🥁 test: Stabilize Host Edit Responsiveness Assertions

[lia-by-librechat]

Summary

API shard 4/4 intermittently fails an existing skill-edit assertion across unrelated PRs. Introduced in #16677, the assertion treats a worker reply arriving before setTimeout(0) as timer starvation, even when the edit is correctly rejected without writing.

Separate atomicity from responsiveness: skill handler tests keep real amplification rejection and no-write assertions. Worker lifecycle tests hold a reply pending and check timer progress with both cold and reused workers. Existing real-amplification timer coverage remains intact. No production code changes.

Related to #16707, #16719, #16715.

How it works

handler tests: real amplification → budget error → no writes
worker tests: held reply → timer fires while unsettled → abort and close

Type of change

• Tests / tooling / CI

Testing

Environment: Node.js 24.16.0, repository npm/Jest configuration.

Automated checks at c935cf5b7530619266cb35f6da01f015e0ee3b1e:

• Handler and worker suites: 376 tests passed with coverage scoped to handlers.ts, processing.ts and matching.ts.
• Timer/atomicity regression cases: 4 passed in each of 5 successive runs.
• npm exec --workspace packages/api -- tsc --noEmit: passed.
• Touched-file ESLint, Prettier, import ordering and npm run static-checks -- --against origin/dev: passed.
• Data-provider, data-schemas and API builds: passed.

The default whole-workspace coverage collection hit the local command limit after both suites passed. The scoped coverage run completed successfully. Full suites and Lighthouse were not run locally; this changes tests only.

Investigation: Original handler suite passed locally. A disposable Node inspector set a conditional breakpoint before clearTimeout(timer); 200 repeated original cases passed without hitting it. The CI failure was not reproduced locally.

Screenshots / recordings

No user-facing change.

Risk / compatibility

Test-only. Worker budgets, matching, persistence and runtime behavior are unchanged.

[lia-by-librechat]

Head: c935cf5

Separates skill-edit atomicity from timer ordering. Adds cold/reused-worker timer checks with a deliberately pending reply. Existing real-amplification timer coverage is unchanged. No production changes.

[lia-by-librechat]

Verified head: c935cf5

• 376 focused tests passed with edit-code coverage.
• 4 timer/atomicity cases passed in each of 5 successive runs.
• API typecheck, touched-file lint, formatting, import order and static checks passed.
• Backend package builds passed.

Default whole-workspace coverage collection exceeded the local command limit; scoped coverage completed. Full suites and Lighthouse were not run locally. CI is running.

[lia-by-librechat]

Reviewed head: c935cf5

Independent review complete: no findings. Confirmed amplification rejection, no-write assertions and existing real-amplification timer coverage remain intact.

CI: all four @librechat/api shards and TypeScript checks passed. Remaining checks are running with no failures reported.

Local evidence: 376 focused tests passed with scoped coverage; four regression cases passed across five runs. API typecheck and touched-file static checks passed. Independent reviewer did not rerun Jest.