Summary

• Problem: main WebChat could appear stuck or lose visible continuity because accepted browser sends were not surfaced early enough, brief Control UI detach/reattach could tear down websocket state, and high-context sessions could wait too long before native compaction.
• Why it matters: users could send a prompt, see no acknowledgement/progress, refresh to recover, and risk losing confidence that the backend accepted the turn.
• What changed: added early WebChat preflight/receipt handling, pending user-turn history continuity, delayed Control UI disconnect cleanup, safer session reset transcript allocation, earlier preflight compaction gates, and focused tests.
• What did NOT change (scope boundary): default WebChat tool capability policy is not restricted; explicit/custom reset transcript paths are preserved; user-authored System: text is no longer globally stripped.

Change Type (select all)

• Bug fix
• Feature
• Refactor required for the fix
• Docs
• Security hardening
• Chore/infra

Scope (select all touched areas)

• Gateway / orchestration
• Skills / tool execution
• Auth / tokens
• Memory / storage
• Integrations
• API / contracts
• UI / DX
• CI/CD / infra

Linked Issue/PR

• Closes #
• Related #
• This PR fixes a bug or regression

Root Cause (if applicable)

• Root cause: WebChat accepted browser sends and streamed assistant progress were coupled too tightly to later transcript/history and socket lifecycle updates; reset could reuse stale default transcript file metadata for a new session id; and preflight compaction could miss fresh persisted token totals.
• Missing detection / guardrail: coverage did not lock in pending user-turn visibility, stale-default reset transcript replacement, user-authored System: preservation, or the exact preflight compaction threshold behavior.
• Contributing context (if known): local reproduction showed replies/pending turns sometimes became visible only after refresh, and short visible chats could still carry high cached context from app-server trajectory/tool history.

Regression Test Plan (if applicable)

• Coverage level that should have caught this:
  • Unit test
  • Seam / integration test
  • End-to-end test
  • Existing coverage already sufficient
• Target test or file: src/gateway/server.sessions.reset-models.test.ts, src/auto-reply/reply/strip-inbound-meta.test.ts, src/gateway/server-methods/chat.inject.parentid.test.ts, src/gateway/server.chat.gateway-server-chat.test.ts, ui/src/ui/controllers/chat.test.ts, ui/src/ui/views/chat.test.ts, ui/src/ui/chat/message-normalizer.test.ts, plus compaction/memory-flush tests already in the branch.
• Scenario the test should lock in: accepted WebChat turns remain visible through history refresh, reset avoids stale default transcript reuse while preserving explicit custom transcript paths, user-authored System: lines remain intact, preflight acknowledgement text stays ephemeral, and preflight compaction gates trigger from fresh token totals.
• Why this is the smallest reliable guardrail: these tests cover the server history/reset/sanitizer contracts and the UI message normalization/rendering paths without requiring a browser E2E harness.
• Existing test that already covers this (if any): existing chat history/UI tests cover parts of message normalization and rendering; this PR adds/updates focused reset, sanitizer, preflight, pending-history, and message-normalizer coverage for the review findings.
• If no new test is added, why not: N/A.

User-visible / Behavior Changes

Control UI WebChat should show a received/working state sooner, keep the pending user prompt visible during refresh/reconnect windows, and avoid unnecessary context saturation before compaction. Default WebChat tools/capabilities are unchanged by this PR.

Diagram (if applicable)

Before:
[user sends prompt] -> [chat.send accepted] -> [history/socket lag] -> [user may refresh to see progress]

After:
[user sends prompt] -> [preflight receipt + pending user turn] -> [history catches up] -> [final assistant reply]

Security Impact (required)

• New permissions/capabilities? No
• Secrets/tokens handling changed? No
• New/changed network calls? No
• Command/tool execution surface changed? No
• Data access scope changed? No
• If any Yes, explain risk + mitigation: N/A.

