fix(tui): propagate verbose flag from thinking.delta to recordReasoningDelta

[AhmetArif0]

Problem

thinking_callback in tui_gateway/server.py emits thinking.delta without a verbose field, while the sibling reasoning_callback (added in #30225) already sets verbose: True when the session is in verbose mode (/verbose verbose).

On the client side, the thinking.delta case in createGatewayEventHandler.ts calls recordReasoningDelta(value) without the force flag, while reasoning.delta correctly passes Boolean(ev.payload.verbose):

// thinking.delta — BEFORE (force always false)
turnController.recordReasoningDelta(value)

// reasoning.delta — correct
turnController.recordReasoningDelta(ev.payload.text, Boolean(ev.payload.verbose))

recordReasoningDelta bails early when force=false and showReasoning=false:

recordReasoningDelta(text: string, force = false) {
  if (this.interrupted || (!force && !getUiState().showReasoning)) {
    return   // ← thinking blocks silently dropped
  }
  ...
}

Result: in /verbose verbose mode, extended-thinking blocks from the model are silently dropped when the user has reasoning display off (the default), even though the user explicitly enabled verbose output to see them. The reasoning.delta path works correctly for the same scenario.

Fix

Server (tui_gateway/server.py): mirror reasoning_callback in thinking_callback — emit verbose: True under _session_verbose().

Client (createGatewayEventHandler.ts): read ev.payload?.verbose in the thinking.delta case and pass it as the force flag to recordReasoningDelta.

Tests

Added "shows verbose thinking.delta even when normal reasoning display is off" — symmetric with the existing "shows verbose reasoning even when normal reasoning display is off" test from #30225.

✓ shows verbose thinking.delta even when normal reasoning display is off

43/43 tests pass in createGatewayEventHandler.test.ts.

Checklist

• Parity fix: symmetric with reasoning_callback / reasoning.delta (already correct since fix(tui): surface verbose tool details #30225)
• No behavior change when _session_verbose() is false — verbose field is omitted, force stays false
• ruff check tui_gateway/server.py passes
• 43/43 TS tests pass

[teknium1]

Thanks for the focused parity investigation. The current implementation does not support the proposed event semantics.

Problems

• thinking_callback is the quiet-mode spinner path: agent/conversation_loop.py:1073-1079 emits a generated spinner label and clears it later. Model reasoning uses run_agent.py:4745-4750 and is already emitted with verbose support by tui_gateway/server.py:3890-3894. Recording verbose thinking.delta would therefore turn spinner text into persisted reasoning.
• The changed handler reads ev.payload?.verbose, but ui-tui/src/gatewayTypes.ts:620 defines thinking.delta as { text?: string }; ui-tui/tsconfig.json has strict: true. The added test casts its event to any, so it does not exercise that protocol typing mismatch.

Suggested changes

• Confirm a current producer routes model reasoning through thinking.delta before changing this status event; otherwise keep the existing verbose-aware reasoning.delta path.
• If extending the event intentionally, update ui-tui/src/gatewayTypes.ts, ui-tui/README.md, and add a server _agent_cbs payload test.

Automated hermes-sweeper review.

[teknium1]

GatewayEvent currently types thinking.delta payload as { text?: string } in ui-tui/src/gatewayTypes.ts:620. Add verbose?: boolean to that protocol variant (and its documented wire payload) before reading this field; the test's as any cast bypasses the strict TypeScript error.

[GottZ]

This was generated by AI during triage.

Summary

Two PRs address the reported behavior with identical diffs, adding a verbose flag to thinking.delta and forcing those events into recorded reasoning. The diff does not establish the reported cause, however: the contributor review identifies thinking.delta as the quiet-mode spinner/status path, while model reasoning already uses the verbose-aware reasoning.delta path.

Related pull requests

• #30446 related — (+25/-2) — keep open for revision, do not merge as-is: the diff would persist spinner/status text as reasoning and accesses an undeclared verbose field on the strictly typed thinking.delta payload; its any-cast test masks that protocol mismatch. This follows the contributor keep_open review on #30446, which requires evidence of a current model-reasoning producer for thinking.delta or, if the protocol is intentionally extended, corresponding gateway type, documentation, and server coverage.
• #30501 [closed] duplicate — (+25/-2) — closed duplicate, still relevant because its diff is identical to #30446 across the server emitter, client handler, and any-cast test, so it has the same semantic and typing defects. It was correctly superseded by #30446 and provides no independent fix to consolidate.

Duplicates

#30501 is an exact duplicate of #30446: both apply the same +25/-2 changes to the same three files.

Suggested consolidation

Do not merge either PR as currently written. Keep #30446 as the sole revision point, requiring producer-level evidence that model reasoning traverses thinking.delta and a protocol-complete implementation if that event is intentionally extended; keep duplicate #30501 closed.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    subgraph Dup30446 ["PRs duplicating each other"]
        P30446["PR #30446 (open)"]
        P30501["PR #30501 (closed)"]
    end
    class P30446 open
    class P30501 closed
    class P30446 target
    click P30446 "https://github.com/NousResearch/hermes-agent/pull/30446"
    click P30501 "https://github.com/NousResearch/hermes-agent/pull/30501"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 2 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 5 kB of PR diffs, 3 kB of issue/PR text, 2 kB of discussion (3 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.