[waleedlatif1]

Summary

• Keep checked trace content at every depth when the projection budget runs out. Stored traces have one workflow-execution root, so retention that only worked per top-level span withheld everything. This changes the rule pinned by the old "withholds the whole top-level span" test. Each retained span's own content still passes the whole-trace check before its children are walked, and any failure other than the budget still withholds the whole trace
• Budget warnings now count spans across the tree, so a rooted trace no longer reports retainedSpanCount: 0
• Release-review tidying: utf8ByteLength instead of a throwaway encode, redundant mounted guards in the speech hook and page-access button, ExtensionSessionExpired at its three call sites, shared capture copy, a styled page-access error, and a guard that the stale-transcript e2e sends before the socket closes

Type of Change

• Bug fix

Testing

• trace-secret-projection regression test fails on the previous code and passes now, with a rooted fixture and a non-budget failure case
• lib/logs/execution, lib/sim-links and process-contents Vitest: 451 passing; extension type-check; biome. Full audits and e2e run in CI

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing (new tests pass the test-audit authoring gate)
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Nov 5, 2026 2:07am UTC

[cubic-dev-ai]

No issues found across 14 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Turn on auto-fix | Re-trigger cubic}

[greptile-apps]

Confidence Score: 4/5

[Medium impact] The changes reach trace viewing and several browser-extension screens.

The PR appears safe to merge, with a non-blocking timing issue in the voice test.

Findings

1.  Slow restarts fail the test ▶

Summary

Keeps checked trace content at every depth when the content budget runs out.

• Trace projection keeps checked content at every depth until the budget runs out.
• Extension pages share one sign-in view when a session expires.
• Page-capture messages use shared wording across the extension.
• Page-access errors now stand out in the access dialog.
• Small helper and test cleanups leave their workflows unchanged.

No runtime tests were run during this review.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Visit span] --> B[Sanitize and check own content]
  B -->|Pass| C[Keep own content and visit children]
  B -->|Content budget exceeded| D[Keep only remaining structure]
  C --> A
  B -->|Other failure| E[Keep only structure for the whole trace]

Loading

_{Reviews (1) · Last reviewed commit: "fix(logs): keep checked trace content at..." · Reviewed by Greptile}

[greptile-apps]

Slow restarts fail the test

If restarting voice input takes more than two seconds, this assertion fails even when voice input works correctly. stopStreaming closes the old socket after two seconds, but the test waits for several browser actions and a new connection before checking sessions[0].closed.

Pause the browser clock during the restart, then advance it after sending the stale transcript. This would test the guard without making CI speed part of the result.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[waleedlatif1]

Folded into #9692 (commit bddf866, rebased past #9696 with the stale-transcript guard kept on the renumbered session). Closing to keep the release fix in one PR.