@jnmetacode

finishing-a-development-branch guides developers through merging, PR creation, or discarding completed work by validating tests, detecting environment state (worktree vs standard repo), and executing one of four structured options with proper cleanup.

chinese-commit-conventions adapts conventional commits for mandarin-speaking teams, providing standardized templates, tooling configs (commitlint, husky, commitizen), and changelog generation setup with bilingual examples.

mcp-builder documents systematic methodology for designing, implementing, testing and deploying model context protocol servers. covers tool/resource/prompt primitives, schema validation, error handling, testing patterns, security controls and deployment via npm/pip/docker.

dispatching-parallel-agents guides splitting independent failures across focused agents with isolated context, using decision trees and concrete examples to avoid parallel interference while maximizing concurrent troubleshooting throughput.

executing-plans provides a structured workflow for loading, reviewing, and executing multi-step implementation plans with checkpoints and validation gates. emphasizes critical review upfront and blocks progress on ambiguity.

using-git-worktrees provides a decision tree for isolating work through platform-native tools or git worktrees, starting with detection of existing isolation before tool selection.

subagent-driven-development delegates tasks to isolated sub-agents with two-phase review (spec compliance then code quality), aiming to preserve coordinator context and improve iteration speed through task isolation.

using-superpowers establishes a mandatory skill-lookup protocol requiring agents to invoke the skill tool before any response, enforcing procedural discipline across multi-vendor platforms with priority rules and chinese-language routing.

当创建新技能、编辑现有技能或在部署前验证技能是否有效时使用

国内 Git 平台配置参考——Gitee、Coding.net、极狐 GitLab、CNB 的 SSH/HTTPS/凭据/CI 接入差异与镜像同步配置。仅在用户显式 /chinese-git-workflow 时调用，不要根据上下文自动触发。

中文文档排版参考——中英文空格、全半角标点、术语保留、链接格式、中文文案排版指北约定。仅在用户显式 /chinese-documentation 时调用，不要根据上下文自动触发。

在 Claude Code / OpenClaw / Cursor 中直接运行 agency-orchestrator YAML 工作流——无需 API key，使用当前会话的 LLM 作为执行引擎。当用户提供 .yaml 工作流文件或要求多角色协作完成任务时触发。

当你有规格说明或需求用于多步骤任务时使用，在动手写代码之前

在任何创造性工作之前必须使用此技能——创建功能、构建组件、添加功能或修改行为。在实现之前先探索用户意图、需求和设计。

中文 review 沟通参考——话术模板、分级标注（必须修复/建议修改/仅供参考）、国内团队常见反模式应对。仅在用户显式 /chinese-code-review 时调用，不要根据上下文自动触发。

在实现任何功能或修复 bug 时使用，在编写实现代码之前

完成任务、实现重要功能或合并前使用，用于验证工作成果是否符合要求

遇到任何 bug、测试失败或异常行为时使用，在提出修复方案之前执行

收到代码审查反馈后、实施建议之前使用，尤其当反馈不明确或技术上有疑问时——需要技术严谨性和验证，而非敷衍附和或盲目执行

在宣称工作完成、已修复或测试通过之前使用，在提交或创建 PR 之前——必须运行验证命令并确认输出后才能声称成功；始终用证据支撑断言