btspoony/mstar-harness
453Last commit Aug 16, 2026
mstar-harness DSH plugin
Morning Star is a TypeScript Harness Workflow Engine that enforces deterministic workflow gates through @mstar-harness/engine, while mstar-* skills provide role, gate, and workflow judgment. It runs across multiple hosts including dsh, omp, OpenCode, Cursor, Kimi Code, ZCode, and Codex.
How to install the mstar-harness DSH plugin
dsh plugin --profile web add @mstar-harness/dshCopying does not run this command. Review the repository and version before installing the mstar-harness DSH plugin.
mstar-harness DSH plugin data source
mstar-harness DSH plugin snapshot date: Aug 16, 2026
discovered
What the mstar-harness DSH plugin can do
- Deterministic workflow gates enforced by a TypeScript engine (path, status, lease, dispatch, SDD, iteration, lint)
- Cross-host support: works with dsh, omp, OpenCode, Cursor, Kimi, ZCode, and Codex
- Iteration workflow: Phase 1–5 lifecycle with interactive or autonomous modes
- Codebase audit: read-only survey generating prioritized improvement plans
- Portable Agent Plugins v1.0.0 manifest with plugin.json and skills/ directory
Where the mstar-harness DSH plugin fits
- Multi-phase iterative development with structured planning, execution, QC, and QA gates
- Codebase audit to discover improvement opportunities without modifying source
- Single-plan hotfix or feature development with deterministic gate checks
- Multi-agent code delivery with role-based assignment (PM, architect, dev, QA, etc.)
Who the mstar-harness DSH plugin is for
- Developers using DeepSeek Harness (dsh) who want a structured workflow engine
- Teams that need deterministic, multi-agent code delivery with role-based gates
mstar-harness DSH plugin limitations
- On dsh, installation requires the host-specific command (dsh plugin --profile web add @mstar-harness/dsh) rather than the generic CLI init
- PM skill must be manually loaded on most hosts (e.g., via /pm, /skill:pm, or agent selection)
- Recommended host compatibility order is provided (dsh = omp ≥ OpenCode ≥ Cursor > Kimi = ZCode > Codex), indicating varying levels of support
- The project is tagged with prerelease versions and may still be in early development
mstar-harness DSH plugin: from the repository README
Quoted from the btspoony/mstar-harness README, the upstream source of the mstar-harness DSH plugin. Copyright remains with the original authors.
Harness Workflow Engine · Agent Plugin English / [中文](README_CN.md) <a href="https://github.com/btspoony/mstar-harness">GitHub</a> · <a href="https://github.com/btspoony/mstar-harness/issues">Issues</a> [](https://github.com/btspoony/mstar-harness/actions/workflows/ci.yml) [](LICENSE) [](https://github.com/btspoony/mstar-harness/releases) [](https://github.com/btspoony/mstar-harness/commits/main) [](https://dshfind.com/zh/plugins/btspoony/mstar-harness?ref=badge) [](https://whyihaveyou.github.io/dsh
Read the full READMERepository license: MIT
mstar-harness DSH plugin questions
How do I install Morning Star on DeepSeek Harness?
Do not use the generic CLI init command. Instead, run `dsh plugin --profile web add @mstar-harness/dsh` in your terminal. After installation, reload the host (e.g., restart or use the reload command if available).
What hosts does Morning Star support?
Morning Star supports dsh, omp, OpenCode, Cursor, Kimi Code, ZCode, and Codex. The recommended host order is dsh = omp ≥ OpenCode ≥ Cursor > Kimi = ZCode > Codex, but all are usable.
How do I start a workflow iteration?
Use the `/iteration-start` command in dsh, omp, or OpenCode hosts. This initiates a Phase 1 interactive grill-me session, then automatically continues through Phases 2–5. You can also use `/iteration-loop` for a fully autonomous run.
Can I use Morning Star for a simple hotfix without iteration?
Yes. Enter the Project Manager (PM) role by using `/pm` (or equivalent for your host), then follow the per-plan cycle: Prepare → Execute → QC → QA gate → Done. No iteration wrapper is needed.
What is the codebase audit feature?
The `/codebase-audit` command performs a read-only survey of your repository. It generates prioritized, self-contained improvement plans saved to `{PLAN_DIR}/audit-<date>/`. You can refine the scope with keywords like `quick`, `deep`, `security`, `perf`, or `simplify`.