Add or update UniRL model package support. Use when adding diffusion or autoregressive model pipelines, model config dataclasses, Bundle/Pipeline/Stage/Conditions implementations, LoRA targets, FSDP wrapping hints, RolloutReq/RolloutResp plumbing, or multimodal text/image/video conditioning.
Create and repair UniRL pull requests. Use when creating a PR, editing a PR body, responding to PR Body or Semantic Pull Request CI failures, running gh pr create, or when the user mentions PR body, pull request template, body check, title check, or pull request formatting.
Create and repair UniRL pull requests. Use when creating a PR, editing a PR body, responding to PR Body or Semantic Pull Request CI failures, running gh pr create, or when the user mentions PR body, pull request template, body check, title check, or pull request formatting.
代码规范与完整开发工作流，涵盖 Python、PyTorch/深度学习、Shell 脚本、配置文件。贯穿始终的四条设计原则：易读直观、拒绝过度抽象、单一职责、命名自解释。强制三段式流程：（1）写代码前——在已有项目上开发时必须先了解项目全貌避免重复实现，然后先出 plan（思路+实现步骤）并获得用户批准才能开工；（2）写代码中——遵循命名/类型/docstring/张量操作/异常处理等规范，同时时刻守住四条设计原则；（3）写代码后——必须先补齐对应单元测试并本地跑通，再主动调用 simplify 简化新代码，最后调用 review 做审查。触发条件：用户要求写/生成/重构 Python 或 PyTorch 代码；用户要求在已有项目/代码库上开发；用户创建新模块、脚本或训练循环；用户说"按代码规范来"或引用 /code-standards。