← 返回 Skills 市场
Deep Search
作者
aiwithabidi
· GitHub ↗
· v1.0.0
400
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install deep-search
功能描述
Perplexity-powered AI search with auto model tier selection for quick facts, research, or deep analysis across web, academic, news, YouTube, and Reddit.
使用说明 (SKILL.md)
Deep Search 🔍
Multi-tier Perplexity-powered search with automatic Langfuse observability tracing.
When to Use
- Quick facts and simple lookups →
quicktier - Standard research, comparisons, how-to →
protier - Deep analysis, market research, complex questions →
deeptier - Academic paper search, news monitoring, Reddit/YouTube research
Usage
# Quick search (sonar, ~2s)
python3 {baseDir}/scripts/deep_search.py quick "what is OpenClaw"
# Pro search (sonar-pro, ~5-8s)
python3 {baseDir}/scripts/deep_search.py pro "compare Claude vs GPT-4o for coding"
# Deep research (sonar-reasoning-pro, ~10-20s)
python3 {baseDir}/scripts/deep_search.py deep "full market analysis of AI agent frameworks"
# Focus modes
python3 {baseDir}/scripts/deep_search.py pro "query" --focus academic
python3 {baseDir}/scripts/deep_search.py pro "query" --focus news
python3 {baseDir}/scripts/deep_search.py pro "query" --focus youtube
python3 {baseDir}/scripts/deep_search.py pro "query" --focus reddit
Tiers
| Tier | Model | Speed | Best For |
|---|---|---|---|
| quick | sonar | ~2s | Simple facts, quick lookups |
| pro | sonar-pro | ~5-8s | Research, comparisons |
| deep | sonar-reasoning-pro | ~10-20s | Deep analysis, complex questions |
Environment
PERPLEXITY_API_KEY— Required. Perplexity API key.OPENROUTER_API_KEY— Optional. For Langfuse tracing model pricing.
Credits
Built by M. Abidi | agxntsix.ai YouTube | GitHub Part of the AgxntSix Skill Suite for OpenClaw agents.
📅 Need help setting up OpenClaw for your business? Book a free consultation
安全使用建议
Before installing or running this skill:
- Expect to supply a Perplexity API key (PERPLEXITY_API_KEY or PPLX_API_KEY); the registry metadata should state this but currently does not — ask the publisher to correct the manifest.
- Review the bundled script locally. It will try to read ~/.openclaw/openclaw.json for keys — if you don't want that, either remove that code or ensure the file contains no sensitive data.
- The script contains hardcoded Langfuse keys and a default LANGFUSE_HOST. These could cause your queries/results to be sent to a tracing/observability endpoint controlled by the publisher (or by the embedded host). If you are concerned about telemetry or data leakage, remove or override those env defaults before running, or run the skill in an isolated environment.
- Ensure you trust the author (agxntsix) and the included langfuse tracing behavior; request clarification on why default secret keys are embedded and what data is sent to Langfuse.
- Consider running the script in an ephemeral container or VM, or inspect network traffic, until you are satisfied the tracing endpoint and keys are safe.
If the publisher can (a) remove embedded keys, (b) update registry metadata to declare required env vars and dependencies, and (c) clearly document tracing behavior and opt-out, the incoherence and privacy concerns would be largely resolved.
功能分析
Type: OpenClaw Skill
Name: deep-search
Version: 1.0.0
The skill bundle appears benign. The `SKILL.md` instructions are clear and do not contain any prompt injection attempts. The `deep_search.py` script performs web searches using the legitimate Perplexity API and includes optional Langfuse tracing to an internal host (`http://langfuse-web:3000`), which is an observability feature, not data exfiltration to an external malicious entity. The script attempts to retrieve the `PERPLEXITY_API_KEY` from environment variables or a specific path within `~/.openclaw/openclaw.json`, which is a common and likely intended mechanism for OpenClaw skills to access API keys for their operation, not an attempt to steal arbitrary secrets.
能力评估
Purpose & Capability
The name/description claim Perplexity-powered, multi-tier web/academic/news/YouTube/Reddit search and the code indeed calls Perplexity's API. However registry metadata declares no required environment variables while SKILL.md and the script require a Perplexity API key (PERPLEXITY_API_KEY or PPLX_API_KEY) and will also try to read ~/.openclaw/openclaw.json for keys. The mismatch between declared requirements and actual behavior is incoherent and should be corrected.
Instruction Scope
SKILL.md instructs running the provided Python script which is consistent with the purpose. The runtime script however: (1) attempts to read the user's OpenClaw config at ~/.openclaw/openclaw.json (not documented in the registry metadata), (2) sets default Langfuse environment keys inside the process (hardcoded secrets), and (3) will attempt to use an optional langfuse client to trace observations — which may transmit query text and results to a tracing endpoint. Reading the user's config file and sending traces are beyond the narrow 'search' description unless explicitly disclosed.
Install Mechanism
This is instruction + bundled script only (no install spec), which is low install-risk. The script requires third-party Python packages (requests; optionally langfuse) but these dependencies are not declared in the manifest. No external archive downloads or installers are used.
Credentials
The SKILL.md and code expect a Perplexity API key (PERPLEXITY_API_KEY or PPLX_API_KEY), but the registry metadata lists none — a concerning omission. The script also unilaterally sets LANGFUSE_SECRET_KEY and LANGFUSE_PUBLIC_KEY defaults (hardcoded secrets) and LANGFUSE_HOST to http://langfuse-web:3000; these embedded keys could enable telemetry/tracing to a third-party Langfuse instance without the user's explicit configuration or consent. The skill also searches for credentials in a local OpenClaw config file (~/.openclaw/openclaw.json), which increases the scope of local data access.
Persistence & Privilege
The skill does not request always: true, does not modify system-wide settings, and does not persistently write configuration. It only sets process-level environment defaults and attempts to read a local config file for an API key. That read is scoped to a single file path (the OpenClaw config) rather than broad filesystem access.
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 deep-search - 安装完成后,直接呼叫该 Skill 的名称或使用
/deep-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
v1.0.0
Initial release of deep-search: a multi-tier, Perplexity-powered AI web search tool with automatic routing and focus modes.
- Supports three search tiers: quick (facts), pro (comparisons), and deep (analysis), auto-selected by query complexity
- Integrates focus modes for targeted searching: internet, academic, news, YouTube, and Reddit
- Requires a Perplexity API key; optional OpenRouter API key for enhanced tracing
- Suitable for research, fact-checking, and competitive analysis tasks
- Command-line interface for easy tier and focus selection
元数据
常见问题
Deep Search 是什么?
Perplexity-powered AI search with auto model tier selection for quick facts, research, or deep analysis across web, academic, news, YouTube, and Reddit. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 400 次。
如何安装 Deep Search?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install deep-search」即可一键安装,无需额外配置。
Deep Search 是免费的吗?
是的,Deep Search 完全免费(开源免费),可自由下载、安装和使用。
Deep Search 支持哪些平台?
Deep Search 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 Deep Search?
由 aiwithabidi(@aiwithabidi)开发并维护,当前版本 v1.0.0。
推荐 Skills