dsh-mattpocock-skills
在终端中运行以下命令:
dsh plugin install GongYuanCaiJi/dsh-mattpocock-skills
将以下提示词粘贴到 DeepSeek Harness 对话框中:
在 DeepSeek Harness 中执行 dsh plugin install GongYuanCaiJi/dsh-mattpocock-skills 即可安装,源码地址为 https://github.com/GongYuanCaiJi/dsh-mattpocock-skills
插件介绍
The most common pitfall of AI-assisted coding is the lack of structured engineering discipline: the model either guesses or waits for you to re-prompt every round. This plugin ports 25 battle-tested engineering and productivity skills verbatim into DeepSeek Harness, covering TDD red-green loops, two-axis code review, stress-test questioning, requirement-to-spec and requirement-to-ticket workflows, domain modeling, codebase design, and bug diagnosis on the engineering side, plus grilling, teach, handoff, and writing-for-agents on the productivity side. The model loads a full structured workflow when a task matches, instead of starting from scratch every turn.
All 76 skill files are byte-identical to the upstream release, with hashes pinned and verifiable via a simple diff command. There are zero runtime dependencies and no build scripts, and skills load lazily so there is no boot-time cost. You can let the model auto-match a skill during a session, or invoke one explicitly by starting a message with the skill name.
This is for developers who treat dsh as a daily engineering companion and want the AI to behave like a disciplined team member rather than a chat partner. If you just want a ready-to-use set of engineering workflows without worrying about upstream versioning details, this plugin is all you need.
使用场景
- 开发时用 /tdd 驱动模型按红绿循环写测试和实现
- 提交前用 /code-review 让模型执行双轴审查
- 需求阶段用 /to-spec 或 /to-tickets 把想法拆成规格和票
适合人员
- 把 AI 作为日常工程伙伴的全栈开发者
- 希望在编码流程中引入结构化工具链的团队
- 喜欢 Matt Pocock 工程方法论的独立开发者