[Hermes Desktop][hermes-desktop-local-ai-thinking-off-default][1/n] Keep thinking off on session start

[OmarB97]

Why

Hermes desktop can start with Thinking Off, but the lazy session-start path was still allowed to hydrate an agent with the default reasoning behavior and then report Medium back to the desktop. That made the status/menu flip from Off to Medium as soon as a chat session started.

What changed

• Default blank desktop/TUI reasoning config to disabled instead of letting it fall through to the agent default.
• Report missing or disabled session reasoning as none, including the lightweight session.create response emitted before the real agent finishes hydrating.
• Keep desktop model/status labels on Off instead of silently relabeling disabled thinking as Med.

How to review

• Start in tui_gateway/server.py around _load_reasoning_config, _reasoning_effort_for_session_info, and session.create.
• Then check the desktop label/menu expectations in apps/desktop/src/lib/model-status-label.test.ts and apps/desktop/src/app/shell/model-edit-submenu.test.ts.
• Confirm this fork mirror is only the two thinking-off commits on top of fork main.

Evidence

• Live Hermes desktop relaunched from /Users/obaradei/.hermes/hermes-agent after rebasing onto fork main at PR fix(cli): preserve chat-local provider flags #86.
• Live /api/ws session.create returned reasoning_effort: none for qwen3.6-27b, then the probe session was closed.

Verification

• venv/bin/python -m pytest tests/test_tui_gateway_server.py -q -k "reasoning_none or blank_reasoning or session_create_lazy_info"
• npm run test:ui -- src/lib/model-status-label.test.ts src/app/shell/model-edit-submenu.test.ts

Risks / gaps

• Full tests/test_tui_gateway_server.py was not green locally because test_browser_manage_connect_default_local_reports_launch_hint depends on local browser plugin/environment behavior; accepted risk for this PR because it is outside the changed reasoning/session metadata path and the focused gateway tests cover the regression.
• This PR mirrors the thinking-off fix into the fork while the upstream contribution remains tracked separately on Add opt-in "reasoning off for local models" preference NousResearch/hermes-agent#39164; covered by MeshBoard task hermes-desktop-local-ai-thinking-off-default, so no work item needed here.

Collaborators

• Omar Baradei: operator, ko-mac, reported the 2026-06-04 session-start Off to Medium regression and confirmed fork PR fix(desktop): emit preflight context usage #85 had merged.
• Codex: AI coding agent on ko-mac, 2026-06-04 11:00-11:25 PT, MeshBoard task hermes-desktop-local-ai-thinking-off-default, implemented, rebased live checkout, tested, and verified the live session-start contract.

Task: hermes-desktop-local-ai-thinking-off-default

[github-actions]

🔎 Lint report: codex/hermes-thinking-off-session-start-fork vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 9848 on HEAD, 9846 on base (🆕 +2)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5108 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.