← 返回 Skills 市场
223
总下载
0
收藏
1
当前安装
9
版本数
在 OpenClaw 中安装
/install novada-search
功能描述
AI Agent search platform with 9 engines, Google 13 sub-types, vertical scene search, and intelligent auto/multi/extract modes. Designed for LLM and AI agent...
安全使用建议
This package appears to be a coherent multi-engine search client that needs one API key (NOVADA_API_KEY) to call Novada's scraper API. Before installing, verify the API key source (novada.com) and that you trust the provider; confirm the registry metadata mismatch (some registry metadata shows no required env but SKILL.md/_meta.json require NOVADA_API_KEY). Review novada_search.py for any additional outbound hosts or unexpected behavior (especially if you plan to run the MCP server). Consider running the package in an isolated environment or container, and avoid reusing high-privilege credentials (use a limited/billing-restricted key). If you have doubts about the upstream project, check the referenced repository (github.com/NovadaLabs/novada-search) and the provider's documentation before supplying your API key.
功能分析
Type: OpenClaw Skill
Name: novada-search
Version: 1.0.8
The novada-search skill is a legitimate multi-engine search platform designed for AI agents. It provides structured access to various search engines (Google, Bing, etc.) via the Novada Scraper API. The code (primarily novada_search.py) is well-structured, includes extensive tests, and follows the declared permissions for network access to https://scraperapi.novada.com. No evidence of data exfiltration, malicious execution, or prompt injection was found.
能力评估
Purpose & Capability
The name/description (multi-engine search for agents) matches the included code (novada_search.py, SDK, CLI, MCP server, LangChain integration) and the only service credential (NOVADA_API_KEY) is appropriate for a scraper/search API. The declared network target (scraperapi.novada.com) fits the stated purpose.
Instruction Scope
SKILL.md describes running the CLI/SDK/MCP server and enumerates limited filesystem permissions (tool files, samples, tests) and a single network endpoint. The runtime instructions do not ask the agent to read arbitrary home directories or unrelated credentials (they explicitly state they won't scan home directories and only support CLI flag, NOVADA_API_KEY, or local .env).
Install Mechanism
There is no install spec provided by the registry (instruction-only), and project files are included. References to pip (pip install novada-search) and github are normal project metadata. Nothing downloads arbitrary code from unknown URLs during install.
Credentials
The runtime requires a single NOVADA_API_KEY which is proportional to a scraper/search API. However there is an inconsistency: the registry summary at the top of the evaluation said 'Required env vars: none' while SKILL.md and _meta.json declare NOVADA_API_KEY as required. Confirm which registry metadata is authoritative before installing.
Persistence & Privilege
The skill does not request always: true or system-wide persistent privileges. It exposes an MCP stdio server and a LangChain tool (normal for this type of integration). It does not modify other skills or request global configs.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install novada-search - 安装完成后，直接呼叫该 Skill 的名称或使用
/novada-search触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.8
No changes detected in this version.
- Version 1.0.8 released with no file updates or user-facing changes.
- Previous features and integrations remain the same.
- No new functionality, bug fixes, or documentation updates in this version.
v1.0.7
Novada-search 1.0.7 changelog:
- Added Python SDK with agent-friendly error handling and method interfaces.
- Introduced MCP server with skill/tools config (see `novada_mcp_server.py` and `mcp.json`).
- Provided LangChain integration (see `integrations/langchain_tool.py`).
- Expanded test coverage with new tests for engine params, error handling, extractors, and agent output.
- Added benchmarks and sample queries for verification and performance analysis.
- Included initial marketing materials and updated documentation for quickstart and agent API usage.
v1.0.6
- No code or documentation changes detected in this release.
- Version bump only; no updates to features, functionality, or documentation.
- All behavior and interface remain unchanged from the previous release.
v1.0.5
Novada Search v1.0.5 — Major agent-mode & output overhaul
- Added unified best-answer ranking across all engines in `agent-json` via new `unified_results`, including scores, agreement counts, and deduplication rationale.
- Improved multi-engine deduplication with aggressive URL normalization; number of duplicates removed now reported to agents.
- `agent-json` output is now explainable and ready for immediate LLM/AI agent use (with scoring and domain fields).
- API key handling is stricter and more flexible: accepts environment variable or CLI flag (home-directory scan removed).
- Introduced regression fixtures and ranking tests to ensure future upgrades don’t degrade answer quality.
- Added new sample JSON and ranking tests in `samples/` and `tests/` for reproducibility and CI.
v1.0.4
Novada Search v1.0.4 introduces improved documentation and troubleshooting guidance.
- Added a new Troubleshooting section in SKILL.md, including common pitfalls and solutions.
- Clarified that script entry point is now `{baseDir}/novada_search.py` (was previously under `scripts/`).
- Provided information about HTTP error handling, server IP issues, default dynamic fetch mode, and debugging options.
- Added an "Optional: AI Analysis" section with instructions on integrating your own LLM for post-search reasoning.
- Added _meta.json file for package metadata.
v1.0.3
Novada Search v2.0 — Docs overhaul and improved clarity
- Completely rewritten documentation for faster onboarding and better real-world examples, including output samples.
- Added practical table-format comparison of engines, Google sub-types, and scenes.
- Expanded usage instructions and included ready-to-copy command examples for every feature.
- Detailed output format options, including `enhanced`, `agent-json`, and others.
- General UX, automation, and integration tips for both human users and LLM/AI agent applications.
- No code changes in this release; documentation only.
v1.0.2
- Three files removed: _meta.json, manifest.yaml, and scripts/novada_search.py.
- Core executable and metadata files deleted, meaning command-line usage and runtime integration are no longer available in this skill version.
- Documentation and feature explanations remain in SKILL.md, but with no underlying implementation files in this package.
- Novada Search skill now contains only documentation; functionality must be restored from another version or source.
v1.0.1
Major update: Novada Search v2.0 introduces multi-layer AI agent support, more search engines, vertical scenes, and intelligent modes.
- Added support for 9 search engines (Google, Bing, Yahoo, DuckDuckGo, Yandex, YouTube, eBay, Walmart, Yelp)
- Enabled all 13 Google sub-types (e.g. shopping, news, images, flights, jobs, scholar, etc.)
- Introduced "vertical scenes" for purpose-driven, cross-engine searches (shopping, local, news, travel, etc.)
- New AI Agent Modes: auto (intent detection), multi (parallel engines), extract (clean content for LLMs)
- New output format: agent-json (structured for LLM/agent integration)
- Unified interface and command syntax for all features
- Optimized for direct consumption by LLMs and AI agents
v1.0.0
Web search with local business extraction and actionable URLs
元数据
常见问题
Multi Engine Search for Agent 是什么？
AI Agent search platform with 9 engines, Google 13 sub-types, vertical scene search, and intelligent auto/multi/extract modes. Designed for LLM and AI agent... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 223 次。
如何安装 Multi Engine Search for Agent？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install novada-search」即可一键安装，无需额外配置。
Multi Engine Search for Agent 是免费的吗？
是的，Multi Engine Search for Agent 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Multi Engine Search for Agent 支持哪些平台？
Multi Engine Search for Agent 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Multi Engine Search for Agent？
由 Memijashi（@goldentrii）开发并维护，当前版本 v1.0.8。
推荐 Skills