← Back to Skills Marketplace
juan-xin-cai

deep-research

by juan-xin-cai · GitHub ↗ · v1.0.2 · MIT-0
cross-platform ✓ Security Clean
349
Downloads
0
Stars
0
Active Installs
3
Versions
Install in OpenClaw
/install google-deep-research
Description
使用 Gemini Deep Research Agent 进行自主深度研究,生成带引用的详细研究报告。当用户需要深度调研话题、生成研究报告、或需要多轮搜索和分析时使用。触发词:深度研究、deep research、帮我调研、研究一下、写研究报告、deep dive、详细分析。
README (SKILL.md)

Deep Research Agent

使用 Gemini Deep Research Pro 执行自主多步搜索-阅读-分析的深度研究,通常需要 5-20 分钟完成。

依赖来源

  • 首选 SDK 来源:npm 包 @google/genai
  • 可选兼容模式:通过 GOOGLE_GENAI_SDK_PATH 显式指定一个本地 SDK 路径

不要依赖某台开发机上的隐式 bundled SDK 路径,也不要在技能里写死个人机器目录。

工作流程

1. 确认研究主题

  • 与用户确认研究主题和范围
  • 如果话题模糊,帮助用户明确研究方向和关注点
  • 将用户需求转化为清晰的英文或中文研究查询

2. 准备 API Key

必须显式提供:

export GEMINI_API_KEY="\x3Capi_key>"

不要自动读取本地 openclaw.json 或扫描多个默认配置位置。发布型技能应让密钥来源保持明确、可审查、可控。

3. 执行研究

提醒用户:深度研究通常需要 5-20 分钟,请耐心等待。

脚本与 SKILL.md 位于同一目录。执行时先把脚本路径解析为当前技能目录下的 deep-research.mjs,不要写死开发机绝对路径。

运行脚本:

GEMINI_API_KEY="\x3Capi_key>" node \x3Cskill-dir>/deep-research.mjs "\x3C研究主题>"

可选参数:

  • --timeout \x3Cseconds> — 超时时间,默认 3600 秒(1 小时)
  • --no-stream — 使用轮询模式(如果流式出错可尝试)

可选 SDK 路径覆盖:

GOOGLE_GENAI_SDK_PATH="/path/to/@google/genai/dist/node/index.mjs" \
GEMINI_API_KEY="\x3Capi_key>" \
node \x3Cskill-dir>/deep-research.mjs "\x3C研究主题>"

重要:由于研究耗时较长,使用 exec 工具时设置足够的超时时间(至少 1200 秒)。

4. 后续追问

研究完成后,脚本会输出 Interaction ID。如果用户需要追问或深入某个方面:

GEMINI_API_KEY="\x3Capi_key>" node \x3Cskill-dir>/deep-research.mjs --follow-up \x3Cinteraction_id> "\x3C追加问题>"

5. 呈现结果

  • stdout 输出为 Markdown 格式的研究报告
  • 保持原始 Markdown 格式呈现给用户
  • 报告包含引用来源链接
  • 如果用户需要,可以将报告保存为文件

输出说明

  • stdout — 最终研究报告(Markdown 格式)
  • stderr — 进度信息:
    • 🔬 研究开始
    • 📊 状态更新
    • 💭 思考摘要(Agent 的中间思考过程)
    • ✅ 研究完成 + 耗时
    • 📎 Interaction ID(用于后续追问)

错误处理

退出码 含义 处理方式
0 成功 正常呈现报告
1 参数错误 检查研究主题是否提供
2 环境或 API 错误 检查 GEMINI_API_KEY@google/genai 安装和配额
3 超时 建议缩小研究范围或增加超时

