A settled row carries on only its own part's reveal, so a looping agent's thoughts no longer flicker (#569) #571

Merged
cmoriarty merged 2 commits from claude/reveal-569 into main 2026-11-20 15:22:38 -05:00
Owner

A settled Thought or answer picked up a live row's reveal by matching its text. On run 204 the code review looped, and every thought opened with the same words. So each earlier Thought the list mounted took the new thought's reveal, drew itself again from that place, and ended the reveal. The new thought then started again from nothing, and the agent pane switched layouts every 40 to 120 ms (the owner's screen recording). braid#552's fix was already live: this is a second cause of the same flicker.

OpenSpec change reveal-handover-by-part, for #569.

What changed

  • Lines carry the ids of their parts: a thought's one part, or each text part joined into a message's prose.
  • One function, revealKey, names a reveal for both the live row and the hand-over.
  • adoptReveal looks only at the reveals of the line's own parts. The text still places the reveal within the settled text, as before, so a part that settles mid-reveal carries on at its pace.
  • The browser tests' server gains /_test/delta, which publishes a frame of a part being written as the engine does, and the alike_thoughts seed state. No journey could stream live text before.

Tests

Each check failed on main first.

Check On main Here
Journey ui/e2e/live-thought.spec.ts: the earlier thought's first frame, once scrolled into view only the new thought's text its whole text
Same journey: the earlier thought's height over 120 frames 4 heights 1
Same journey: the new thought's first frame, back at the tail empty its text
vitest ui/src/lib/reveal.test.ts: an earlier part's row that opens like the live part handed the live part's reveal null, live reveal untouched

A probe outside the suite reproduced run 204's conditions on the test server: a window 2200 px wide with scrollbars drawn, the looping thought streaming in, and the pane wheeled up and down for ten seconds. The earlier thought changed height 6 times in each of two runs on main's UI, and 0 times with the fix.

./tools/test.sh passed in 57 s.

🤖 Generated with Claude Code

A settled Thought or answer picked up a live row's reveal by matching its text. On run 204 the code review looped, and every thought opened with the same words. So each earlier Thought the list mounted took the new thought's reveal, drew itself again from that place, and ended the reveal. The new thought then started again from nothing, and the agent pane switched layouts every 40 to 120 ms (the owner's screen recording). braid#552's fix was already live: this is a second cause of the same flicker. OpenSpec change `reveal-handover-by-part`, for #569. ## What changed - Lines carry the ids of their parts: a thought's one part, or each text part joined into a message's prose. - One function, `revealKey`, names a reveal for both the live row and the hand-over. - `adoptReveal` looks only at the reveals of the line's own parts. The text still places the reveal within the settled text, as before, so a part that settles mid-reveal carries on at its pace. - The browser tests' server gains `/_test/delta`, which publishes a frame of a part being written as the engine does, and the `alike_thoughts` seed state. No journey could stream live text before. ## Tests Each check failed on main first. | Check | On main | Here | |---|---|---| | Journey `ui/e2e/live-thought.spec.ts`: the earlier thought's first frame, once scrolled into view | only the new thought's text | its whole text | | Same journey: the earlier thought's height over 120 frames | 4 heights | 1 | | Same journey: the new thought's first frame, back at the tail | empty | its text | | vitest `ui/src/lib/reveal.test.ts`: an earlier part's row that opens like the live part | handed the live part's reveal | `null`, live reveal untouched | A probe outside the suite reproduced run 204's conditions on the test server: a window 2200 px wide with scrollbars drawn, the looping thought streaming in, and the pane wheeled up and down for ten seconds. The earlier thought changed height 6 times in each of two runs on main's UI, and 0 times with the fix. `./tools/test.sh` passed in 57 s. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
A settled row takes over a live row's reveal by its part, not its text.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fix(console): a settled row carries on only its own part's reveal (#569)
All checks were successful
test / suite (pull_request) Successful in 1m29s
e9e269f0b7
A settled Thought or answer picked up a live row's reveal by its text. On run 204 the code review
looped and every thought opened with the same words, so each earlier Thought the list mounted took
the new thought's reveal, drew itself again from that place and ended the reveal. The new thought
then started again from nothing, and the pane flickered as the heights swung.

Lines now carry the ids of their parts, one function names a reveal for the live row and the
hand-over, and `adoptReveal` looks only at the reveals of the line's own parts. The text only
places the reveal, as before.

Tests: a vitest test of the hand-over, failing first on the text matching, and the journey
`ui/e2e/live-thought.spec.ts`, which streams a thought through the test server's new
`/_test/delta` route. On main it fails three ways: the earlier thought's first frame showed only
the new thought's text, its height changed 4 times in 120 frames, and back at the tail the new
thought's text was empty.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid!571
No description provided.