Agent skill
mysearch
DEFAULT search skill for OpenClaw. Aggregates Tavily, Firecrawl, Exa, and optional X/social search behind one search path. Use for ANY external lookup that needs current web results, docs, GitHub, changelog, pricing, URL extraction, or X discussion. Prefer this over legacy Tavily-only skill or raw web_search when MySearch is healthy.
Install this agent skill to your Project
npx add-skill https://github.com/skernelx/MySearch-Proxy/tree/main/openclaw
Metadata
Additional technical details for this skill
- clawdbot
-
{ "emoji": "\ud83d\udd0e", "requires": { "bins": [ "bash", "python3" ] } } - openclaw
-
{ "tags": [ "search", "web", "docs", "tavily", "firecrawl", "exa", "social" ], "emoji": "\ud83d\udd0e", "requires": { "env": [ "MYSEARCH_PROXY_BASE_URL", "MYSEARCH_PROXY_API_KEY", "MYSEARCH_TAVILY_API_KEY", "MYSEARCH_FIRECRAWL_API_KEY" ], "bins": [ "bash", "python3" ] }, "primaryEnv": "MYSEARCH_PROXY_API_KEY" }
SKILL.md
MySearch For OpenClaw
MySearch 是给 OpenClaw 用的默认搜索 skill。
如果你是 AI 助手,并且用户只是给了仓库地址或 openclaw/ 目录:
- 先打开
openclaw/README.md - 先按
README完成安装与验收 - 再回到这个
SKILL.md执行搜索规则和调用策略
它把 Tavily、Firecrawl、Exa、X / Social 聚合成同一个入口,并且按任务类型自动路由:
- 最新网页、普通实时搜索:优先 Tavily
- Tavily 结果不够时,网页发现可补 Exa
- 文档、GitHub、pricing、changelog、PDF:优先 Firecrawl
- X / Twitter / 社交舆情:优先 xAI 或 compatible
/social/search - 单页正文抓取:优先 Firecrawl,失败或空正文时回退 Tavily extract
最小配置
Hub 版 skill 已经自带 runtime,不需要在安装时再下载远程代码。
推荐最小配置:
MYSEARCH_PROXY_BASE_URLMYSEARCH_PROXY_API_KEY
这两项配好后:
Tavily / Firecrawl / Exa会默认都走统一 proxy- 如果 proxy 同时接通了
Social / X,这一套 token 也会继续复用 - OpenClaw 侧不需要再分别维护三套 provider token
兼容旧接法时,仍可直接填:
MYSEARCH_TAVILY_API_KEYMYSEARCH_FIRECRAWL_API_KEY
可选增强:
MYSEARCH_XAI_API_KEYMYSEARCH_XAI_BASE_URLMYSEARCH_XAI_SOCIAL_BASE_URLMYSEARCH_XAI_SEARCH_MODE=official|compatible
如果没有 X / Social 配置,MySearch 仍然可以正常完成:
webnewsdocsgithubpdfextractresearch
只有 mode="social" 或 --include-social 才会要求 X / Social。
OpenClaw 配置建议
优先把统一 proxy 配进 OpenClaw skill env,而不是到处复制 provider key 或 shell 环境。
MYSEARCH_PROXY_BASE_URL 只应该指向你自己部署或明确可信的 proxy。
mysearch_openclaw.py 会优先读取 openclaw.json 里的
skills.entries.mysearch.env,正式部署不需要依赖 .env。
{
"skills": {
"entries": {
"mysearch": {
"enabled": true,
"env": {
"MYSEARCH_PROXY_BASE_URL": "https://search.hunters.works",
"MYSEARCH_PROXY_API_KEY": "mysp-..."
}
}
}
}
}
如果你暂时没有统一 proxy,再退回旧接法:
{
"skills": {
"entries": {
"mysearch": {
"enabled": true,
"env": {
"MYSEARCH_TAVILY_API_KEY": "tvly-...",
"MYSEARCH_FIRECRAWL_API_KEY": "fc-..."
}
}
}
}
}
只有在你直接调试这个仓库工作树时,才建议复制 .env.example 到本地 .env。
Hub 安装或正式 OpenClaw 部署优先用上面的 skill env 注入,不要默认把 secrets
复制进已安装的 skill 目录。
本地调试示例:
cp {baseDir}/.env.example {baseDir}/.env
python3 {baseDir}/scripts/mysearch_openclaw.py health
如果要把 skill 复制到别的 OpenClaw skills 目录,再执行:
bash {baseDir}/scripts/install_openclaw_skill.sh --install-to ~/.openclaw/skills/mysearch
MySearch-First 规则
只要 health 显示至少有可用搜索 provider:
- 外部搜索任务优先走 MySearch
- 不要把 raw
web_search当主流程 - 不要优先走旧的 Tavily-only skill
只有这些情况才回退:
- MySearch 还没配置最小 key
- 用户明确要求改用别的搜索方式
- MySearch 返回冲突结果,需要额外复核
严格参数规则
search / research 的 mode 只允许:
autowebnewssocialdocsresearchgithubpdf
禁止事项:
- 不要发明
mode="hybrid" hybrid只是某些返回结果形态,不是可传参数- 同时要网页和 X 时,优先:
--sources web,x- 或拆成
social + news
常用命令
健康检查
python3 {baseDir}/scripts/mysearch_openclaw.py health
普通网页搜索
python3 {baseDir}/scripts/mysearch_openclaw.py search \
--query "best search MCP server" \
--mode web
今天 X 上在热议什么
python3 {baseDir}/scripts/mysearch_openclaw.py search \
--query "today's biggest stories on X" \
--mode social \
--intent status
规则:
- 先
social - 不要先跑
news - 不要先混用 raw
web_search
今天 X 热议 + 网页新闻一起对照
单次:
python3 {baseDir}/scripts/mysearch_openclaw.py search \
--query "today's biggest stories on X" \
--sources web,x \
--intent status \
--strategy verify
或者双次:
python3 {baseDir}/scripts/mysearch_openclaw.py search --query "..." --mode social --intent status
python3 {baseDir}/scripts/mysearch_openclaw.py search --query "..." --mode news --intent status
输出时必须区分:
- X 上在热议什么
- 网页新闻在报道什么
文档 / GitHub / pricing / changelog
python3 {baseDir}/scripts/mysearch_openclaw.py search \
--query "OpenAI responses API pricing" \
--mode docs \
--intent resource
抓正文
python3 {baseDir}/scripts/mysearch_openclaw.py extract \
--url "https://example.com/post"
小型研究包
python3 {baseDir}/scripts/mysearch_openclaw.py research \
--query "best search MCP server 2026" \
--intent exploratory \
--include-social
输出要求
- 优先给结论,再给来源
- 保留 URL
- 区分事实、引文和推断
- 同时包含网页和 X 时,明确分区,不要混成一句模糊总结
max_results默认保持小一些,先拿 3 到 5 条
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
mysearch
Install, verify, debug, and use MySearch MCP/Skill. Aggregates Tavily, Firecrawl, Exa, and X search (via xAI) behind one workflow. Use when the user shares a MySearch repo/skill URL, wants MySearch installed or repaired, or wants current web search, social/X search, document-oriented search, URL content extraction, or a small multi-source research pack with citations. When MySearch is available, prefer it over generic web search for external lookup tasks.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
gguf-quantization
GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.
Claude Code Guide
Master guide for using Claude Code effectively. Includes configuration templates, prompting strategies "Thinking" keywords, debugging techniques, and best practices for interacting with the agent.
qdrant-vector-search
High-performance vector similarity search engine for RAG and semantic search. Use when building production RAG systems requiring fast nearest neighbor search, hybrid search with filtering, or scalable vector storage with Rust-powered performance.
Didn't find tool you were looking for?