一个 MIT 协议的 Agent Plugin，在 DeepSeek Harness 上跑确定性、带关卡强制的多智能体工作流——一行 `npx` 命令即可安装，dsh 为最优宿主，并随附可移植的 Agent Plugins v1.0.0 包。
DSH 适配
兼容
作者声明
安全审计
未审计
最后核验
2026-09-02
许可证
MIT
01它能帮你完成什么？
Run a deterministic, spec-driven multi-agent code-delivery workflow on DeepSeek Harness — from a product brief through plan, implement, QC/QA gates, to a merge-ready PR.
Git-branched code changes plus tracked plans (`status.json`), workflow snapshots, and a project residual register under `.mstar/`; the audit path emits prioritized, ready-to-execute improvement plans.
Engineering teams and solo developers running DeepSeek Harness (dsh) who want enforced workflow gates and multi-agent delivery across dsh / omp / OpenCode / Cursor / Codex hosts.
Audit an existing codebase or run a deep pre-merge PR review inside DeepSeek Harness without editing any source.
Read-only survey of what's worth doing (prioritized plans) or a single verdict (`ship it` / `needs fixes` / `blocked`) with every finding, optionally posted to GitHub by the review agent.
Maintainers and reviewers on DeepSeek Harness who need an advisory, no-write pass (codebase audit or PR review at quick / default / deep strength).
02如何接入 DeepSeek Harness？
前置条件
- DeepSeek Harness (dsh) CLI on PATH — the dsh install path runs `dsh plugin --profile web add @mstar-harness/dsh`.
- A Node.js runtime with npx/npm — the one-command install is `npx @mstar-harness/cli init --target dsh`; the engine check installs via `npm i -g @mstar-harness/cli`.
- An Agent Plugins v1.0.0 conformant client — the repo ships a portable `plugin.json` + `skills/` package usable by any compliant host.
安装步骤
- 01
One-command install for dsh: `npx @mstar-harness/cli init --target dsh` — this runs two independent `dsh plugin --profile web add` installs (`@mstar-harness/dsh` + `dsh-llm-fallbacks`; add `--no-fallbacks` to skip the latter).
$ npx @mstar-harness/cli init --target dsh
- 02
Or install the two packages manually: `dsh plugin --profile web add @mstar-harness/dsh` then `dsh plugin --profile web add dsh-llm-fallbacks`.
$ dsh plugin --profile web add @mstar-harness/dsh
- 03
For engine gate checks, install the CLI globally: `npm i -g @mstar-harness/cli` (puts the `mstar-harness` binary / alias `mstar` on PATH); `init` auto-installs it after a successful run unless you pass `--no-global-cli`.
$ npm i -g @mstar-harness/cli
验证接入成功
- `npx @mstar-harness/cli doctor --target dsh` — verifies the install for the dsh target.
- `npx @mstar-harness/cli plugin validate` — validates the portable Agent Plugins v1.0.0 manifest (`plugin.json`) and the `skills/` component.
03DSH 适配与能力边界
Installs on DeepSeek Harness (dsh) via `dsh plugin --profile web add @mstar-harness/dsh` (plus `dsh-llm-fallbacks`); the one-command `npx @mstar-harness/cli init --target dsh` also works. dsh is a first-class, best-tier host of the shared TS engine.
Deterministic workflow gates (TS engine)
per-plan dispatch requests and workflow state→enforced path/status/lease/dispatch/sdd/iteration/lint gates via `@mstar-harness/engine`
Pluggable JSON persistence (ArtifactStore)
coordination docs (status.json, workflow snapshots, residuals, review envelopes)→persisted artifacts through an ArtifactStore; default FsStore keeps `.mstar/` paths
coordination docs (`status.json`, workflow snapshots, project residuals, review envelopes) persist through an `ArtifactStore`; the default `FsStore` keeps the existing `.mstar/` pathsMulti-agent delivery (mstar-* skills)
a product brief / plan→role-routed multi-agent code delivery driven by `mstar-*` judgment skills (single source of truth for roles, gates, workflow)
Read-only audit & review (mstar-audit)
a codebase, PR, branch, or diff→read-only prioritized improvement plans, or a PR review verdict with findings
04适合谁？何时不该用？
适合
- Engineering teams and solo developers running DeepSeek Harness (dsh) who want enforced workflow gates and multi-agent delivery across dsh / omp / OpenCode / Cursor / Codex hosts.
