fix: stabilize hold-mode assistant autoscroll#533
Conversation
Hold mode now latches after manual escape or when the Assistant answer text reaches the viewport top, and it remains held until the user manually reaches the true bottom. Content growth, hold-target changes, and stream completion no longer re-enable follow or snap the viewport while held. Hold target eligibility is restricted to user-readable Assistant answer text. Tool/status/output blocks and reflection/reasoning-only content are excluded from hold targeting, and held Assistant text anchors are restored across subsequent content renders. Verification: node --test packages/ui/src/components/virtual-follow-behavior.test.ts; npm run typecheck --workspace @codenomad/ui; npm run build --workspace @codenomad/ui.
|
PR builds are available as GitHub Actions artifacts: https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/27159386614 Artifacts expire in 7 days.
|
Prompt submission now performs an explicit bottom jump that suppresses stale Hold latches before and after sending, so the optimistic user prompt and the new exchange return to the true bottom after manual scrolling in Hold mode. The virtual follow controller also exposes a focused hold-clear transition for disabling Hold while held, and optimistic user rendering now keeps only the primary submitted text visible while preserving synthetic helper-part hiding. Verified with the virtual follow behavior node test, UI typecheck, UI build, and diff whitespace checks.
Adds a durable submit/new-stream bottom-follow intent so Hold mode keeps the newly submitted prompt and assistant stream at the bottom until the new exchange is mounted and settled. Suppresses stale previous Assistant hold anchors and snapshot restores while that intent is active, then preserves suppression if the old target is still current so it cannot reclaim the viewport at stream start. Validated with node --test packages/ui/src/components/virtual-follow-behavior.test.ts, npm run typecheck --workspace @codenomad/ui, npm run build --workspace @codenomad/ui, and git diff --check.
|
I want you to perform the gatekeeper checks for latest PR state check for regressions |
|
PR builds are available as GitHub Actions artifacts: https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/27163704081 Artifacts expire in 7 days. |
|
Gatekeeper review complete for the latest PR state. Request understood: perform a ruthless review for regressions and better implementations. I treated this as review-only and did not modify code. Findings:
Validation:
Gatekeeper verdict: do not merge as-is. The active-hold latching behavior is directionally correct, but the broad -- |
|
PR builds are available as GitHub Actions artifacts: https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/27163817328 Artifacts expire in 7 days. |
Limit the virtual follow bottom-pin suspension gate to external suspension or an actively latched Hold target, so escaped Assistant streaming can rejoin near the bottom while active Hold remains latched until true bottom. Adds behavior coverage for future eligible hold targets versus active hold latches.
|
PR builds are available as GitHub Actions artifacts: https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/27164365361 Artifacts expire in 7 days.
|
|
Gatekeeper blocker from #533 (comment) addressed in c236b7f. The bottom-pin suspension now applies only to external suspension or an actively latched Hold target, so ordinary escaped streaming can rejoin near the bottom while active Hold still remains latched until true bottom. Added regression coverage; local validation passed 37/37 virtual-follow tests, UI typecheck, UI build, and diff checks. The medium note about measuring a dedicated inner Assistant text element remains a documented non-blocking follow-up. |
|
@CodeNomadBot |
|
Gatekeeper re-review complete for the latest PR state after Request understood: “again” means re-run the same ruthless PR gatekeeper check against the current branch. I treated this as review-only and did not modify code. Findings:
Validation from this bot checkout:
Gatekeeper verdict: mergeable from code-review perspective. I do not see a remaining blocking regression in the latest diff. The CI/local dependency-backed typecheck and build results should be trusted over this bot checkout for those two commands because this environment is missing required packages. -- |
|
PR builds are available as GitHub Actions artifacts: https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/27164936545 Artifacts expire in 7 days.
|
|
PR builds are available as GitHub Actions artifacts: https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/27478538808 Artifacts expire in 7 days.
|
1 participant
Summary
Validation