Update 2.24

v2.24.0BETASeymour

Monster AI Editor: 1-click AUTOMATIONS (the software's "AI" assembles the bytecode)

• The owner asked for layperson buttons where the editor assembles the ATEL bytecode without breaking + validates before saving (the editor was great at editing, bad at authoring from scratch). New "✨ Automations (1-click)" card at the top of the module, with ➕ Add ability (friendly spell/ability dropdown via AiCommandId + "when": Always / Sometimes 1-in-K → the AI auto-picks the combat worker + entrypoint, grows the jump-table via AppendGuardedAction, validates and saves grow-aware) and ➖ Remove an action (lists the command actions the monster ALREADY performs, decoded by name; removes the stack-neutral [target][command][CALLPOPA] triplet via Rebuild + relink, the validator blocks the dangle case). + a "Technical mode (real names)" toggle that relabels the list with raw mnemonic/hex (for nerdier users). Free-edit (AI Assembler + Behavior Library) stays intact as advanced mode. Pure logic extracted to FfxLib/Ai/AiAutomation.cs (no editor deps, like AiCommandId/AiSnippetLibrary) → testable and gateable. New gate AiScriptLab --ai3 proves over the corpus (361 m*.bin, 346 with script): add-ability 346/346 + rng-guard 346/346 (auto-pick validates + re-parses clean as a grow), detect 1354 actions / 255 scripts, remove 241/241 (Rebuild shrinks by exactly the dropped triplet, re-parses clean) + 14 correctly blocked by the validator (something branches to the action). --ai2 still PASS. Build 0 errors. Honest: structure proven OFFLINE; the in-game behaviour of an ADDED ability depends on WHICH entrypoint runs per turn — NOT byte-proven (only the structure is, 692/692), so the entrypoint auto-pick is a heuristic (largest-span), reported transparently and flagged experimental/RT2 (confirm via the DINPUT8 probe). Multi-cast/multi-target/HP%-turn conditionals remain pending (corpus/anchor)

v2.23.3