Repro + Verification

Environment

• OS: macOS local development checkout
• Runtime/container: local OpenClaw source checkout
• Model/provider: N/A for automated tests
• Integration/channel (if any): Control UI WebChat / gateway session APIs
• Relevant config (redacted): main WebChat session agent:main:main

Steps

1. Send a main WebChat prompt while the session is slow/high-context or while the page refreshes/reconnects.
2. Observe whether the user turn and assistant working/receipt state are visible before final transcript history catches up.
3. Reset a session that has a stale default transcript path and a session that has an explicit custom transcript path.
4. Display/history-sanitize a user message beginning with System:.
5. Exercise ordinary prompts that contain exactly to make sure they still receive preflight acknowledgement, while explicit output-only/silent prompts do not.

Expected

• User turn remains visible after accepted send/refresh.
• Assistant preflight/progress state can render before final reply without becoming a durable assistant message.
• Reset allocates a new default transcript for a new session id, but preserves explicit custom transcript files.
• User-authored leading System: text is preserved.
• The Control UI transport label is hidden only for local user messages, not assistant/tool messages.

Actual

• Before this PR, visibility could lag until refresh, reset could reuse a stale default transcript path, and the original branch globally stripped leading System: text. After the latest patch, focused tests cover the intended behavior and the latest Copilot review comments are addressed in the current head.

Evidence

• Failing test/log before + passing after
• Trace/log snippets
• Screenshot/recording
• Perf numbers (if relevant)

Recorded verification for the current head 4cb8cce017b0ce976c691066b524f417dd2f633a:

• corepack pnpm test src/agents/prompt-overlay-runtime-contract.test.ts src/auto-reply/reply/get-reply-run.media-only.test.ts src/auto-reply/reply.raw-body.test.ts src/gateway/server.sessions.reset-models.test.ts passed: 3 Vitest shards, 74 tests.
• corepack pnpm test src/gateway/server.chat.gateway-server-chat.test.ts src/gateway/server.chat.gateway-server-chat-b.test.ts ui/src/ui/controllers/chat.test.ts ui/src/ui/views/chat.test.ts ui/src/ui/chat/message-normalizer.test.ts passed: 2 Vitest shards, 169 tests.
• corepack pnpm exec oxfmt --check --threads=1 src/agents/gpt5-prompt-overlay.ts src/agents/prompt-overlay-runtime-contract.test.ts src/auto-reply/reply/get-reply-run.media-only.test.ts src/gateway/server-methods/chat.ts ui/src/ui/controllers/chat.ts ui/src/ui/chat/run-lifecycle.ts ui/src/ui/chat/message-normalizer.ts ui/src/ui/views/chat.test.ts ui/src/ui/chat/message-normalizer.test.ts passed.
• corepack pnpm exec tsc --noEmit --pretty false -p tsconfig.core.json passed.
• git diff --check passed.
• Live GitHub PR checks for current head are successful, with only expected skipped backfill-pr-labels.

Note: a prior non-elevated gateway test run hit local sandbox listen EPERM 127.0.0.1, then passed when rerun with local loopback binding allowed.

Human Verification (required)

• Verified scenarios: source diff reviewed against BunsDev, ClawSweeper, Barnacle, and Copilot findings; focused prompt-overlay, server, UI, and session tests passed; PR branch head is 4cb8cce017b0ce976c691066b524f417dd2f633a.
• Edge cases checked: custom reset transcript path preservation, stale default reset transcript replacement, preservation of user-authored System: lines, no hard-coded WebChat toolsAllow remains, Control UI transport sender label is hidden only for user messages, ordinary prompts containing exactly still receive acknowledgement, upstream heartbeat-ack filtering remains combined with this PR's ephemeral preflight-stream guard, and the global GPT-5 stable prefix no longer asks the model to emit a second durable acknowledgement.
• What you did not verify: full browser E2E on GitHub CI, maintainer-specific release packaging, or a production OpenClaw update containing this PR.

