Skip to content

CI: fix the two flaky macOS checks (e2e prompt target, bridge backlog) - #12

Merged
sheehanmunim merged 3 commits into
mainfrom
e2e-live-prompt-everywhere
Sep 30, 2026
Merged

sheehanmunim merged 3 commits into
mainfrom
e2e-live-prompt-everywhere

Conversation

@sheehanmunim

@sheehanmunim sheehanmunim commented Sep 30, 2026 •

Copy link
Copy Markdown
Member

Two checks made the macOS CI job flaky:

  1. Browser e2e, credentials prompt. After e2e: approve the site prompt in the live DevTools target #10 fixed the approval-prompt check, main's CI (dc4fd0b) failed the credentials check instead: Cannot read properties of undefined (reading 'origin'). Same race: the evaluation landed in a stale DevTools target for the prompt URL, which returned nothing. All prompt reads and clicks now go through evaluateInLive with a readiness guard, so a target that has not run prompt.js is skipped. 14/14 locally, three runs in a row.
  2. Bridge stress test (120 peers). It failed on this PR with "peer 64 of 120 was refused". The bridge socket's listen backlog was 64, and a burst of connections outran accept() on a busy runner. This is a real limit, not only a test one: enough agents starting at once would hit it. The backlog is now SOMAXCONN (kern.ipc.somaxconn, 128 by default), the most macOS will queue anyway. Passes 3/3 locally.

🤖 Generated with Claude Code

After the approval check was fixed, main's CI failed the credentials check
the same way: its evaluation landed in a stale target for the prompt URL and
returned nothing ('reading origin' of undefined). All prompt reads and
clicks now go through evaluateInLive with a readiness guard, so a target
that has not run prompt.js is skipped instead of answered.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Copilot AI balanced review requested due to automatic review settings September 30, 2026 04:18

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

The 120-peer stress check failed in CI with peer 64 refused: the socket's
listen backlog was 64, and a burst of connections outran accept() on a busy
runner. Use SOMAXCONN (kern.ipc.somaxconn, 128 by default), the most macOS
will queue anyway.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@sheehanmunim sheehanmunim changed the title e2e: every prompt interaction uses the live DevTools target CI: fix the two flaky macOS checks (e2e prompt target, bridge backlog) Sep 30, 2026
The Linux job hung for 30 minutes: a stale prompt target accepted the
DevTools socket and never replied, and the helper waited on it forever. Each
evaluation now resolves null on error, close, or two seconds of silence.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@sheehanmunim
sheehanmunim merged commit c6a4d79 into main Sep 30, 2026
4 checks passed
@sheehanmunim
sheehanmunim deleted the e2e-live-prompt-everywhere branch September 30, 2026 04:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants