Cross-platform resource search orchestrator for AI Agents. Search 15+ sources in one query — find Skills, MCP Servers, Prompt templates, and open-source projects.
- 15 search sources — Skills.sh, GitHub, Prompts.chat (MCP), ClawHub, SkillHub, MCPServers.org, and more
- 3-tier search strategy — Tier 1 (CLI/MCP) → Tier 2 (WebSearch) → Tier 3 (broad search), converges fast
- Pre-install security scanning — deterministic code scanning + LLM context evaluation, quantified score 0-100
- Natural language trigger — just say "find me a xxx tool", no commands to memorize
- Extensible registry — discover and add new sources on the fly
User query
↓
Step 1: Intent classification → skill / mcp / prompt / repo
Step 2: Read registry.json, filter available sources
Step 3: Tier 1 parallel search (CLI + MCP simultaneously)
Step 4: Enough results? (≥3 high-relevance → skip to 6)
Step 5: Tier 2/3 supplementary search
Step 6: Relevance evaluation
Step 7: Safety quick-check ([SAFE] / [CAUTION] / [RISK])
Step 8: Dedup + sort (relevance > safety > popularity)
Step 9: Display recommendations
Step 10: User action → "install" triggers deep security scan
Step 11: New site discovery → prompt to add to registry
| Source | Method | Categories |
|---|---|---|
| Skills.sh | npx skills find |
Skill |
| GitHub | gh search repos |
Repo / MCP / Skill |
| ClawHub | clawhub search |
Skill |
| Prompts.chat | MCP search_prompts |
Prompt |
| Prompts.chat | MCP search_skills |
Skill |
SkillHub, MCPServers.org, SkillsMP, Prompts.chat, ClawHub, AI Short, NanoPrompts, AI Art Pics, LocalBanana
# Install as a Claude Code skill
npx skills add ZadAnthony/find-everything --skill find-everything
# Or manually: symlink skill/ directory to your agent's skill path
ln -s /path/to/find-everything/skill ~/.claude/skills/find-everything/find-everything find a code review skill
/find-everything search aggregator MCP
/find-everything blockchain analyst prompt
Or just use natural language — "is there a tool for xxx", "find me a xxx".
Unlocks Tier 1 prompt search:
claude mcp add prompts-chat -- npx -y @fkadev/prompts.chat-mcpTwo layers of protection before installing any resource:
Layer 1 — Quick check (metadata-based)
- [SAFE]: from known platform + installs > 100
- [CAUTION]: low installs or unknown source
- [RISK]: suspected typosquat or red flags
Layer 2 — Deep scan (triggered on install)
security_scan.py detects:
- Prompt injection (15 patterns, including base64/URL-encoded/HTML-hidden)
- Dangerous commands (curl|sh, rm -rf, sudo, eval, etc.)
- Data exfiltration (curl POST, netcat, etc.)
- Credential access (.ssh, .aws, .env, etc.)
