test: state the timeout budget for the two workflow suites that execute externally

[jpr5]

The failure

main @ 079f37e failed the test (24) leg of Unit Tests (run 34731065578):

FAIL src/__tests__/publish-pin-workflow.test.ts
 > the PyPI publish path runs no Python bytes it has not pinned
 > EXECUTED against real pip > SUBSTITUTED backend bytes are REFUSED by the repo's own pin
Error: Test timed out in 5000ms.        (file: 21075ms)

Not a correctness failure. The test shells out to real pip — deliberately, that is the guard — and inherited vitest's 5000ms default. It passes on a fast machine and fails on a slow one.

Audit

Every src/__tests__/*workflow*.test.ts and sibling that spawns a process was measured. Three already state a budget (fix-drift-workflow 30s, unreleased-check-workflow 30s, npm-publish-verify-workflow 60s). Two did not:

file	slowest test, measured	new budget
publish-pin-workflow.test.ts	172ms / 183ms warm local pip; 21075ms for the file on the CI runner	60_000
test-drift-workflow.test.ts	3039ms idle 18-core; 5490ms → timed out under concurrent load	30_000

drift-sync-core.test.ts also shells out to real git, but its slowest test is 116ms — 43x of headroom under the default. Left alone. fix-drift.test.ts mocks child_process. adoption-wall.test.ts is out of scope.

RED

test-drift-workflow at the inherited default, with two full suites running concurrently:

× EXECUTED POSITIVE CONTROL: base exits 0 and 2 stay non-fatal 5490ms
 Test Files  1 failed | 1 passed (2)
      Tests  1 failed | 32 passed (33)

publish-pin-workflow's RED is the CI run above — local pip is warm here and never approaches 5s.

GREEN

Both files, 5 consecutive runs:

run 1 exit=0    Tests  33 passed (33)
run 2 exit=0    Tests  33 passed (33)
run 3 exit=0    Tests  33 passed (33)
run 4 exit=0    Tests  33 passed (33)
run 5 exit=0    Tests  33 passed (33)

And once under the exact condition that produced RED — concurrent with two full suites:

CONCURRENT EXIT=0
 Test Files  2 passed (2)
      Tests  33 passed (33)
slowest under load: 4751ms, 4379ms, 3898ms

4751ms against a 5000ms budget is the flake, stated plainly.

The budgets still bite

A raised-so-far-it-can-never-fire timeout is a removed guard. Both were confirmed against an injected hang (sleep in front of the external call, reverted after):

publish-pin  → Test timed out in 60000ms.   Tests  2 failed | 7 passed (9)
test-drift   → Test timed out in 30000ms.   Tests  1 failed | 1 passed | 22 skipped (24)

Deliberately not done

• No global testTimeout — that would stop genuinely hung tests from failing everywhere.
• No stubbing. These are marked EXECUTED against real pip because running the repo's own pin file against the real resolver is the assertion. What they exercise is unchanged.
• npm-publish-verify-workflow.test.ts untouched — already fixed in test(ci): drive the npm-verify retry loop on a virtual clock #436 on a stubbed clock.

Gates (raw exit codes)

gate	exit
prettier --check	0
npx eslint .	0
pnpm typecheck	0
npx vitest run (full)	0 — 191 passed / 1 skipped, 5864 tests
npx commitlint --from origin/main --to HEAD	0

🤖 Generated with Claude Code

https://claude.ai/code/session_01AvkmhXVLqSSEW6FvPQHSu5

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@copilotkit/aimock@437

commit: 0b343d8