- Maintainers and reviewers on DeepSeek Harness who need an advisory, no-write pass (codebase audit or PR review at quick / default / deep strength).
不适合
- The short `mstar` alias is a shared bin namespace — an unrelated third-party npm package named `mstar` claims the same command name, and globally co-installing both silently overwrites the `mstar` shim (last install wins). Use the long name `mstar-harness` on any conflict.
05兼容性、维护与安全提示
- The short `mstar` alias is a shared bin namespace — an unrelated third-party npm package named `mstar` claims the same command name, and globally co-installing both silently overwrites the `mstar` shim (last install wins). Use the long name `mstar-harness` on any conflict.
- Process artifacts (`plans/`, `iterations/`, `status.json`, `workflows/`, `projects/`, `sdd/`, …) are gitignored; only `{HARNESS_DIR}/AGENTS.md`, `knowledge/`, and `specs/` are tracked results.
- Without a global CLI install the harness still works and gate checks stay advisory; set `enforcement: hard` in an iteration compass to make dispatch preflights fail-fast.
MIT; actively maintained — latest release v3.5.1 (2026-08-28), last push 2026-09-01 (per repo metadata)
06常见问题
怎么在 DeepSeek Harness 上安装 Morning Star？
执行 `npx @mstar-harness/cli init --target dsh`——它会运行两次独立的 `dsh plugin --profile web add` 安装（`@mstar-harness/dsh` + `dsh-llm-fallbacks`；加 `--no-fallbacks` 可跳过后者）。也可手动添加：`dsh plugin --profile web add @mstar-harness/dsh`，再 `dsh plugin --profile web add dsh-llm-fallbacks`。
DeepSeek Harness 是一等宿主还是仅兼容？
dsh 是一等、最优宿主。README 把推荐宿主排为 `dsh = omp ≥ OpenCode ≥ Cursor > Kimi = ZCode > Codex`，同一套 TS 引擎加 `mstar-*` 技能既驱动 dsh，也驱动 omp、OpenCode、Cursor、Kimi Code、ZCode 和 Codex。
它会改我的代码吗？能不能只读运行？
两种都行。常规流程在特性分支上派发计划并产出可合并的 PR；而审计/评审命令（`/codebase-audit`、`/amazing-pr-review`）是只读、建议性的——绝不改源码，发现项会变成可选择性执行的计划。
怎么验证安装是否成功？
运行 `npx @mstar-harness/cli doctor --target dsh` 检查 dsh 目标；再用 `npx @mstar-harness/cli plugin validate` 校验可移植的 `plugin.json` 清单与 `skills/` 组件。
工作流产物存在哪里？
协调文档——`status.json`、工作流快照、项目残留、评审信封——通过可插拔的 `ArtifactStore` 持久化；默认 `FsStore` 存在 `.mstar/` 下。过程产物被 gitignore，只有 `AGENTS.md`、`knowledge/` 与 `specs/` 会纳入跟踪。
07相关的 DSH 工作流
archify
作者 tt-a1i
为编码智能体生成美观可验证的架构图、时序图与数据流图，输出自包含 HTML，支持动效与清晰导出。
openviking
作者 volcengine
为 AI 智能体打造的自进化上下文数据库，统一智能体记忆、知识 RAG 与技能。
nocobase
作者 nocobase
开源的 AI + 无代码应用搭建平台：AI 在久经生产验证的底层基础设施与所见即所得的无代码界面之上工作，帮你快速构建 CRM、ERP 等业务系统，兼顾速度与可靠性。
learn-harness-engineering
作者 walkinglabs
Harness 工程新手教程，从 0 到 1 系统学习智能体工作流框架。
08数据与来源
the same engine + skills power dsh (DeepSeek Harness), omp, OpenCode, Cursor, Kimi Code, ZCode, and Codex
or `dsh plugin --profile web add @mstar-harness/dsh`<br>+ `dsh plugin --profile web add dsh-llm-fallbacks`
The repo ships a portable **Agent Plugins v1.0.0** manifest (`plugin.json`) at its root; `skills/` is the Agent Skills c…
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-09-02。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。