ZeeLin Deep Research 深度研究

ZeeLin Deep Research 深度研究是一款 AI 驱动的专业研究辅助平台，支持一句话生成与多步骤生成，提供深度、专家两大研究路径。从快速信息梳理、系统分析到超万字专家报告全流程覆盖，依托多轮推理与多源数据整合，高效完成企业分析、市场洞察、招商研究等复杂任务，一站式提升研究效率与决策质量。

Security Scan

The skill's code largely matches its stated purpose (create a remote research task, poll status, and push a PDF link), but it reads/saves environment values, calls an external API, and invokes an undeclared CLI ('openclaw') — behaviors that are plausible but not fully documented and warrant caution.

Purpose & Capability

The skill's scripts call ZeeLin endpoints (desearch.zeelin.cn) to create and monitor tasks and then push a PDF link to a user — this aligns with the 'deep research' description. It requires an API key stored in config.json which is consistent. However, the runtime expects an 'openclaw' CLI to deliver notifications but that binary is not declared in the skill's requirements/metadata (the SKILL.md lists python3 and requests). The missing declaration is an inconsistency developers should clarify.

Instruction Scope

The SKILL.md and included scripts read environment variables (ZEELIN_TARGET_USER, ZEELIN_CHANNEL, ZEELIN_SESSION_ID, ZEELIN_CONTENT) and will persist target_user/channel into the skill's config.json. The skill also forwards the PDF download URL returned by the external API directly to the target via an 'openclaw message send' subprocess call. While these actions match the feature list (auto-config, background push), they involve reading env vars not declared in requires.env and automatically persisting identifiers — a privacy and transparency gap the user should be aware of.

Install Mechanism

There is no install spec (instruction-only with small helper scripts). No remote downloads or archive extraction are used, which is lower risk. The SKILL.md metadata asks for pip 'requests' and python3, which matches the scripts' use of requests.

Credentials

The skill requires an API key (stored in config.json) to call the external ZeeLin API — reasonable for its purpose. However it also implicitly depends on environment variables (ZEELIN_TARGET_USER, ZEELIN_CHANNEL) and will persist them into config.json without declaring them as required; it also expects the 'openclaw' CLI to exist for notifications. Those additional implicit dependencies increase the attack surface and persistence of identifiers and should have been declared and justified.

Persistence & Privilege

The skill spawns a background monitor process (using nohup) that polls the remote API and writes logs under /tmp and state into the skill-local config.json. always is false and the skill does not modify other skills or system-wide settings. Background execution and saving target info are expected for this use-case, but they do create persistent artifacts the user should know about.

What to consider before installing

This skill appears to implement what it claims, but there are a few things to check before installing: - It requires you to put a ZeeLin API key in the skill's config.json; treat that file as sensitive. Consider using a limited key and rotating it if possible. - The scripts will read ZEELIN_TARGET_USER and ZEELIN_CHANNEL environment variables (if set) and persist them into config.json. If you don't want those values stored, ensure those env vars are unset or review/clean config.json after first run. - The watch script uses an 'openclaw message send' subprocess to deliver the PDF link; make sure the openclaw CLI is trusted and available in your environment. If you don't want the skill to send messages automatically, do not provide target/channel or modify the notify_done function. - The skill calls an external domain (https://desearch.zeelin.cn) and will forward the PDF URL returned by that service. If data confidentiality is a concern, review what content is sent to ZeeLin and whether the returned links point to trusted hosts. If you need higher assurance, ask the publisher for: a) a declared list of required binaries (including openclaw), b) justification for the env vars it reads and where data is sent/stored, and c) a privacy/data-retention policy for generated reports. If you cannot verify those, treat this skill as potentially privacy-sensitive and consider running it in a restricted environment.

ZeeLin Deep Research 深度研究是一款 AI 驱动的专业研究辅助平台，支持一句话生成与多步骤生成，提供深度、专家两大研究路径。从快速信息梳理、系统分析到超万字专家报告全流程覆盖，依托多轮推理与多源数据整合，高效完成企业分析、市场洞察、招商研究等复杂任务，一站式提升研究效率与决策质量。

功能特点

• ✅ 自动推送：任务完成后自动发送 PDF下载链接 给用户
• ✅ 后台执行：通过子代理执行，不阻塞主会话
• ✅ 自动配置：首次使用时自动获取用户渠道和 ID

没有 API Key？

前往 https://skills.zeelin.cn/console/apps 免费注册获取

获取后，编辑 skill 目录下的 config.json 文件：

{
  "api_key": "你的API_KEY"
}

配置文件位置：config.json

thinking 参数

模式	用途
deep	概念解释、信息聚合、快速调研
major	行业分析、企业研究、万字报告

模式选择指南

使用深度模式 (deep) 的场景：

• 概念解释/知识梳理 — "什么是 XXX"、"XXX 的原理是什么"
• 事件/趋势分析 — 某个热点事件的来龙去脉、发展趋势
• 信息聚合 — 围绕某个话题汇总多方信息和观点
• 快速调研 — 用户没有明确要求深度报告的一般性研究问题

使用专家模式 (major) 的场景：

• 行业/市场分析 — 涉及特定行业的全景分析、市场格局、竞争态势
• 企业/公司研究 — 某个企业的财务分析、战略评估、业务拆解
• 政策/法规研究 — 需要系统梳理政策影响、合规要求
• 技术/产品深度对比 — 多维度的技术路线对比、产品竞品分析
• 投资/商业决策 — 需要数据支撑的投资分析、可行性评估

如果不确定使用哪种模式：可以询问用户确认

使用方式

spawn 子代理执行

spawn 执行 python3 scripts/zeelin_start.py "调研内容" deep

示例：

spawn 执行 python3 scripts/zeelin_start.py "调研合肥地理位置分析" deep

工作流程

用户发送调研请求
        ↓
spawn 子代理执行 zeelin_start.py
        ↓
创建 ZeeLin 任务，获取 session_id
        ↓
后台监控任务状态（每30秒检查）
        ↓
状态变为完成（status=2）
        ↓
自动推送PDF下载链接到用户

文件位置

• 配置文件：config.json
• 启动脚本：scripts/zeelin_start.py
• 监控脚本：scripts/zeelin_watch.py

状态码

状态码	含义
1	进行中
2	✅ 完成
3	用户终止
4	❌ 失败
5	排队中

常见错误

错误	解决方法
code=315 当前接口调用已超限	⚠️ 请等待当前任务执行完成后再提交新任务
code=315 试用已超限	前往 https://skills.zeelin.cn/console/recharge 充值