🤖 fix: make hidden-message indicators follow omission seams

[ammar-agent]

Summary

Refactors transcript truncation omission handling into an explicit planner that emits history-hidden marker rows at omission seams, removes inline user-gap reminder rendering, and hardens deferred snapshot bypass logic so stale deferred ordering cannot temporarily misplace omission indicators.

Background

The previous omission path mixed a single global marker with per-user hidden-gap reminders, which encoded placement policy in implicit loop state (insertionIndex, pendingOmittedCount, consumedFirstOmittedRun). That made omission placement harder to reason about and produced edge cases where indicators felt disconnected from omitted runs.

Implementation

• Added buildTranscriptTruncationPlan in src/browser/utils/messages/transcriptTruncationPlan.ts:
  • collects omission runs in old history,
  • emits explicit history-hidden rows per run at seam positions,
  • caps marker row count (MAX_HISTORY_HIDDEN_SEGMENTS) by merging older runs to protect DOM size.
• Switched StreamingMessageAggregator truncation path to call the planner.
• Removed hiddenCountBeforeUser from DisplayedMessage user type and removed inline reminder UI in UserMessage.
• Updated deferred bypass logic (shouldBypassDeferredMessages) to bypass when snapshots are same length but diverge in row identity/order.
• Added/updated tests:
  • new planner unit tests (transcriptTruncationPlan.test.ts),
  • updated aggregator truncation expectations for multi-segment markers,
  • updated deferred-bypass tests,
  • updated UI truncation test expectations to aggregate across marker rows.

Validation

• make static-check
• bun test src/browser/utils/messages/transcriptTruncationPlan.test.ts src/browser/utils/messages/StreamingMessageAggregator.test.ts src/browser/utils/messages/messageUtils.test.ts
• Attempted bun test ./tests/ui/chat/truncation.test.ts -t "shows a generic hidden indicator and preserves assistant meta rows" (environmental failure in this workspace: Cannot find module '@/version' from telemetry service import path)

Risks

• Behavior change: omission UI now surfaces explicit seam markers (capped) instead of per-user reminder text.
• Risk is primarily in transcript rendering expectations and any downstream assumptions about marker cardinality.
• Mitigated with planner-focused and aggregator/UI expectation updates.

---

Generated with mux • Model: openai:gpt-5.3-codex • Thinking: xhigh • Cost: $0.00

[ammar-agent]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Hooray!

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[ammario]

Manually verified that a long chat appears to render more correctly after this change.