emrg: TUI status bar shows current context message count (rant 21:52:18)

[argszero]

Fixes host rant 2026-08-13T21:52:18: the TUI status bar · N msgs · displayed the locally-tracked total message count (including compacted history), which misleads — it should show the number of messages actually sent to the LLM in the current context.

Daemon (emrg/server/daemon.py): every done frame now carries context_messages: len(messages) — the authoritative current-context size (system + history + user + all tool results + assistant replies; re-derived after auto-compact rebuild). Added to both the Case-1 (stop) and Case-3 (max-tokens) final done frames.

TUI (emrg/client/app.py): the done handler sets msg_count = context_messages when present (daemon-authoritative, reflects tool results and auto-compact), falling back to the local +1 approximation otherwise. Status-bar copy unchanged.

Test: test_streaming_task_with_tool_calls extended to assert context_messages == 5 for a one-round tool flow (system + user + assistant-tool_calls + tool-result + final assistant).

Verified: pytest 801/801, import + CLI OK. GUI untouched.

[argszero]

✅ LGTM — cycle (1/3)

Verified head 32442e1 (3 files, +21/-1):

• daemon.py: both final done frames (Case 1 stop @2271, Case 3 max-tokens @2445) now carry context_messages: len(messages) — authoritative current-LLM-context size (system + history + user + tool results + assistant replies), correctly re-derived after auto-compact rebuild (messages rebuilt @2170).
• app.py done handler: msg_count = int(context_messages) when present, +1 fallback otherwise; status-bar copy unchanged (· N msgs · path).
• e2e test: one-round tool flow asserts context_messages == 5 (system + user + assistant-tool_calls + tool-result + final assistant) — matches the actual message flow (initial 2, +assistant_msg@2313, +tool result@2397, +final assistant@2435).
• pytest 801/801 local, CI test + test-windows PASS (run 31708272704).

[argszero]

✅ LGTM — cycle (2/3)

Re-verified head 32442e1 (unchanged since 1/3):

• Diff vs master 44b09d8 = 3 files +21/-1: daemon.py adds context_messages: len(messages) to both final done frames (Case 1 stop, Case 3 max-tokens); app.py consumes it with +1 fallback; e2e asserts context_messages == 5 for the one-round tool flow.
• No overlap with the #771 GUI merge (daemon/app/test files vs GUI files) — no rebase needed, MERGEABLE.
• CI test + test-windows PASS (run 31708272704).

[argszero]

✅ LGTM — cycle (2/3, head 32442e1)

Fresh independent verification this cycle:

• daemon.py: BOTH final done frames now carry context_messages: len(messages) — Case 1 (loop exit @~2277) and Case 3 (max-tokens @~2455) — authoritative current LLM context (system + history + user + tool results + assistant replies), correctly re-derived after auto-compact rebuild.
• app.py done handler: int(context_messages) when present, +1 local fallback otherwise (backward compatible with older daemons); status-bar copy unchanged.
• e2e test test_streaming_task_with_tool_calls: asserts context_messages == 5 (system + user + assistant-tool_calls + tool-result + final assistant) — ran locally: 1 passed.
• CI test + test-windows PASS (run 31708272704); branch = 3 files (+21/-1), based on master 44b09d8.