Real behavior proof

• Behavior or issue addressed: Main Control UI WebChat reliability for accepted browser sends: pending user turns should remain visible through history refresh/reconnect, preflight/working state should not become a persisted assistant final, generic runtime system events should still reach the reply run, and the PR branch should remain mergeable after upstream main churn.
• Real environment tested: macOS local OpenClaw source checkout at PR head 4cb8cce017b0ce976c691066b524f417dd2f633a, branch codex/openclaw-ui-reliability-upstream. GitHub CLI/API was run against the live PR openclaw/openclaw#75776; focused Control UI/Gateway behavior was run from the local source checkout.
• Exact steps or command run after this patch: rebased the branch onto current main, resolved the system-event test conflict created by upstream's drainFormattedSystemEventBlock shape, removed the global GPT-5 stable-prefix progress acknowledgement so the WebChat receipt remains gateway/UI-owned and ephemeral, ran the focused verification listed above, pushed the branch, then checked live GitHub mergeability/check state for the current head.
• Evidence after fix: Terminal capture from the live GitHub PR and current head:

$ gh pr view 75776 --repo openclaw/openclaw --json headRefOid,baseRefOid,mergeable,mergeStateStatus,reviewDecision
{"baseRefOid":"9c389487002ca7d8558bd7ae98e310a44dfee0fa","headRefOid":"4cb8cce017b0ce976c691066b524f417dd2f633a","mergeStateStatus":"UNSTABLE","mergeable":"MERGEABLE","reviewDecision":"CHANGES_REQUESTED"}

$ gh api repos/openclaw/openclaw/pulls/75776 --jq '{mergeable,mergeable_state,rebaseable,head_sha:.head.sha,base_sha:.base.sha}'
{"base_sha":"9c389487002ca7d8558bd7ae98e310a44dfee0fa","head_sha":"4cb8cce017b0ce976c691066b524f417dd2f633a","mergeable":true,"mergeable_state":"unstable","rebaseable":true}

Current checks:
Real behavior proof: pass
Socket Security: Project Report: pass
Socket Security: Pull Request Alerts: pass
auto-response: pass
dependency-change-awareness: pass
dispatch: pass
label: pass
label-issues: pass
backfill-pr-labels: skipped

• Current-code review proof for the latest Copilot findings:
  • ui/src/ui/chat/message-normalizer.ts now hides openclaw-control-ui only when role === "user".
  • src/gateway/server-methods/chat.ts no longer treats exactly alone as a blanket acknowledgement-suppression phrase; suppression is limited to explicit silence/output-only wording.
• Current-code review proof for the latest ClawSweeper finding:
  • src/agents/gpt5-prompt-overlay.ts no longer contains the global <progress_acknowledgement> stable-prefix block.
  • src/agents/prompt-overlay-runtime-contract.test.ts now asserts that the GPT-5 stable prefix does not contain <progress_acknowledgement>.
• Observed result after fix: The latest PR head is mergeable against current main, checks are green/skipped as expected, the active code addresses the two Copilot findings plus the ClawSweeper P2 durable-acknowledgement finding, and focused source-run verification covers the server/UI/prompt behavior this PR changes.
• What was not tested: Full browser E2E against a packaged release build, maintainer release packaging, and production OpenClaw update delivery were not tested in this PR checkout.

Review Conversations

• I replied to or resolved every bot review conversation I addressed in this PR.
• I left unresolved only the conversations that still need reviewer or maintainer judgment.

ClawSweeper issue-comment findings were addressed in commit a4396316 and later follow-up commits; BunsDev's system-event review was addressed in the current branch; the two Copilot comments from May 9 are addressed at current head 4cb8cce017b0ce976c691066b524f417dd2f633a; the ClawSweeper P2 about a second durable GPT-5 acknowledgement is addressed by removing the global <progress_acknowledgement> stable-prefix block.

Compatibility / Migration

• Backward compatible? Yes
• Config/env changes? No
• Migration needed? No
• If yes, exact upgrade steps: N/A.

Risks and Mitigations

• Risk: WebChat acknowledgement/progress copy could be perceived as an assistant response if final streaming fails.
  • Mitigation: preflight receipt is ephemeral UI state and final transcript/history remains authoritative.
• Risk: reset transcript handling could accidentally preserve stale default files or drop custom files.
  • Mitigation: reset now preserves only non-default/custom sessionFile paths and adds coverage for stale default replacement plus existing custom-file preservation.
• Risk: earlier context compaction may run more often for high-context sessions.
  • Mitigation: threshold is bounded and only triggers from persisted/fresh token pressure; normal low-context sessions are unaffected.