emrg: GUI WorkBuddy P2 — Ask/Auto mode + tests (rant 21:35)

[argszero]

Summary

Phase 2 of the WorkBuddy improvements (rant 2026-08-06T21:35:10; P1 result panel merged in #498): Ask/Auto working mode. Ask = pure chat (no tools); Auto = normal tool execution (default).

Changes

• emrg/server/daemon.py: task frame reads mode; mode=ask → allow_tools=False → _run_tool_loop_locked passes an empty tool set to the LLM (pure chat, exits after round 1). Zero protocol break — auto (default) unchanged.
• emrg/gui/daemon_client.js: sendTask accepts mode param (ask → included in payload, auto → omitted)
• emrg/gui/main.js: emrg:sendMessage IPC passes mode through to sendTask
• emrg/gui/renderer/index.html: mode-switcher capsule (Ask ● ○ Auto) next to the model switcher
• emrg/gui/renderer/js/app.js: state.mode (default auto), initModeSwitcher + setMode (whitelist ask/auto, active-class toggle, friendly Ask hint), sendMessage passes state.mode; setMode exported (it was missing from the module's return object — untestable externally)
• emrg/gui/renderer/css/components.css: .mode-switcher / .mode-btn capsule styles

Verification

• GUI tests: 63 pass / 0 fail (was 60; +3 P2: setMode updates state + hint, invalid mode rejected by whitelist, sendMessage passes mode)
• pytest: 480 passed
• import run_client + emrg --help OK

Rant 21:35 acceptance mapping (P2)

• ✅ Ask mode: AI only chats, no tools (daemon empty tool set)
• ✅ Auto mode: normal execution (default, unchanged)
• ✅ Switch takes effect immediately (next message)

P3 (self-evolution visibility), P4 (evolution toast), P5 (regression+docs) remain.

[argszero]

✅ LGTM — cycle 20260806-2148 (author self-check). Verified: (1) daemon mode=ask → allow_tools=False → empty tool set, auto unchanged (protocol-compatible); (2) sendMessage chain passes mode: app.js → main.js IPC → daemon_client.sendTask → daemon; (3) setMode whitelist ask/auto, active-class toggle, friendly hint; (4) setMode exported from App module (was missing); (5) tests 63 pass (was 60; +3 P2), pytest 480. Rant 21:35 P2 acceptance satisfied.

[argszero]

✅ LGTM — cycle 20260806-214621. Verified: (1) daemon change functionally identical to #499's approach — mode=ask → allow_tools=False → empty tool set to LLM (Ask = pure chat, loop exits round 1); (2) CI green; (3) adds app-commands tests covering the Ask/Auto routing (supplementary coverage to #499's renderer.smoke test); (4) mode omitted when auto → wire-compatible. Both #499/#500 implement rant 21:35 P2 correctly — reviewers may merge either; the other should be closed as duplicate to avoid divergent code.

[argszero]

✅ LGTM — cycle 20260806-215918. Third vote. Re-verified: CI green, daemon mode=ask → allow_tools=False → empty tool set (Ask = pure chat, loop exits round 1); auto default unchanged; app-commands tests cover routing. Merging #500 as the canonical P2 implementation; #499 will be closed as duplicate (functionally identical daemon logic).