dictcli
개인 어휘 그래프 — 한↔영 크로스링귀얼 쿼리 확장 + 한국어 형태소 분석. expand로 한글 단어에서 영어 태그 후보를 찾고, stem으로 한국어 어간을 추출한다. knowledge_search의 3층 역할. '단어 확장', '태그 찾기', '한영 매핑', 'expand', 'stem', '형태소'.
来源信息
- 仓库
- junghan0611/agent-config
- 最近来源活动
- 2026年9月3日 02:04
- 检测到的 SKILL.md 语言
- 韩语
- 星标
- 5
- 分支
- 0
安装方式
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
检查来源文件
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
文件资源管理器
2 个文件正在显示 SKILL.md
SKILL.md
来源说明 · 只读预览- name
- dictcli
- description
- 개인 어휘 그래프 — 한↔영 크로스링귀얼 쿼리 확장 + 한국어 형태소 분석. expand로 한글 단어에서 영어 태그 후보를 찾고, stem으로 한국어 어간을 추출한다. knowledge_search의 3층 역할. '단어 확장', '태그 찾기', '한영 매핑', 'expand', 'stem', '형태소'.
# dictcli — Personal Vocabulary Graph (Layer 3)
Korean↔English triple graph. Expands Korean words to English Denote tags.
Layer 3 of knowledge_search — cross-lingual query enrichment.
## API
Binary bundled in skill directory. **Must cd to {baseDir}** (reads CWD/graph.edn).
| Command | Usage | Output |
|---------|-------|--------|
| **expand** | `cd {baseDir} && ./dictcli expand "보편" --json` | `["universal","universalism","particular","paideia"]` |
| graph | `cd {baseDir} && ./dictcli graph "보편"` | All triples for the word |
| stats | `cd {baseDir} && ./dictcli stats` | Graph statistics |
| validate | `cd {baseDir} && ./dictcli validate` | Invariant check |
### expand — Core Command
Traverses the graph: direct translation → opposite's translation → related word's translation.
```bash
cd {baseDir} && ./dictcli expand "기술" --json # → ["art","technology","technique"]
cd {baseDir} && ./dictcli expand "도피" --json # → ["escape","flight","avoidance","evasion"]
```
Empty array = word not in graph.edn. Fall back to original query.
### knowledge_search Integration Pattern
```
1. User: "보편 학문 관련 노트"
2. expand "보편" → ["universal","universalism","paideia"...]
3. Search: "보편 학문 universal universalism paideia"
```
Agent autonomously: extract concept words → expand → enrich query → knowledge_search.
## stem — Korean Morphological Analysis
**JVM-only** (Kiwi JNI — not in native binary). Called via dictcli repo's `run.sh stem`.
Agents rarely call directly — andenken batch-indexes via `--batch` / `--serve` mode.
```bash
cd ~/repos/gh/dictcli && ./run.sh stem "설계했다" # → 설계
cd ~/repos/gh/dictcli && ./run.sh stem --serve 9876 # socket server (1ms/query)
```
Pipeline: `"설계했다" → stem → "설계" → expand → ["design","architecture"]`
## Data
- `graph.edn`: 3,989 triples, 2,453 `:trans` mappings, 4,735 words, 526 clusters
- Sources: meta-note clusters, Syntopicon 102 Great Ideas, philosophy glossary
- Relations: `:trans`(2453), `:source`(1422), `:related`(58), `:synonym`(44)
## Architecture Note
| Component | Runtime | Latency |
|-----------|---------|---------|
| expand/lookup/stats | GraalVM native-image | ~9ms |
| stem | JVM (Kiwi JNI) | ~1ms (server) / ~3s (cold start) |
Native binary cannot include stem (JNI incompatible with native-image).
Two separate execution paths — no conflict.
## Layer 3 in the Stack
| Layer | Tool | Role |
|-------|------|------|
| 1 | knowledge_search | Embedding vector search |
| 2 | denotecli | Exact match + graph links |
| **3** | **dictcli expand + stem** | **Korean→English query expansion + stemming** |
在 GitHub 查看