RikkuBETA

Update 2.84

6/11/2026

v2.84.0BETAMINOR

Rikku

Full UI responsiveness: fluid frame + app-wide sweep of "forced sizes" across every window + reusable breakpoint infra

Delivers the owner's order: "no more forced sizes, everything responsive and auto-resizing, beauty over performance, push the cost to the user's PC". New infra: Styles/Responsive.cs — a Responsive.Breakpoints attached property that tags the root with narrow/medium/wide classes based on the REAL rendered width, making Style Selector work like a CSS media query. Frame (Main_Window): fixed 290/320px sidebars → Auto columns with MinWidth/MaxWidth; the right inspector collapses (Auto column → 0) when the window isn't wide, keeping 3 columns only on wide screens; MinWidth 1280→1000; top-bar fixed-width ComboBox → MinWidth; breakpoints active. App-wide sweep (65 of 71 .axaml adjusted via multi-agent orchestration — one agent per file, SURGICAL layout-only edits): pixel ColumnDefinitions 137→59 (−78); StackPanel Orientation="Horizontal" → WrapPanel (×26; 252 total) so button/input rows wrap instead of clipping; safety-net ScrollViewer where the root could overflow; Viewbox on SphereGridCanvas; WebView hosts (*Embedded) stretched; TextWrapping where missing; fixed Width/Height 231→211 and 35→27 (the rest are intrinsic: icons, numeric fields, DataGrid columns, and alignment widths inside an already-reflowing WrapPanel — they don't clip). (new responsiveness infra + cross-cutting UI capability). Honesty: all 75 screens compile (Release build 0 errors; & / < / > escaping guaranteed by the green build) and no {Binding}/Click/x:Name was changed (layout-only edits) — but per-screen visual QA at small window sizes is a pending human step (I did not run the app across all 75 screens; structure/compile proven, appearance not). Does not touch logic/save/AI/runtime/RT2/Spira Forge. Files: Styles/Responsive.cs (new) + Modules/Main/Main_Window.axaml + 65 module/control/template .axaml

v2.83.0

v2.84.10BETAPATCH

Lulu

Monster AI Editor: consolidation of 10 small bug/UX/guardrail patches, no new feature

RT2/user attempts showed the human Monster AI surface needed contract fixes, not another tab. This cut removes the Monster AI Editor 2 sub-tab from navigation and concentrates authoring back into the main Monster AI Editor; makes large blocks collapsible; relocates AEON/YUNALESCA/BIBLE/Live/actions for human reading; fixes local Maybe insertion so it no longer repoints the whole entrypoint; adds controlled Stop here/RET; demotes false combo badges and uses direct-sequence language when no native combo is proven; prevents forced from appearing just because chance was selected; clears YUNALESCA condition leakage; switches NulAll to the real spell; blocks start-of-battle spell casts where the runtime does not enqueue an action; improves condition/worker labels; fixes linked target/first ability/same-target flows; adds Shred-style computed target recipes + lowest-HP target; extends AiScriptLab --ai3 gates for local chance, linked rite, target recipe, computed target, and stop-after; keeps real blockers for natural Multi-* + Forbidden Rite and pending RT2 semantics. Honesty: this is bugfix/hardening for the existing Monster AI Editor, not a new writer outside scope

v2.84.0

v2.84.20BETAPATCH

Lulu

Monster AI Editor: 10 more bugfix patches for Forbidden Rite/Anti-Ribbon and human behavior reading

fixes the Forbidden Rite LAB surface that still exposed only a subset of what the real writeChrProperty(target, field, value) pattern already supports. The Anti-Ribbon dropdown now lists direct negative fields: Poison, Petrify, Power Break, Magic Break, Armor Break, Mental Break, Berserk, Sleep(255), Slow(255), plus advanced Provoke/Threaten, while keeping Zombie, Confuse, Silence(255), Darkness(255), Curse, Doom, and Doom counters. Zombie/Confuse/Silence/Darkness are explicitly marked as RT2/in-game proven in the current pattern; the expansion is recorded as operator-tested in-game through the same direct route. Death/KO stays out of the normal dropdown because the btlActorProperty map exposes isAlive, not a clean StatusDeath; that needs a separate recipe/gate. BIBLE and the SIN/Anti-Ribbon plan were updated so the next agent does not reopen a gate that is already solved. This also versions the recent design notes around Multi-* vs Sequential-*, dynamic cycles, Dark Flan, and Seymour Natus as future-design knowledge. Honesty: bugfix/hardening for the existing Monster AI Editor; no new writer outside the Forbidden Rite LAB

v2.84.10

v2.84.20.1BETAREVISION

Lulu

Research: feasibility of creating NEW Auto-Abilities (weapon/armor abilities) — ALL methods + the byte-recipe formula cookbook