安全

  • 不要将 API Key 直接展示给用户
  • 使用环境变量传递 API Key,不要把真实密钥写进技能文件
  • 不要在未获用户确认前读取本地配置中的潜在密钥
Usage Guidance
This skill appears to do what it says: run a Gemini deep-research agent using your GEMINI_API_KEY and the @google/genai SDK. Before installing/using: (1) confirm you install @google/genai from the official npm package or provide a vetted local SDK path, (2) treat GEMINI_API_KEY as sensitive—use a scoped/limited key if possible and rotate it after use, (3) be aware stderr includes intermediate agent 'thoughts' (may reveal internal chain-of-thought), so avoid exposing stderr content to end users without review, and (4) run the script in an isolated environment if you have any doubts about provenance. If you want higher assurance, review the included deep-research.mjs source (already inspected here) and verify the npm package contents match upstream releases.
Capability Analysis
Type: OpenClaw Skill Name: google-deep-research Version: 1.0.2 The skill bundle is a legitimate implementation of a research agent using the Gemini Deep Research API. The code in `deep-research.mjs` correctly handles API interactions, streaming, and environment variables without any signs of data exfiltration, malicious execution, or obfuscation. It follows security best practices by explicitly instructing the agent not to auto-read local configuration files for secrets and instead relies on the `GEMINI_API_KEY` environment variable.
Capability Assessment
Purpose & Capability
Name and description claim a Gemini deep-research agent and the skill requires a GEMINI_API_KEY and the @google/genai SDK (or an explicit SDK path). Those requirements match the stated purpose.
Instruction Scope
SKILL.md instructs running the included deep-research.mjs with GEMINI_API_KEY in the environment, and explicitly disallows reading local config files. It also warns not to display the API key. Note: the script intentionally prints intermediate 'thinking summaries' to stderr (agent internal summaries), which may reveal chain-of-thought details; this is expected for debugging/progress but may leak internal reasoning if you present stderr output directly to end users.
Install Mechanism
There is no install spec; the script dynamically imports '@google/genai' or a user-provided SDK path. This is low-risk but means the runtime must have the npm package installed from the registry (or a trusted local build). Verify the package source before installing.
Credentials
The only required credential is GEMINI_API_KEY (and an optional GOOGLE_GENAI_SDK_PATH). No unrelated secrets, config paths, or extra credentials are requested.
Persistence & Privilege
Skill is not always-enabled and does not request elevated or persistent system privileges, nor does it modify other skills' configs. It runs as an invoked script only.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install google-deep-research
  3. After installation, invoke the skill by name or use /google-deep-research
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.2
- Clarifies dependency handling: now requires explicit use of the @google/genai NPM package and/or GOOGLE_GENAI_SDK_PATH; stops scanning multiple local config paths. - Sets GEMINI_API_KEY as the explicit and only way to provide API credentials; local file/config scan is no longer supported. - Adds OpenClaw metadata (primaryEnv, homepage) to the manifest. - Updates process instructions to ensure SDK path and API key are provided in a clear, auditable, and controlled way. - Improves clarity on security and avoids use of personal/dev machine paths or bundled SDKs.
v1.0.1
- 增强了 API Key 加载逻辑,优先使用环境变量,无需硬编码路径,同时兼容本地 OpenClaw 配置。 - 优化 Google GenAI SDK 的加载机制,支持多种路径自动发现,提高节点兼容性。 - 移除了开发者绝对路径说明,改为相对 skill 目录下脚本,便于跨环境部署。 - 更新错误处理,覆盖 SDK 路径和环境问题。 - 文档细化 API Key 获取步骤及运行环境兼容性说明。
v1.0.0
feat: publish deep research skill
Metadata
Slug google-deep-research
Version 1.0.2
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 3
Frequently Asked Questions

What is deep-research?

使用 Gemini Deep Research Agent 进行自主深度研究,生成带引用的详细研究报告。当用户需要深度调研话题、生成研究报告、或需要多轮搜索和分析时使用。触发词:深度研究、deep research、帮我调研、研究一下、写研究报告、deep dive、详细分析。 It is an AI Agent Skill for Claude Code / OpenClaw, with 349 downloads so far.

How do I install deep-research?

Run "/install google-deep-research" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is deep-research free?

Yes, deep-research is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does deep-research support?

deep-research is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created deep-research?

It is built and maintained by juan-xin-cai (@juan-xin-cai); the current version is v1.0.2.

💬 Comments