Source-available DSH short-drama workbench: infinite canvas, 1–30 episode generation, deterministic checks, shootability reports, per-episode sign-off and a release gate — all local, experimental status.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-11
License
PolyForm-Noncommercial-1.0.0
01What can it help you accomplish?
Write a 1–30 episode short drama from a one-line story to exportable scripts on a DSH canvas
A canvas with nine node types, a batch generation pipeline (beats → scene list → script → deterministic checks → QC/compliance), pre-generation snapshots and docx / md / txt export
Short-drama writers and small studios that need batch drafts plus a quality gate
Decide whether the finished episodes are actually shootable
A per-episode shootability report — estimated duration with ±25% tolerance, scene count, speaking roles, end-of-episode hook, canon membership — with sign-off bound to the script version and a release gate
Producers and lead writers who must sign off episode by episode before release
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness >= 0.2.1-rc.5 — all three bundles declare `0.2.1-rc.5+` in `dsh.compatibility`
- Node.js `^22.19.0 || >=24.0.0` for the DSH CLI and source builds
- Optional Python 3 with python-docx for docx export; without it the export degrades to md and says why
Installation steps
- 01
Clone the repo and run `bash dsh/profile/install.sh` — it creates the `moxi` profile from the `web` template and mounts the three bundles with `dsh plugin add`
$ dsh plugin add
- 02
Optionally install the creation skill with `MOXI_INSTALL_SKILL=1 bash dsh/profile/install.sh` (existing skills are backed up, not overwritten)
- 03
Start with `dsh --profile moxi`
$ dsh --profile moxi
Verify the integration
- The terminal prints a tokenised `http://127.0.0.1:<port>/?token=...` address; opening it shows 「墨戏画布」 and 「墨戏后台」 in the sidebar
- Sidebar → 「墨戏画布」 → 「+ 新开一部剧」 fills in title, episode count, audience, genre and one-line story
Rollback
- `bash dsh/profile/install.sh --uninstall` for a source install, or `dsh plugin --profile moxi-exp remove moxi-core-bundle moxi-canvas-bundle moxi-admin-bundle` for the prebuilt package
- Uninstalling never deletes project data — it stays in `~/Documents/Moxi/projects`
03DSH integration and capability boundaries
Three DSH bundles in a dedicated profile: `moxi-core` host plugin plus `moxi-canvas` and `moxi-admin` client plugins
Infinite canvas
one drama project→nine node types (project root, setting collection, assets, episode beats, scene list, script scenes, QC, compliance, storyboard/export) with drag, zoom, links, rename, undo/redo and persisted viewport
Batch episode generation
a start/end episode range→episodes generated through the beats → scenes → script → deterministic checks → QC/compliance pipeline, with live progress, cancel, resume and breakpoint recovery
one task is capped at 30 episodesa snapshot is taken before generation startsShootability report and sign-off
generated episodes→per-episode estimated duration, scene count, speaking roles, end hook, canon membership and role distribution; each episode can be signed off
sign-off is bound to the script version, so editing invalidates itRelease gate
sign-off state and hook detection→releases blocked when an episode is unsigned, its sign-off is stale or the end hook is missing; forced release requires a written reason and is audited
forced releases are recorded in the audit event streamAdmin console (5 pages)
projects, rules, models and logs→workspace KPIs, projects, rules and skills, models and usage with a link probe, and logs with range export
Local data with audit trail
every write operation→append-only `history/events.jsonl`, WAL write transactions that can roll back, and soft deletes recoverable from `.trash/`
all project data lives on the local disk under `~/Documents/Moxi/projects/`
04Who is it for? When not to use it?
Good for
- Short-drama writers and small studios that need batch drafts plus a quality gate
- Producers and lead writers who must sign off episode by episode before release
Not for
- PolyForm Noncommercial 1.0.0: free for personal study, research and non-commercial creation; any commercial use needs written authorisation.
- The project is explicitly experimental — interfaces, data formats and UI may change, so back up project data before upgrading.
- Shootability duration and hooks are estimates and text-feature detection; they cannot replace real timing, location and cast feasibility judgement.
- AI compliance review is advisory and does not participate in the release gate; red lines come from a deterministic rule table and other high-risk items need human confirmation.
- Scale was not stress-tested at 100 episodes (a 30-episode, 157-node project runs smoothly), so very long projects may stutter.
