QMD Plus
by @thehappyboy
Enhanced QMD search with LLM-powered query expansion. Better recall and precision for multilingual knowledge bases. Use when users ask to search notes, find...
clawhub install qmd-plus📖 About This Skill
name: qmd_plus description: Enhanced QMD search with LLM-powered query expansion. Better recall and precision for multilingual knowledge bases. Use when users ask to search notes, find documents, or look up information with intelligent query expansion. license: MIT compatibility: Requires qmd CLI or MCP server. Install via
npm install -g @tobilu/qmd.
metadata:
author: 虾爬爬 (based on qmd by tobi)
version: "1.0.0"
slug: qmd-plus
allowed-tools: Bash(qmd:*), Bash(node), Bash(jq), Bash(npx), Bash(your-llm-cli), mcp__qmd__*
⚠️ 安全与隐私说明
重要: 本技能需要将查询发送到外部 LLM 进行扩展。这意味着:
1. 你的查询内容会暴露给 LLM 提供商 — 包括搜索关键词、collection 名称等 2. 不要搜索敏感内容 — 避免查询包含密码、API 密钥、私人数据等 3. LLM 响应需要人工确认 — 在安装前查看元数据/集成不一致
运行时指令:
npx @anthropic/claude-code、kimi 等)数据流:
用户查询 → 本地生成提示词 → 外部 LLM → JSON 响应 → 本地构造 qmd 查询 → 本地搜索
建议: 在生产环境使用前,审查 scripts/expand-query.js 和 scripts/qmd-query-llm.sh 了解数据如何发送。
QMD Plus - QMD with LLM Query Expansion
Enhanced local search engine for markdown content with intelligent LLM-powered query expansion.
Install
# Install via ClawHub
clawhub install qmd-plusOr clone manually
git clone ~/workspace/skills/qmd_plus
Status
!qmd status 2>/dev/null || echo "Not installed: npm install -g @tobilu/qmd"
MCP: query
{
"searches": [
{ "type": "lex", "query": "CAP theorem consistency" },
{ "type": "vec", "query": "tradeoff between consistency and availability" }
],
"collections": ["docs"],
"limit": 10
}
Query Types
| Type | Method | Input |
|------|--------|-------|
| lex | BM25 | Keywords — exact terms, names, code |
| vec | Vector | Question — natural language |
| hyde | Vector | Answer — hypothetical result (50-100 words) |
Writing Good Queries
lex (keyword)
"connection pool" (quoted)performance -sports (minus prefix)handleError asyncvec (semantic)
"how does the rate limiter handle burst traffic""in the payment service, how are refunds processed"hyde (hypothetical document)
expand (auto-expand)
expand: question on its own lineexpand: with other typed lines — it's either a standalone expand query or a full query documentIntent (Disambiguation)
When a query term is ambiguous, add intent to steer results:
{
"searches": [
{ "type": "lex", "query": "performance" }
],
"intent": "web page load times and Core Web Vitals"
}
Intent affects expansion, reranking, chunk selection, and snippet extraction. It does not search on its own — it's a steering signal that disambiguates queries like "performance" (web-perf vs team health vs fitness).
Combining Types
| Goal | Approach |
|------|----------|
| Know exact terms | lex only |
| Don't know vocabulary | Use a single-line query (implicit expand:) or vec |
| Best recall | lex + vec |
| Complex topic | lex + vec + hyde |
| Ambiguous query | Add intent to any combination above |
First query gets 2x weight in fusion — put your best guess first.
Lex Query Syntax
| Syntax | Meaning | Example |
|--------|---------|---------|
| term | Prefix match | perf matches "performance" |
| "phrase" | Exact phrase | "rate limiter" |
| -term | Exclude | performance -sports |
Note: -term only works in lex queries, not vec/hyde.
Collection Filtering
{ "collections": ["docs"] } // Single
{ "collections": ["docs", "notes"] } // Multiple (OR)
Omit to search all collections.