(4th digit: doc/RE/research that enters the repo and does NOT change behavior — no writer/module/binding touched). Multi-agent research (3 workflows with parallel finders + adversarial verification) consolidating whether/how new auto-abilities can be created on FFX HD PC. Verdict: YES, with 3 senses of "new" and one hard split — DATA-DRIVEN vs HARDCODED-BY-ID. (1) Effects via flag combinations are data-driven and proven by offset in IDA (sub_79C610/sub_7861B0 OR-accumulate the 0x6C record fields into the actor with no per-ID switch) → any slot inherits the effect just by setting bytes; full byte cookbook decoded by diffing the shipped a_ability.bin (elements 0x11-0x15 Fire01/Ice02/Thunder04/Water08/Holy10; status touch=0x16+slot=50/strike=100; stat% 0x55+0x56; auto/SOS 0x5A+0x10; Ribbon 0x3C+slot=255+0x60). (2) 5 genuinely free slots 129–133 (0x81-0x85) (all-zero data) → repurpose without growing the table. (3) Grow >0x86: the surprise — the engine ACCEPTS a 135th entry (sub_7AB890 reads count/range from the file header, no 0x86 literal; buffer sized from file length); blocked only by 3 C# guards + RT2 + a byte-size cap 0x25800 + a 2nd equip-menu RAM structure capped at 0xC8=200. Hard limit: ~31 abilities are hardcoded-by-id (one bit in 0x62/0x64/0x66, handler in C — Sensor/First Strike/Counter/Piercing/Break Limits/Capture/etc.) → not reproducible by bytes; a new KIND of effect (lifesteal/reflect%) needs a DLL hook (the DINPUT8 probe + FfxHooksDll PolyHook2 infra is already proven) or an exe patch. ability-as-script (ATEL) refuted as a hook-free path for a player-equippable ability (binding gap). Cross-validated by 3 independent streams + the open-source fahrenheit struct (byte-for-byte). Honesty: real gameplay of any edited slot and all runtime/hook work = RT2/LAB pending; a_ability.bin stays reader+no-edit guard in production (no ship authorization). Docs: docs/reverse/FFX_AUTOABILITIES_NOVAS_MASTER_2026-06-11.md (index/verdict) + FFX_NEW_AUTO_ABILITY_FEASIBILITY_2026-06-11.md (static) + FFX_AUTO_ABILITY_NEW_VIA_HOOK_DLL_RUNTIME_FEASIBILITY_2026-06-11.md (runtime/DLL) + FFX_AUTOABILITIES_NOVAS_VIAS_E_FORMULAS_COOKBOOK_2026-06-11.md (12 methods + cookbook)

v2.84.20

v2.84.25BETAPATCH

Lulu

Monster AI Editor: YUNALESCA is now a clean preset area and conditions moved into the main authoring flow

fixes the screen UX/contract that still mixed quick rites, conditions, and advanced editing in the same block. YUNALESCA now stays only as ready-made status/blessing presets, without combat rites, without Alternate 1/2, without Alternate 1/3, without Duplicate cast, and without the redundant condition preview. The condition builder moved to Add / change behavior, the editor's main authoring surface: onTurn, Start, Always, onHit, Any Hit, HP < %, and Maybe 1 in K now prepare the next created/changed behavior. The main toolbar is focused only on the selected action (Change ability, Duplicate action, Move up/down, Mode: queue/now, Remove action, Undo backup). Backend paths now support explicit guards when creating an ability, copying an action template, and linking Forbidden Rite under the mounted condition, without presenting this as a new writer outside the Monster AI Editor. Forbidden Rite copy was corrected to tell the truth: it applies Anti-Ribbon right after the created/changed behavior at the authored point; no false warning that Multi-* necessarily breaks. Separate-output build passed with 0 errors

v2.84.20.1

Back to updates

RikkuBETA

Update 2.84

6/11/2026

v2.84.0BETAMINOR

Rikku

Full UI responsiveness: fluid frame + app-wide sweep of "forced sizes" across every window + reusable breakpoint infra