- Unicode homoglyph attacks
- Zero-width character injection
- Typosquat detection (edit distance + homoglyphs, against 200+ known skills)
Combined with security-checklist.md for LLM-assisted 5-dimension scoring (0-100):
source trust / code transparency / permission scope / network risk / community signals
find-everything/
├── skill/ # Skill package (what gets installed)
│ ├── SKILL.md # Core logic (11-step search orchestration)
│ ├── references/
│ │ ├── registry.json # Search source registry
│ │ ├── security-checklist.md # Security scoring rubric
│ │ └── known_skills.txt # Known skill names (for typosquat detection)
│ └── scripts/
│ └── security_scan.py # Deterministic security scanner
├── tests/ # Tests for security_scan.py
├── docs/ # Design documents
├── LICENSE
└── README.md
python -m pytest tests/ -v跨平台资源搜索编排器 for AI Agent。一句话搜遍 15 个源，找到你需要的 Skill、MCP Server、Prompt 模板或开源项目。
- 15 个搜索源 — Skills.sh、GitHub、Prompts.chat (MCP)、ClawHub、SkillHub、MCPServers.org 等
- 3 层搜索策略 — Tier 1 (CLI/MCP) → Tier 2 (WebSearch) → Tier 3 (广域搜索)，快速收敛
- 安装前安全扫描 — 确定性代码扫描 + LLM 上下文评估，量化评分 0-100
- 自然语言触发 — 说"找个 xxx 工具"自动搜索，无需记命令
- 注册表可扩展 — 发现新站点一键加入
用户查询
↓
Step 1: 意图分类 → skill / mcp / prompt / repo
Step 2: 读取 registry.json，筛选可用源
Step 3: Tier 1 并行搜索（CLI + MCP 同时跑）
Step 4: 结果够了吗？（≥3 条高匹配 → 跳到 6）
Step 5: Tier 2/3 补充搜索
Step 6: 相关性评估
Step 7: 安全快筛（[SAFE] / [CAUTION] / [RISK]）
Step 8: 去重 + 排序（相关度 > 安全 > 热度）
Step 9: 格式化展示推荐
Step 10: 用户操作 → "安装"触发深度安全扫描
Step 11: 新站点发现 → 提示加入注册表
| 源 | 方法 | 覆盖类别 |
|---|---|---|
| Skills.sh | npx skills find |
Skill |
| GitHub | gh search repos |
Repo / MCP / Skill |
| ClawHub | clawhub search |
Skill |
| Prompts.chat | MCP search_prompts |
Prompt |
| Prompts.chat | MCP search_skills |
Skill |
SkillHub、MCPServers.org、SkillsMP、Prompts.chat、ClawHub、AI Short、NanoPrompts、AI Art Pics、LocalBanana
# 作为 Claude Code skill 安装
npx skills add ZadAnthony/find-everything --skill find-everything
# 或手动：将 skill/ 目录链接到你的 agent skill 路径
ln -s /path/to/find-everything/skill ~/.claude/skills/find-everything/find-everything 找一个代码审查的 skill
/find-everything search aggregator MCP
/find-everything 区块链分析师 prompt
也可以自然语言触发："有没有 xxx 工具"、"帮我找个 xxx"。
解锁 Tier 1 级别的 prompt 搜索：
claude mcp add prompts-chat -- npx -y @fkadev/prompts.chat-mcp安装任何资源前有两层防护：
第一层 — 快筛（基于元数据）
- [SAFE]：来自已知平台 + 安装量 > 100
- [CAUTION]：安装量低或来源不明
- [RISK]：疑似 typosquat 或其他红旗
第二层 — 深度扫描（安装时触发）
security_scan.py 检测：
- Prompt 注入（15 种模式，含 base64/URL 编码/HTML 注释隐藏）
- 危险命令（curl|sh、rm -rf、sudo、eval 等）
- 数据外传（curl POST、netcat 等）
- 凭证访问（.ssh、.aws、.env 等）
- Unicode 同形字攻击
- 零宽字符注入
- Typosquat 检测（编辑距离 + 同形字，对比 200+ 已知 skill）
配合 security-checklist.md 做 LLM 辅助 5 维度评分（0-100）：
来源可信度 / 代码透明度 / 权限范围 / 网络风险 / 社区信号
find-everything/
├── skill/ # Skill 本体（安装到 agent 的内容）
│ ├── SKILL.md # 核心逻辑（11 步搜索编排流程）
│ ├── references/
│ │ ├── registry.json # 搜索源注册表
│ │ ├── security-checklist.md # 安全评估评分体系
│ │ └── known_skills.txt # 已知 skill 名单（typosquat 检测用）
│ └── scripts/
│ └── security_scan.py # 确定性安全扫描脚本
├── tests/ # 测试
├── docs/ # 设计文档
├── LICENSE
└── README.md
python -m pytest tests/ -vMIT