/executable.md/actions/runs/32184873043
• Job: https://github.com/taras/executable.md/actions/runs/32184873043/job/95866085048
• Tracked by the self-closing main is red: CI failure at c21e598 #499

What the test does

It spawns the real CLI from a scratch temp directory and asserts the worker
relaunch exits 0:

runCli(["test", path.resolve("smoke-test/test-agent/README.md")], {
  cwd: elsewhere, inheritEnv: true, timeout: 180_000,
})

The relaunch entry path comes from import.meta.url, so it is meant to stay
valid wherever the CLI started. The worker exited 1 after ~33s.

Open question

Regression or flake — one run cannot tell them apart, and no intermittence
evidence exists yet, so this is deliberately not labelled flake.

composability runs only on a main push and on a ci-main-red-fix pull
request, so no ordinary PR check would have caught this before merge.

Next step

Re-run the composability job on the identical revision. A green re-run makes
this a flake issue with the evidence attached; a second failure makes it a
regression to bisect against #495.