Delivers the owner's order: "no more forced sizes, everything responsive and auto-resizing, beauty over performance, push the cost to the user's PC". New infra: Styles/Responsive.cs — a Responsive.Breakpoints attached property that tags the root with narrow/medium/wide classes based on the REAL rendered width, making Style Selector work like a CSS media query. Frame (Main_Window): fixed 290/320px sidebars → Auto columns with MinWidth/MaxWidth; the right inspector collapses (Auto column → 0) when the window isn't wide, keeping 3 columns only on wide screens; MinWidth 1280→1000; top-bar fixed-width ComboBox → MinWidth; breakpoints active. App-wide sweep (65 of 71 .axaml adjusted via multi-agent orchestration — one agent per file, SURGICAL layout-only edits): pixel ColumnDefinitions 137→59 (−78); StackPanel Orientation="Horizontal" → WrapPanel (×26; 252 total) so button/input rows wrap instead of clipping; safety-net ScrollViewer where the root could overflow; Viewbox on SphereGridCanvas; WebView hosts (*Embedded) stretched; TextWrapping where missing; fixed Width/Height 231→211 and 35→27 (the rest are intrinsic: icons, numeric fields, DataGrid columns, and alignment widths inside an already-reflowing WrapPanel — they don't clip). (new responsiveness infra + cross-cutting UI capability). Honesty: all 75 screens compile (Release build 0 errors; & / < / > escaping guaranteed by the green build) and no {Binding}/Click/x:Name was changed (layout-only edits) — but per-screen visual QA at small window sizes is a pending human step (I did not run the app across all 75 screens; structure/compile proven, appearance not). Does not touch logic/save/AI/runtime/RT2/Spira Forge. Files: Styles/Responsive.cs (new) + Modules/Main/Main_Window.axaml + 65 module/control/template .axaml

v2.83.0

v2.84.10BETAPATCH

Lulu

Monster AI Editor: consolidation of 10 small bug/UX/guardrail patches, no new feature

RT2/user attempts showed the human Monster AI surface needed contract fixes, not another tab. This cut removes the Monster AI Editor 2 sub-tab from navigation and concentrates authoring back into the main Monster AI Editor; makes large blocks collapsible; relocates AEON/YUNALESCA/BIBLE/Live/actions for human reading; fixes local Maybe insertion so it no longer repoints the whole entrypoint; adds controlled Stop here/RET; demotes false combo badges and uses direct-sequence language when no native combo is proven; prevents forced from appearing just because chance was selected; clears YUNALESCA condition leakage; switches NulAll to the real spell; blocks start-of-battle spell casts where the runtime does not enqueue an action; improves condition/worker labels; fixes linked target/first ability/same-target flows; adds Shred-style computed target recipes + lowest-HP target; extends AiScriptLab --ai3 gates for local chance, linked rite, target recipe, computed target, and stop-after; keeps real blockers for natural Multi-* + Forbidden Rite and pending RT2 semantics. Honesty: this is bugfix/hardening for the existing Monster AI Editor, not a new writer outside scope

v2.84.0

v2.84.20BETAPATCH

Lulu

Monster AI Editor: 10 more bugfix patches for Forbidden Rite/Anti-Ribbon and human behavior reading

fixes the Forbidden Rite LAB surface that still exposed only a subset of what the real writeChrProperty(target, field, value) pattern already supports. The Anti-Ribbon dropdown now lists direct negative fields: Poison, Petrify, Power Break, Magic Break, Armor Break, Mental Break, Berserk, Sleep(255), Slow(255), plus advanced Provoke/Threaten, while keeping Zombie, Confuse, Silence(255), Darkness(255), Curse, Doom, and Doom counters. Zombie/Confuse/Silence/Darkness are explicitly marked as RT2/in-game proven in the current pattern; the expansion is recorded as operator-tested in-game through the same direct route. Death/KO stays out of the normal dropdown because the btlActorProperty map exposes isAlive, not a clean StatusDeath; that needs a separate recipe/gate. BIBLE and the SIN/Anti-Ribbon plan were updated so the next agent does not reopen a gate that is already solved. This also versions the recent design notes around Multi-* vs Sequential-*, dynamic cycles, Dark Flan, and Seymour Natus as future-design knowledge. Honesty: bugfix/hardening for the existing Monster AI Editor; no new writer outside the Forbidden Rite LAB

v2.84.10

v2.84.20.1BETAREVISION

Lulu

Research: feasibility of creating NEW Auto-Abilities (weapon/armor abilities) — ALL methods + the byte-recipe formula cookbook

