Thanks for the detailed PR, @sixianli! This addresses a real UX pain point — losing a long typed draft after a silent timeout is frustrating, and making the deadline visible + preserving the draft on expiry is the right fix.

Overall assessment: well-scoped and thoughtfully implemented.

A few things to verify before merge:

Timeout metadata source
The PR adds requested_at, timeout_seconds, and expires_at to pending clarify payloads. Please confirm these fields are populated server-side (in api/clarify.py or equivalent) and not just synthesized client-side — if the client invents the expiry, the countdown could drift from the actual backend timeout.

Draft preservation — edge cases

• If the user has content in the main composer before the clarify prompt expires, the moved draft should be appended (or shown in a conflict state), not silently overwrite what was already there. Please confirm the implementation handles this.
• Draft text should not be preserved for prompts the user explicitly dismissed vs. those that timed out silently — the PR description says "avoids preserving already-submitted clarify text," but it's worth confirming that manual close/cancel also does not trigger the preservation path.

Full test suite
The PR notes that pytest tests/ -v --timeout=60 has not been run yet. Please run the full suite before re-requesting review — the regression coverage looks solid but we want to make sure nothing else is affected by the clarify metadata changes.

Minor nib
Urgent countdown styling in the final seconds is a nice touch — just make sure the styling is accessible (not color-only) for users with color vision differences.

Once the full test suite passes and the edge cases above are confirmed, this looks close to ready. Happy to do a follow-up review pass!

Thanks for the follow-up commits, @sixianli! Two new additions since the last review:

• fix: handle clarify review edge cases — good to see edge case coverage being added.
• fix: keep clarify countdown steady — this addresses any jitter or inconsistency in the countdown display, which matters for the UX since users are watching it.

The PR is looking solid across api/clarify.py, static/messages.js, static/index.html, static/style.css, and the two test files. The scope is well-contained.

For the countdown steady fix specifically: is the issue timer drift (e.g. setInterval vs wall-clock drift), or a rendering issue where the countdown was being re-initialized on re-renders? Either way, good catch before merge.

The PR is shaping up well — appreciate the iterative polish!

I pushed two follow-up commits after the review feedback:

• 70aea31 - handle clarify review edge cases
• 1611bc3 - keep clarify countdown steady

Timeout metadata source

Confirmed. The timeout metadata is populated server-side in api/clarify.py, not synthesized by the client.

submit_pending() calls _with_timeout_metadata() before the clarify request is queued or sent to the UI callback. That helper sets:

• requested_at
• timeout_seconds
• expires_at

The frontend only consumes those fields. It uses pending.expires_at first, with pending.requested_at + pending.timeout_seconds as a fallback when both are present.

Draft preservation edge cases

Confirmed and covered with follow-up tests.

If the main composer already contains text, the clarify draft is appended with a blank-line separator instead of replacing the existing composer value.

Explicit user actions do not trigger draft preservation:

• submitted clarify text uses the sent reason
• explicit stream cancellation now uses the cancelled reason
• _stashClarifyDraft() ignores both of those reasons

Draft preservation is limited to timeout / terminal closure paths, not explicit submit or cancel paths.

Countdown accessibility

The urgent countdown state now has a non-color visual cue as well as the error color. I added a regression test to make sure the urgent styling is not color-only.

Additional countdown fix found during manual testing

While testing the deployed PR build, we found one more countdown issue. The pending clarify poll refreshes the same prompt every ~1.5s, and the previous implementation restarted the countdown interval on each refresh. That could make the visible countdown jump by two numbers within one second.

Fixed in 1611bc3: _startClarifyCountdown() now keeps the existing interval running when the pending clarify payload has the same expires_at, so polling no longer resets the timer. Added regression coverage for this case.

Verification

Targeted regression checks:

• uv run --with pytest pytest tests/test_clarify_unblock.py tests/test_sprint30.py -q -> 88 passed
• uv run --with ruff ruff check api/clarify.py tests/test_clarify_unblock.py tests/test_sprint30.py -> passed
• git diff --check -> passed

Full suite:

I ran the full suite in a detached /tmp worktree with isolated HOME / Hermes state:

• pytest tests/ -v --timeout=60
• result: 1 failed, 2791 passed, 54 skipped, 1 warning, 8 subtests passed

The remaining failure is:

• tests/test_sprint31.py::TestProfileCreateAPIWithEndpoint::test_api_route_writes_base_url_to_config

I reproduced that same failure on upstream/master with the sprint31 test class, so it appears unrelated to this PR. I am not claiming the full suite is clean locally, but the remaining failure does not appear to be introduced by these clarify changes.

Manual verification:

• Deployed the updated PR branch to a test WebUI instance.
• Confirmed the clarify countdown now decrements steadily once per second.
• Confirmed the draft preservation behavior works as expected after timeout.

Thanks for the detailed follow-up, @sixianli! The two additional commits address the remaining review feedback thoroughly.

Timeout metadata source (confirmed): Good to have the explicit confirmation that _with_timeout_metadata() is called server-side before the request is queued, and the frontend only consumes expires_at / requested_at + timeout_seconds. The fallback chain is solid.

Draft preservation edge cases (confirmed):

• Append-with-separator behavior when the composer already has text is the right approach — avoids silently overwriting existing content.
• Restricting draft stash to timeout/terminal closure paths (not sent/cancelled) is correct. The elif guard on reasons prevents accidentally stashing drafts from explicit user actions.

Countdown accessibility (confirmed): Non-color visual cue added alongside the error color for the urgent state, with regression coverage. Good.

Countdown stability fix (1611bc3): The _startClarifyCountdown() same-expires_at guard is a nice catch — preventing the interval from resetting on each ~1.5s poll was a real UX issue (visible jump in the countdown). The fix is correct: keep the existing interval running if the payload hasn't changed.

88 passed on the targeted test modules and the full suite at 1 failed / 2791 passed. The remaining failure (test_api_route_writes_base_url_to_config) reproducible on upstream/master confirms it's pre-existing and unrelated.

This looks ready for maintainer review and merge. Great thorough work on the follow-ups! 🎉

Merged in v0.50.237 via #1243. Thank you @sixianli! 🎉