LuluBETA

Update 2.233

8/17/2026

v2.233.0.0BETAMINOR

LuluAug 16, 2026

BYOK AI Assistant: AiAssistant panel connecting the LLM contract (P2B) to UI + full i18n migration + nav wiring

New module FFXProjectEditor/Modules/AiAssistant/ (5 files): BYOK panel consuming the existing Core/LLM/ layer (P2B contract: LlmHttpAdapter + LlmGuard + LlmCredential + ProposalToPlanMapper). AiAssistantSession.cs (key in memory only, Dispose zeroizes, refuses remote host without human confirmation); AiAssistant_DataModel.cs (ObservableObject, Connect/Send/Clear, LlmHistoryEntry audit without payload, ApiKeyText zeroized via TakeApiKey); AiAssistant_Control.axaml (TextBox with PasswordChar, two-way binding, IRestorableModule); AiAssistantLabels.cs (forward to Strings, RejectReason helper). Security: BYOK in-memory + opt-in + loopback endpoint free, remote requires checkbox; no file writes.
Full i18n migration: 23 labels migrated from AiAssistantLabels (raw EN) → Strings.resx (EN) + Strings.pt.resx (PT-BR) + Strings.cs (props AiAssistant_*). Gate: StringsIntegrityTests 3/3 PASS.
Catalog/nav: ai-assistant entry in ModuleRegistry (official, Live cluster, RequiresProject: false) + module strings. The MenuItem_AiAssistant handler + Dispatch case in Main_Window are ready in the worktree but out of this commit: Main_Window.axaml.cs is interleaved with ~33 uncommitted i18n changes from another lane in the same file — coordinated commit when the i18n baseline lands.
Absorbed theory doc: docs/ai/BOT_MAECHEN_EDITOR_ABSORBED_2026-08-16.md (verified against real Core/LLM code).
Gates: build 0 errors; StringsIntegrityTests 3/3 PASS. Suite: 424/432 (8 pre-existing failures: 7 PtStrings mojibake + 1 MagicDll — none from this lane). Honesty: adapter expects PatchProposal JSON endpoint (raw chat API → MalformedPayload); end-to-end needs gateway/template from recipe expansion (next MINOR).

v2.232.1.0

v2.233.1.0BETAPATCH

SeymourAug 17, 2026

Aurora EditViewer restored: actor-overlay bug fixed (mis-nested installKeyboard broke boot → loadAnchors never ran) + markers made visible

Root cause (via Chrome DevTools console): installKeyboard was nested inside togglePick in RuntimeTools/FFXMapViewerWeb/aurora-overlay.js — never exposed at global scope, so buildPanel() threw ReferenceError: installKeyboard is not defined → interrupted boot() → loadAnchors never ran → the EditViewer opened the map with no anchor markers at all (no monster/party/camera spheres or pins; inert editing UI). Fixed by moving installKeyboard to global scope (dated comment in code).
Actor visibility: anchor markers were effectively invisible (radius-0.7 sphere in an ~80-unit map with camera at distance ~88 ≈ 1px). Raised to 4.0 + added a vertical cone pin per anchor (role color) so actors are unmistakable on the canvas; the pin follows the sphere on drag (3 points).
Dual validation (Chrome via DevTools + internal WebView2): panel now shows "8 âncora(s) · 5 monstro(s) · battle bika02_00"; Luna confirmed gold/orange (monsters) + blue/cyan (party) markers on the canvas on both targets.
Honest limit (separate lane-HD front): the monster HD models (/work/phyre_chr_anim/models/mNNN/mNNN_animated.gltf) do not exist on disk → loadMonster falls back to the orange degraded sphere with reason (404). The phyre_chr_anim pipeline remains empty (documented as broken in ModuleCatalogPolicy). Markers now serve as the editable placeholder until lane HD re-exports the glTFs.
Gates: node --check OK; editor build ok. No RT2 test (bug was purely in the web overlay UI). ### MINOR

v2.233.0.0

Back to updates

LuluBETA

Update 2.233

8/17/2026

v2.233.0.0BETAMINOR

LuluAug 16, 2026

BYOK AI Assistant: AiAssistant panel connecting the LLM contract (P2B) to UI + full i18n migration + nav wiring

