chore(test): start the subprocess-stub kill-criterion experiment (#1823)

[thymikee]

Summary

#1823 is the live tracker for the subprocess-stub Vitest project and states its own kill criterion: "if the three members run un-serialized in the default pool for 20 consecutive CI runs with no timeout-shaped failure, delete the project." Nobody had started that experiment yet — the project still existed with all its serialization machinery. This PR runs exactly the steps the issue describes: drops SUBPROCESS_STUB_TESTS from unit-core's exclude and deletes the subprocess-stub project block, so its members run in the default forks pool.

Membership note: the issue's table lists 3 members, but scripts/fuzz/corpus-replay.test.ts already left for its own fuzz-worker project in #1994 (merged earlier today, for an unrelated coverage-lane worker-death bug — see #1824). So only 2 files are actually un-serialized by this change:

• src/__tests__/client-metro.test.ts — real Metro dev server + child spawn per case
• scripts/fuzz/harness.test.ts — the SUT is the subprocess watchdog, one target hangs on purpose (test: nightly fuzz lane — parser inputs must fail as typed AppErrors, never hang #1414)

fuzz-worker (holding corpus-replay.test.ts) is untouched — its serialization is for a different, already-decided reason.

What's the same, what changed

• Mutation lane: SUBPROCESS_STUB_TESTS still feeds SERIALIZED_TESTS, so both files stay excluded from the Stryker mutation scope regardless of this experiment's outcome — thousands of mutant reruns times a real per-case spawn is timeout noise independent of whether Vitest itself serializes the file.
• Coverage lane: unaffected — test:coverage:ci already only skips fuzz-worker, not subprocess-stub.
• package.json / docs/agents/*.md / assorted comments: updated the handful of places that named the subprocess-stub project directly, so nothing left in the repo asserts a project that no longer exists. scripts/gate/check.ts needed no change — it derives vitestProjects from vitest.config.ts by regex, so the gate manifest picked up the deletion automatically (verified below).

Revert plan

If a timeout-shaped failure shows up in either file before 20 consecutive CI runs pass clean, restore the subprocess-stub project block and put SUBPROCESS_STUB_TESTS back in unit-core's exclude — that's the entirety of the revert, symmetric with this diff.

Validation

• pnpm exec vitest run --project unit-core src/__tests__/client-metro.test.ts scripts/fuzz/harness.test.ts — both files pass running alongside the rest of unit-core (previously they only ran alone in subprocess-stub).
• pnpm exec vitest run --project unit-core --project fuzz-worker — full suite: 1035 files, 2 pre-existing failures unrelated to this change (app-log-session-resource.test.ts, durable-capture-resource-adoption.test.ts) reproduced identically against origin/main before this diff — not introduced here.
• pnpm check:gate-manifest — 48 checks wired across 28 lanes (down from the prior run's subprocess-stub-inclusive count), confirming the manifest re-derives cleanly with no manual sync needed.
• pnpm check:gate-manifest:test, pnpm check:agent-guidance, pnpm check:layering, pnpm typecheck, pnpm lint, pnpm format:check — all green.
• docs/agents/testing.md stays at 9,587 of its 10,000-byte gate ceiling (agent-guidance-contract.test.ts).

Refs #1823.

---

Generated by Claude Code

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	2.41 MB	2.41 MB	0 B
JS gzip	805.6 kB	805.6 kB	0 B
npm tarball	930.5 kB	930.5 kB	-17 B
npm unpacked	3.23 MB	3.23 MB	-26 B

npm unpacked components

Component	Base	Current	Diff
JS / dist source	2.56 MB	2.56 MB	0 B
Apple runner source/project	570.6 kB	570.6 kB	0 B
macOS helper source	54.5 kB	54.5 kB	0 B
Android helper artifacts	0 B	0 B	0 B
Other package files	44.7 kB	44.7 kB	-26 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.4 ms	29.1 ms	+1.7 ms
CLI --help	81.4 ms	80.0 ms	-1.4 ms

Top changed chunks: no changes in the largest emitted chunks.

Top changed packed files

Packed file	Base	Current	Diff
package.json	18.5 kB	18.4 kB	-26 B

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-08-24 18:08 UTC

[thymikee]

Clean at ca14f2ad: this implements #1823’s prescribed experiment exactly. The two remaining real-subprocess tests now run in unit-core, corpus-replay correctly remains in the separately justified fuzz-worker project, and mutation exclusion remains independent. Code is ready for human review; the remaining iOS check is still running.