[brentyates-swx]

Refs #10, #26

Found by CI, not by reading

#10's fix scaled the config's expect.timeout and APP_READY_TIMEOUT. CI then showed that was not the whole set. On PR #27 — a branch touching only packages/opencode/src/plugin/ — four e2e specs failed, and the log carries both halves of the story:

Timeout: 30000ms                                          <- scaled, working
Timeout 10000ms exceeded while waiting on the predicate   <- not scaled

The first is the config's 10 s expect bound × 3 under CI, so the earlier change is doing its job. The second is a per-call bound written inline that nothing scaled — review-line-comment.spec.ts:60's .toPass({ timeout: 10_000 }) — and that spec was among the four failures.

The sweep

file	bound
review-line-comment.spec.ts	toPass 10 s
session-timeline-collapse-state.spec.ts	three waits, 10 s each
scroll-interaction.spec.ts	20 s
subagent-child-navigation.spec.ts	15 s
session-tab-switch-probe.ts	30 s
reproduction/timeline-suspense/playwright.config.ts	30 s test, 10 s expect

There are now no unscaled timeouts left under packages/app/e2e — verified by grep, which is how these were found in the first place.

What this does not fix

One of those four failures was a toBeVisible that timed out at the already-scaled 30 s. Thirty seconds is not a marginally slow render, so that one is either a genuine defect or a much deeper stall, and scaling will not touch it. It needs its own investigation against a failing run's Playwright trace, which is what #26 is for.

Saying so explicitly because the easy mistake here is to scale everything, watch the failure rate drop, and assume the remainder is the same problem.

🤖 Generated with Claude Code

https://claude.ai/code/session_017iG7xA9wRvNyFYZSXm8inU

[brentyates-swx]

The lint gate caught my own bug in this PR

Worth recording, because it is the first time a control added in this triage pass has caught a defect that nothing else would have.

CI failed on unit → Lint for correctness bugs, which was surprising on a branch that touches only packages/app/e2e. The cause was mine: the script that added the deadline import to each file inserted it before the first line not starting with import . In scroll-interaction.spec.ts the second line is import { — a multi-line import — so the next line is an indented member, and the new import landed inside the braces:

import {
import { deadline } from "../../utils/deadline"
  defineVisualRegions,

A syntax error, in a file I had just described as verified.

Nothing else would have caught it

• tsc --noEmit on packages/app did not flag it.
• playwright test --list on that exact file reports "Total: 0 tests in 0 files" — the performance specs are not in the default project that test:e2e:local runs, so CI's e2e job never parses this file at all.

My earlier verification listed two regression specs, which do run, and I generalised from them to a file that does not. That is the mistake, and it is the same shape as several others this session: verifying the thing that was convenient to verify rather than the thing that was changed.

The other four files were inserted correctly — checked each rather than assuming the failure was isolated.

On the gate itself

It now has two catches: the duplicate sessionStatus key that motivated it, and this. Both were in files that the existing checks did not cover — one a rule violation, one a parse failure. The typeAware: false decision matters here too: a parse error is found in 1.3 seconds across 3,364 files.

One incidental finding, not fixed here: the performance e2e specs are not run by CI. They can therefore rot without anyone noticing, which is worth a look separately from #26.