fix(desktop,bots): render "(empty)" sentinel as a friendly message in group chat

[RibatTRW]

Problem

A bot in the desktop Bot Mode group chat posts the raw "(empty)" when its turn produces no output. The gateway shows a friendly notice instead. The desktop bridge does not.

Fix

Normalize the sentinel at appendGroupChatEntry, matching the gateway text (gateway/run.py). "(pass)" silence and all other text are unchanged.

Tests

node --test apps/desktop/src/plugins/hermes-bots/tests/group-chat-empty-sentinel.test.mjs (2 pass). The test fails when the fix is reverted.

Fixes #94308

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

Right fix shape, and the exact-trim equality faithfully mirrors the canonical matcher (agent/turn_finalizer.py compares _stripped == "(empty)", so no divergence there). Three points:

1. The normalization hits user-authored entries too — apps/desktop/src/plugins/hermes-bots/plugin.js:6887: appendGroupChatEntry is also how "You" messages land in the room log (the test fixture itself shows { from: { kind: 'user', name: 'You' } } entries). A member who literally types (empty) into the room gets their message silently replaced by the ⚠️ model-failure text — wrong attribution and confusing in the opposite direction. Gate on the author kind (from.kind === 'member' / bot) so only machine replies are rewritten.

2. Test-file hygiene — tests/group-chat-empty-sentinel.test.mjs:12-19 carries a TRACE_ATOM env-gated console.error debug scaffold inside the atom setter, and the harness drags in substantial machinery this file never uses (busyUntilResumeCall, conflictOnce conflict injection, clarify/approval payloads, session.create/resume/prompt.submit scripting). Two tests need maybe a tenth of it. Trim to what's exercised (or factor a shared harness module if one doesn't exist yet — three files now duplicate this loader nearly verbatim).

3. Comment points at the wrong file — the docstring cites "gateway/run.py's user-friendly substitution", but the sentinel replacement lives in agent/turn_finalizer.py's completion explainer. Since the whole guarantee here is "match whatever the agent loop emits", name the authoritative site precisely (and its exact-comparison semantics) so a future widening there triggers someone to update this mirror.

[teknium1]

Merged via #96239 (rebase-merge) — your normalization fix is on main as-authored: a588685. Thanks @RibatTRW — you were the earliest submitter on the sentinel-protocol seam (00:08!), and your append-time rewrite of the exact (empty) sentinel to the gateway's friendly notice is the backstop half of the consolidated fix (landed together with #94386's selection half). Live-verified: the literal (empty) bubble on main renders as the friendly warning on current main.