(4th digit: doc/RE/research that enters the repo and does NOT change behavior — no writer/module/binding touched). Multi-agent research (3 workflows with parallel finders + adversarial verification) consolidating whether/how new auto-abilities can be created on FFX HD PC. Verdict: YES, with 3 senses of "new" and one hard split — DATA-DRIVEN vs HARDCODED-BY-ID. (1) Effects via flag combinations are data-driven and proven by offset in IDA (sub_79C610/sub_7861B0 OR-accumulate the 0x6C record fields into the actor with no per-ID switch) → any slot inherits the effect just by setting bytes; full byte cookbook decoded by diffing the shipped a_ability.bin (elements 0x11-0x15 Fire01/Ice02/Thunder04/Water08/Holy10; status touch=0x16+slot=50/strike=100; stat% 0x55+0x56; auto/SOS 0x5A+0x10; Ribbon 0x3C+slot=255+0x60). (2) 5 genuinely free slots 129–133 (0x81-0x85) (all-zero data) → repurpose without growing the table. (3) Grow >0x86: the surprise — the engine ACCEPTS a 135th entry (sub_7AB890 reads count/range from the file header, no 0x86 literal; buffer sized from file length); blocked only by 3 C# guards + RT2 + a byte-size cap 0x25800 + a 2nd equip-menu RAM structure capped at 0xC8=200. Hard limit: ~31 abilities are hardcoded-by-id (one bit in 0x62/0x64/0x66, handler in C — Sensor/First Strike/Counter/Piercing/Break Limits/Capture/etc.) → not reproducible by bytes; a new KIND of effect (lifesteal/reflect%) needs a DLL hook (the DINPUT8 probe + FfxHooksDll PolyHook2 infra is already proven) or an exe patch. ability-as-script (ATEL) refuted as a hook-free path for a player-equippable ability (binding gap). Cross-validated by 3 independent streams + the open-source fahrenheit struct (byte-for-byte). Honesty: real gameplay of any edited slot and all runtime/hook work = RT2/LAB pending; a_ability.bin stays reader+no-edit guard in production (no ship authorization). Docs: docs/reverse/FFX_AUTOABILITIES_NOVAS_MASTER_2026-06-11.md (index/verdict) + FFX_NEW_AUTO_ABILITY_FEASIBILITY_2026-06-11.md (static) + FFX_AUTO_ABILITY_NEW_VIA_HOOK_DLL_RUNTIME_FEASIBILITY_2026-06-11.md (runtime/DLL) + FFX_AUTOABILITIES_NOVAS_VIAS_E_FORMULAS_COOKBOOK_2026-06-11.md (12 methods + cookbook)

v2.84.20

v2.84.25BETAPATCH

Lulu

Monster AI Editor: YUNALESCA is now a clean preset area and conditions moved into the main authoring flow

fixes the screen UX/contract that still mixed quick rites, conditions, and advanced editing in the same block. YUNALESCA now stays only as ready-made status/blessing presets, without combat rites, without Alternate 1/2, without Alternate 1/3, without Duplicate cast, and without the redundant condition preview. The condition builder moved to Add / change behavior, the editor's main authoring surface: onTurn, Start, Always, onHit, Any Hit, HP < %, and Maybe 1 in K now prepare the next created/changed behavior. The main toolbar is focused only on the selected action (Change ability, Duplicate action, Move up/down, Mode: queue/now, Remove action, Undo backup). Backend paths now support explicit guards when creating an ability, copying an action template, and linking Forbidden Rite under the mounted condition, without presenting this as a new writer outside the Monster AI Editor. Forbidden Rite copy was corrected to tell the truth: it applies Anti-Ribbon right after the created/changed behavior at the authored point; no false warning that Multi-* necessarily breaks. Separate-output build passed with 0 errors

v2.84.20.1