New module FFXProjectEditor/Modules/AiAssistant/ (5 files): BYOK panel consuming the existing Core/LLM/ layer (P2B contract: LlmHttpAdapter + LlmGuard + LlmCredential + ProposalToPlanMapper). AiAssistantSession.cs (key in memory only, Dispose zeroizes, refuses remote host without human confirmation); AiAssistant_DataModel.cs (ObservableObject, Connect/Send/Clear, LlmHistoryEntry audit without payload, ApiKeyText zeroized via TakeApiKey); AiAssistant_Control.axaml (TextBox with PasswordChar, two-way binding, IRestorableModule); AiAssistantLabels.cs (forward to Strings, RejectReason helper). Security: BYOK in-memory + opt-in + loopback endpoint free, remote requires checkbox; no file writes.
Full i18n migration: 23 labels migrated from AiAssistantLabels (raw EN) → Strings.resx (EN) + Strings.pt.resx (PT-BR) + Strings.cs (props AiAssistant_*). Gate: StringsIntegrityTests 3/3 PASS.
Catalog/nav: ai-assistant entry in ModuleRegistry (official, Live cluster, RequiresProject: false) + module strings. The MenuItem_AiAssistant handler + Dispatch case in Main_Window are ready in the worktree but out of this commit: Main_Window.axaml.cs is interleaved with ~33 uncommitted i18n changes from another lane in the same file — coordinated commit when the i18n baseline lands.
Absorbed theory doc: docs/ai/BOT_MAECHEN_EDITOR_ABSORBED_2026-08-16.md (verified against real Core/LLM code).
Gates: build 0 errors; StringsIntegrityTests 3/3 PASS. Suite: 424/432 (8 pre-existing failures: 7 PtStrings mojibake + 1 MagicDll — none from this lane). Honesty: adapter expects PatchProposal JSON endpoint (raw chat API → MalformedPayload); end-to-end needs gateway/template from recipe expansion (next MINOR).

v2.232.1.0

v2.233.1.0BETAPATCH

SeymourAug 17, 2026

Aurora EditViewer restored: actor-overlay bug fixed (mis-nested installKeyboard broke boot → loadAnchors never ran) + markers made visible

Root cause (via Chrome DevTools console): installKeyboard was nested inside togglePick in RuntimeTools/FFXMapViewerWeb/aurora-overlay.js — never exposed at global scope, so buildPanel() threw ReferenceError: installKeyboard is not defined → interrupted boot() → loadAnchors never ran → the EditViewer opened the map with no anchor markers at all (no monster/party/camera spheres or pins; inert editing UI). Fixed by moving installKeyboard to global scope (dated comment in code).
Actor visibility: anchor markers were effectively invisible (radius-0.7 sphere in an ~80-unit map with camera at distance ~88 ≈ 1px). Raised to 4.0 + added a vertical cone pin per anchor (role color) so actors are unmistakable on the canvas; the pin follows the sphere on drag (3 points).
Dual validation (Chrome via DevTools + internal WebView2): panel now shows "8 âncora(s) · 5 monstro(s) · battle bika02_00"; Luna confirmed gold/orange (monsters) + blue/cyan (party) markers on the canvas on both targets.
Honest limit (separate lane-HD front): the monster HD models (/work/phyre_chr_anim/models/mNNN/mNNN_animated.gltf) do not exist on disk → loadMonster falls back to the orange degraded sphere with reason (404). The phyre_chr_anim pipeline remains empty (documented as broken in ModuleCatalogPolicy). Markers now serve as the editable placeholder until lane HD re-exports the glTFs.
Gates: node --check OK; editor build ok. No RT2 test (bug was purely in the web overlay UI). ### MINOR

v2.233.0.0