dsh-pinyin-annotation
Pinyin annotation skill pack for DeepSeek Harness (DSH) — teacher-grade拼音, not machine transliteration.
DeepSeek Harness 拼音标注技能包 —— 面向国际中文教学的拼音注音,不是机器音译。
模型不缺"拼音怎么拼",缺的是"怎么给外国学生标注":分词连写的词界、 专名大写、儿化与隔音撇、成语半字线、轻声辨义——一处写错,学生就记成错的。 本技能包把 GB/T 16159-2012 正词法与 HSK 教学体例提炼成可执行的规则 + 分布检查清单,模型匹配后按同一口径产出。
Models can spell syllables; what they get wrong is how to annotate for learners — where the word boundaries go, which letters are capitalized, when
rfolds in, where the apostrophe belongs. One wrong word break and the student memorizes it. This pack turns GB/T 16159-2012 orthography plus HSK classroom conventions into executable rules and a pre-delivery checklist.
English | 中文
What it does / 做什么
Feed it Chinese, get back pinyin only — tone marks on every syllable, word groupings written together, proper nouns capitalized, half-width punctuation:
| Source / 原文 | Output / 输出 |
|---|---|
| 我们明天去北京语言大学。 | Wǒmen míngtiān qù Běijīng Yǔyán Dàxué. |
| 风平浪静 | fēngpíng-làngjìng |
| 第九 / 七八个 | dì-jiǔ / qī-bā gè |
| 花儿 / 皮袄 | huār / pi'ǎo |
No 汉字+pinyin side-by-side tables, no per-word explanations, no 。 or
, — the deliverable is plain pinyin text a learner can read aloud.
What's inside / 包含内容
| Skill | Covers / 覆盖 |
|---|---|
pinyin-annotation |
三条铁律、三步工作流(音节 → 分词 → 篇章)、教学口径默认值、必错清单、交付前自检 |
Four reference notes ship beside it and are loaded on demand, so the main
SKILL.md stays short:
| Reference | Topic / 主题 |
|---|---|
01-音节拼写与标调.md |
声母韵母、标调位置、ü、零声母、隔音符号、儿化、多音字 |
02-分词连写.md |
词界、重叠式、助词、动宾/动补、方位词、数词量词、成语、虚词 |
03-专名大写标点移行.md |
人名地名机构名大写、标点、移行、分段硬规则 |
04-两可边界与等同写法.md |
标准允许的两种写法、择一后全篇一致 |
Install / 安装
From npm / 从 npm:
dsh plugin --profile desktop add dsh-pinyin-annotation
Or from the GitHub tag / 或从 GitHub 的 tag:
dsh plugin --profile desktop add "git+https://github.com/mouse33333/dsh-pinyin-annotation.git#v1.0.2"
From a local checkout / 从本地目录:
dsh plugin --profile desktop add /path/to/dsh-pinyin-annotation
The bundle inserts one host plugin (dsh-pinyin-annotation) into the profile's
cordis.patch.yml. It needs nothing else — the base bundle already mounts the
skill registry it publishes to.
Usage / 使用
Just ask. The skill routes on its own description — 注音、标拼音、加拼音、拼音转写、 生词表/例句/课文的带拼音版本 all trigger it:
给这段课文标拼音:我们明天去北京语言大学。
把这 10 个生词做成带拼音的生词表。
Annotate this dialogue with pinyin for HSK 2 students.
How it works / 工作原理
index.js registers one ctx.skills provider and reads every
skills/<name>/SKILL.md from disk on each discovery pass. On top of that it
fs.watches the skills/ tree and calls the provider's invalidate(), so an
edited SKILL.md, a changed reference note, and a newly added or deleted skill
folder all take effect without restarting the harness:
export const inject = ['skills']
export function apply(ctx) {
ctx.skills.registerProvider(() => ({
name: 'dsh-pinyin-annotation',
async list(options) { /* … */ },
async get(candidate, options) { /* … */ },
}))
}
Each skill is published with source: 'bundled', rank: 600 (the same rank as
the shipped skill packs, so a same-name user skill still wins) and a
resourceBase pointing at its own directory — that is what makes
references/01-音节拼写与标调.md resolvable from inside the skill body.
Frontmatter follows the standard SKILL.md shape (name, description, optional
user-invocable / disable-model-invocation); description is the routing
text, so it leads with the concrete triggers.
Structure / 目录结构
dsh-pinyin-annotation/
├── index.js # bundle entry: registers the skills provider
├── cordis.patch.yml # inserts the plugin into the profile
├── package.json # dsh.bundle.patch points at cordis.patch.yml
├── icon.png
└── skills/
└── pinyin-annotation/
├── SKILL.md
└── references/
├── 01-音节拼写与标调.md
├── 02-分词连写.md
├── 03-专名大写标点移行.md
└── 04-两可边界与等同写法.md
Add your own skill / 添加自己的技能
Drop a skills/<your-skill>/SKILL.md in — the watcher picks it up, no code change
and no restart needed. The name must be kebab-case
(/^[a-z0-9]+(?:-[a-z0-9]+)*$/) and the frontmatter must carry name and
description. Duplicate names fail at load with the offending name, rather than
silently shadowing each other.
License / 许可
MIT. The skill content is an independent reimplementation of the rules in GB/T 16159-2012《汉语拼音正词法基本规则》and common HSK classroom practice. The entry point has no runtime dependencies — frontmatter is parsed by a small built-in reader rather than a YAML library.
No comments yet. Be the first to write one.