Summary

• Rebased PR fix(ui): stabilize WebChat message ordering #74733 onto current origin/main and replaced the old XL patch with a focused WebChat reliability fix.
• Keep optimistic user messages visible when a stale in-flight chat.history response resolves after a local send.
• Avoid clearing active assistant stream/tool state while a run is still in progress.
• Clear tool-stream artifacts during /clear/sessions.reset so old stream/tool cards cannot survive a reset.
• Add a small temp-path guard cleanup for the current-main context treemap output path so pnpm check passes.

Problem

WebChat can start a chat.history request, then the user sends a message before that request resolves. The old response was reconciled against the request-start message list, so it could overwrite the newer optimistic user message and clear the active stream. Visually, the sent message could disappear and then come back only after a later response/history refresh.

Approach

• Track local chat mutations separately from history request versions.
• When a history response resolves after local state changed, merge against the latest visible chat state and preserve unanchored optimistic messages that are not present in the stale history payload.
• Only clear streaming/tool state from history reloads when there is no active stream/run.
• Reset tool-stream collections in the synchronous /clear reset path before reloading history.

Real behavior proof

• Behavior or issue addressed: WebChat sent messages could disappear while an older chat.history request resolved after the local send; this patch keeps the optimistic send and active stream state visible.
• Real environment tested: Vladyslav's Mac mini running the real OpenClaw launchd gateway (ai.openclaw.gateway) on port 18789, installed from this fork via the global /opt/homebrew/bin/openclaw npm symlink.
• Exact steps or command run after this patch: Built and installed the final fork head, then restarted the real gateway: pnpm build, pnpm ui:build, npm install -g /Users/vladyslavlevchuk/Projects/openclaw-pr-74733, launchctl kickstart -k gui/$(id -u)/ai.openclaw.gateway, openclaw --version, curl -I http://127.0.0.1:18789/, and log inspection of ~/.openclaw/logs/gateway.log.
• Evidence after fix: Copied live output from the real setup:

$ openclaw --version
OpenClaw 2026.5.10-beta.1 (48efa0e)

$ launchctl print gui/501/ai.openclaw.gateway
state = running
pid = 82744
last exit code = 0
arguments = /opt/homebrew/bin/openclaw gateway --port 18789

$ tail ~/.openclaw/logs/gateway.log
2026-05-10T14:18:23.620-04:00 [gateway] http server listening (10 plugins: acpx, active-memory, bonjour, browser, canvas, device-pair, file-transfer, memory-core, phone-control, talk-voice; 2.5s)
2026-05-10T14:18:24.289-04:00 [gateway] ready

$ curl -I http://127.0.0.1:18789/
HTTP/1.1 200 OK
Content-Type: text/html; charset=utf-8
Cache-Control: no-cache

• Observed result after fix: The fork is running in a real OpenClaw gateway setup and serving the rebuilt Control UI from the final PR head. The WebChat race is covered by a regression test that reproduces the stale-history response landing after a local send and asserts the optimistic user message remains visible while the active stream remains intact.
• What was not tested: I did not manually send a production WebChat message through the operator's live session after installing; the live setup is installed for operator verification, and stale 2026.5.7 native clients may need an app restart or browser hard refresh to reconnect to the beta gateway.

Verification

• pnpm build passed.
• pnpm ui:build passed.
• pnpm check passed.
• pnpm --dir ui exec vitest run src/ui/controllers/chat.test.ts src/ui/chat/history-merge.test.ts src/ui/app-chat.test.ts --config vitest.config.ts passed: 101 tests.
• pnpm format:check CHANGELOG.md src/auto-reply/reply/context-treemap.ts ui/src/ui/controllers/chat.ts ui/src/ui/controllers/chat.test.ts ui/src/ui/chat/history-merge.test.ts ui/src/ui/app-chat.ts ui/src/ui/app-chat.test.ts passed.
• pnpm exec oxlint src/auto-reply/reply/context-treemap.ts ui/src/ui/controllers/chat.ts ui/src/ui/controllers/chat.test.ts ui/src/ui/chat/history-merge.test.ts ui/src/ui/app-chat.ts ui/src/ui/app-chat.test.ts passed.
• pnpm tsgo:test:ui passed.
• Final head 48efa0e303: pnpm check:changed passed.
• Final head 48efa0e303: env -u OPENCLAW_ALLOW_INSECURE_PRIVATE_WS -u OPENCLAW_GATEWAY_HOST -u OPENCLAW_GATEWAY_TOKEN -u OPENCLAW_VAPID_SUBJECT pnpm test:changed:max passed: 21 shards.

Full Test Note

I also ran pnpm test after pnpm build. The branch-relevant shards passed, but the full local run failed in unrelated areas because this machine exports live OpenClaw gateway/VAPID env vars and because some broad current-main tests are sensitive to built dist/ or unrelated mock drift. I did not fold those unrelated repairs into this WebChat PR.

AI Assistance

AI-assisted. I reviewed the resulting diff and understand the behavior it changes.