Other MCP Tools
| Tool | Use |
|------|-----|
| get | Retrieve doc by path or #docid |
| multi_get | Retrieve multiple by glob/list |
| status | Collections and health |
CLI
qmd query "question" # Auto-expand + rerank
qmd query $'lex: X\nvec: Y' # Structured
qmd query $'expand: question' # Explicit expand
qmd query --json --explain "q" # Show score traces (RRF + rerank blend)
qmd search "keywords" # BM25 only (no LLM)
qmd get "#abc123" # By docid
qmd multi-get "journals/2026-*.md" -l 40 # Batch pull snippets by glob
qmd multi-get notes/foo.md,notes/bar.md # Comma-separated list, preserves order
HTTP API
curl -X POST http://localhost:8181/query \
-H "Content-Type: application/json" \
-d '{"searches": [{"type": "lex", "query": "test"}]}'
Setup
npm install -g @tobilu/qmd
qmd collection add ~/notes --name notes
qmd embed
🚀 LLM 查询扩展(QMD Plus)
使用外部 LLM 生成更高质量的查询变体,替代内置的 expand 功能。
为什么用 LLM 扩展?
| 内置 expand | LLM 扩展 | |------------|---------| | lex 扩展质量不稳定 | 术语更准确 | | hyde 固定英文 | 可指定语言 | | 无法利用上下文 | 可结合笔记内容 | | 模板化生成 | 智能语义理解 |
快速使用
假设 skill 安装在 ~/workspace/skills/qmd_plus/:
# 方式 1:wrapper 脚本生成提示词
~/workspace/skills/qmd_plus/scripts/qmd-query-llm.sh "汽车测试流程" -c memory-root-main -l zh方式 2:wrapper 脚本执行(传入 LLM 响应)
~/workspace/skills/qmd_plus/scripts/qmd-query-llm.sh --response '{"lex":[...],"vec":[...]}' -c memory-root-main方式 3:手动扩展 + 搜索
node ~/workspace/skills/qmd_plus/scripts/expand-query.js "汽车测试流程" zh
→ 复制 LLM 输出的 lex/vec → 构造 qmd query
添加到 PATH(可选)
# 在 ~/.zshrc 或 ~/.bashrc 中添加:
export PATH="$HOME/workspace/skills/qmd_plus/scripts:$PATH"然后可以直接使用:
qmd-query-llm "汽车测试流程" -c memory-root-main -l zh
qmd-query-llm 命令
自动完成:LLM 扩展 → 构造查询 → 执行搜索 → 返回结果
# 生成 LLM 提示词(Mode 1)
qmd-query-llm "汽车测试流程" -c memory-root-main -l zh执行搜索(Mode 2,传入 LLM JSON 响应)
qmd-query-llm --response '{"lex":["汽车测试","整车试验"],"vec":["测试流程是什么"]}' -c memory-root-main显示评分详情
qmd-query-llm --response '' -c memory-root-main --explain
脚本直接使用
# 生成 LLM 提示词
node expand-query.js "汽车测试流程" zh输出示例:
你是一个专业的知识库搜索查询优化器...
(将上述提示词发送给 LLM,获取 JSON 响应)
LLM 响应格式
{
"lex": ["汽车测试", "整车试验", "VTS 验证"],
"vec": ["汽车测试流程是什么样的", "整车试验包括哪些步骤"]
}
构造 qmd 查询
# 将 LLM 响应转换为 qmd query 格式
qmd query $'lex: 汽车测试\nlex: 整车试验\nvec: 汽车测试流程是什么样的' -c memory-root-main
代码示例
#!/bin/bash
qmd-query-llm wrapper
QUERY="$1"
COLLECTION="${2:-.openclaw}"
LANG="${3:-auto}"
Step 1: Generate LLM prompt
PROMPT=$(node scripts/expand-query.js "$QUERY" "$LANG")Step 2: Call LLM (implement according to your LLM provider)
⚠️ 注意:查询会发送到外部 LLM,避免敏感内容
RESPONSE=$(your-llm-cli "$PROMPT")Step 3: Parse and execute qmd query
LEX=$(echo "$RESPONSE" | jq -r '.lex[]' | sed 's/^/lex: /')
VEC=$(echo "$RESPONSE" | jq -r '.vec[]' | sed 's/^/vec: /')qmd query "$(echo -e "$LEX\n$VEC")" -c "$COLLECTION"
LLM 集成示例
使用 Kimi:
RESPONSE=$(kimi --prompt "$PROMPT")
使用 Claude Code:
RESPONSE=$(npx @anthropic/claude-code --prompt "$PROMPT" --max-tokens 1000)
使用 OpenClaw 内置模型:
RESPONSE=$(openclaw run --model modelstudio/qwen3.5-plus --prompt "$PROMPT")
最佳实践
1. 中文笔记用中文扩展 — 避免跨语言损失 2. 术语用 lex,概念用 vec — 组合使用效果最好 3. 指定 collection — 缩小搜索范围提高准确度 4. --explain 调试 — 查看哪个变体匹配到了结果
⚙️ Configuration
npm install -g @tobilu/qmd
qmd collection add ~/notes --name notes
qmd embed