Loping151

Port a new text-to-image model architecture (a new DiT/UNet base model that Forge/A1111 or diffusers cannot load natively) into this project — first as a standalone inference pipeline to verify correctness, then as a native Forge diffusion engine so the checkpoint loads and generates in the WebUI/API. Use when asked to "support / 兼容 / 整合 a new model architecture", run a model whose baseModel is unfamiliar (e.g. a Cosmos/Lumina/Qwen-Image/Z-Image/Anima-style DiT), or reproduce a ComfyUI-only model in plain diffusers/torch.

Write or adapt a Forge / A1111 (Stable-Diffusion-WebUI) extension — a script under extensions/ that adds UI and hooks into the generation pipeline (callbacks, attention/UNet monkeypatching). Use when asked to build a new extension, modify one like the regional prompter, or make an attention-based extension work across model architectures (UNet vs DiT).

鸣潮·当期逆境深塔·深境区 / 冥歌海墟 / 矩阵叠兵三大周期玩法配队顾问。用户问「本期深塔/海墟/矩阵怎么打 / 怎么配队 / 用啥角色」时调用，整合本期 buff、敌人抗性、用户练度，输出可执行的队伍方案。

Delegate full-stack field/feature delivery to StackWeft — a compiled, evidence-gated pipeline for React + Express + Sequelize repos. Use when the user wants to ADD or EXTEND a cross-stack field (Sequelize model → controller / API payload → frontend service → list-card AND detail-page DOM) and have it verified by counterexample sentinel probes. Cheaper and steadier than doing it yourself in a free agentic loop, and it remembers repeated work (0-LLM replay). Do NOT use for non-field work (algorithms, infra, cross-service refactors) or repos that don't fit this full-stack field-flow shape — handle those yourself.