05Compatibility, maintenance and safety notes
- PolyForm Noncommercial 1.0.0: free for personal study, research and non-commercial creation; any commercial use needs written authorisation.
- The project is explicitly experimental — interfaces, data formats and UI may change, so back up project data before upgrading.
- Shootability duration and hooks are estimates and text-feature detection; they cannot replace real timing, location and cast feasibility judgement.
- AI compliance review is advisory and does not participate in the release gate; red lines come from a deterministic rule table and other high-risk items need human confirmation.
- Scale was not stress-tested at 100 episodes (a 30-episode, 157-node project runs smoothly), so very long projects may stutter.
- The Windows install script and paths are compatibility-handled but were not fully tested on a real Windows machine; issues should come with `moxi-diagnose.txt`.
- There is no cloud backend, database service or telemetry; model calls reuse DSH's llm service, keys stay in the DSH profile and the plugin neither stores keys nor sends HTTP itself.
PolyForm-Noncommercial-1.0.0 · README evidence only, no tagged release in the facts pack, safety not audited
06Frequently asked questions
How do I install Moxi on DeepSeek Harness?
Clone the repository and run `bash dsh/profile/install.sh`: it creates a dedicated `moxi` profile from the `web` template and mounts `moxi-core`, `moxi-canvas` and `moxi-admin` with `dsh plugin add`, without touching your `desktop` profile. Start it with `dsh --profile moxi` and open the tokenised address the terminal prints. Requirements are DSH >= 0.2.1-rc.5 and Node.js ^22.19.0 or >=24.
Does the model decide whether an episode passes?
No. The split is deliberate: the model writes and scores, deterministic scripts judge word counts, scene continuity, scene-header format, duplicate paragraphs, compliance red lines, duration and hooks, and a human signs off each episode. The release gate blocks unsigned episodes, stale sign-offs and missing end hooks; forcing past it requires a written reason and is audited.
Can I use it commercially?
Not without authorisation. The license is PolyForm Noncommercial 1.0.0: personal study, research and non-commercial creation are free, while any commercial use — client work, SaaS, paid courses, internal company use, resale — needs written authorisation from the author.
Where is my project data?
Under `~/Documents/Moxi/projects/` by default, as Markdown, JSON and JSONL. There is no cloud backend, database service or telemetry; write operations append to `history/events.jsonl`, transactions use WAL, and deletes are soft deletes recoverable from `.trash/`. Model calls reuse DSH's llm service, so keys stay in the DSH profile.
Is it stable enough for production?
It is explicitly experimental: interfaces, data formats and UI may change, so back up project data before upgrading. Known limits include no 100-episode stress test (30 episodes / 157 nodes runs smoothly), advisory-only AI compliance review, and a Windows install path that was compatibility-handled but not fully tested on a real Windows machine.
07Related DSH workflows
huashu-design
by alchaincyf
Creates selectable design directions and delivers interactive prototypes, PPTX, animations, and visual assets.
dsh-tui
by ccch1mneyyy
Claude Code 风格全屏终端 TUI：像素鲸鱼顶栏、双流光大字、实时工作状态行、思考流式展开、双击 Esc 时间回溯、上下文进度条 + TPS 仪表；接入 Agent preset（standard/code/minimal/cordis）、Skills、MCP、Goals、Todos、子代理，/ 命令注册表与随附技能（audit/bug/review/practice/pr_comments/release-notes/vuln-check）；纯插件挂载零核心改动。
dsh-better-sidebar
by omdsh-dev
Full sidebar workbench with file rendering and editing, terminal, Git, and subagents; third-party plugins can register new tabs.
dsh-media-desk
by woniuniuniu
Capture ideas and generate scripts, titles, covers, images, storyboards, schedules, and review data in a sidebar media workspace.
08Data and sources
DSH 是 DeepSeek 的本地 Agent 运行时（「一切皆插件」），墨戏生花以三个插件的形式接入它。
| `moxi-core` | Host 插件（无 UI） | 项目 / 节点 / 连线的唯一读写入口；状态机、审计、确定性检查、导出、可拍性报告、放行闸门；对 Agent 暴露 15 个工具，对前端提供 36 条 `/api/moxi/*…
安装**不会动** `desktop` profile，你现有的 DSH 界面不受影响。
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-11. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.