Description

提供完整多阶段深度研究，覆盖规划、跨源检索、质量筛选、深入分析、交叉验证及结构化报告生成。

README (SKILL.md)

Skill: Deep Research Pro

Name: Deep Research Pro v2
Author: xueylee-dotcom

版本：2.0.0 描述：世界领先的深度研究技能，支持多阶段迭代、交叉验证、批判分析

触发条件

当用户要求进行深度研究、市场调研、学术论文检索、竞品分析时自动激活。

支持的命令格式

/research \x3C主题> [--depth shallow|medium|deep]
深度调研 \x3C主题>
研究 \x3C主题> 的市场/技术/竞品
帮我查一下 \x3C主题> 的学术论文

研究阶段（强制顺序执行）

Phase 1: 研究规划 (Research Planning)

目标：明确研究问题、定义边界、设计检索策略输出：research-plan.md 必须包含：

核心研究问题（3-5 个）
检索关键词矩阵（主关键词 + 同义词 + 相关概念）
数据源清单（学术/行业/政策/专利）
质量评估标准（纳入/排除标准）
预期产出结构

Phase 2: 多源检索 (Multi-Source Retrieval)

目标：从多源获取信息，避免单一来源偏见输出：sources/raw-sources.json 必须包含：

学术：OpenAlex/PubMed/Google Scholar（至少 15 篇核心论文）
行业：公司官网/行业报告/竞品文档（至少 5 个竞品）
政策：政府文件/监管机构/标准组织（如适用）
专利：Google Patents/WIPO（如适用）
每个来源必须有：URL、发布时间、作者/机构、摘要

Phase 3: 质量筛选 (Quality Screening)

目标：过滤低质量信息，保留高可信度来源输出：sources/filtered-sources.json + sources/excluded-sources.json 评估维度：

维度	权重	评估标准
来源权威性	30%	顶刊/知名机构/官方来源
时效性	25%	近 3 年优先，经典文献可放宽
方法论严谨性	25%	样本量、对照组、统计方法
可复现性	10%	数据/代码是否公开
利益冲突披露	10%	是否声明资助方/利益关系

Phase 4: 深度分析 (Deep Analysis)

目标：提取关键洞察，而非简单摘要输出：analysis/insights.md 必须包含：

每个核心发现的证据等级（A/B/C/D）
相互矛盾的研究发现及可能原因
研究局限性分析（样本、方法、地域等）
未解决的问题/研究空白

Phase 5: 交叉验证 (Cross-Validation)

目标：多源信息相互印证，识别偏见输出：analysis/validation-matrix.md 必须包含：

学术 vs 行业数据对比
不同研究机构结论对比
时间维度趋势分析（早期 vs 最新研究）
地域差异分析（如适用）

Phase 6: 综合报告 (Synthesis Report)

目标：生成结构化、可追溯的深度报告输出：reports/final-report.md 必须包含：

执行摘要（300 字内）
方法论透明说明（检索策略、筛选标准）
核心发现（带证据等级）
批判性分析（局限性、偏见、未解问题）
可操作建议（优先级排序）
完整参考文献（带可点击链接）
附录（原始数据、检索查询、排除来源说明）

Phase 7: 用户反馈迭代 (Feedback Loop)

目标：支持用户质疑后重新检索/分析输出：reports/revised-report.md 触发条件：用户对某结论提出质疑或要求深化 必须包含：

用户质疑点记录
补充检索策略
修订说明（什么变了、为什么）

执行参数

参数	默认值	说明
depth	deep	shallow=快速概览，medium=标准深度，deep=完整7阶段
sources	all	academic=仅学术，industry=仅行业，all=全源
min_sources	20	最低来源数量要求
quality_threshold	0.7	质量评分阈值（0-1）

质量门禁（Quality Gates）

报告生成前必须通过以下检查：

核心结论至少有 2 个独立来源支持
所有数据都有明确来源和日期
相互矛盾的发现已标注并分析原因
研究局限性已明确说明
参考文献包含可访问链接

研究输出目录结构

research/domains/[领域]/[主题]/
├── research-plan.md          # Phase 1: 研究规划
├── sources/
│   ├── raw-sources.json      # Phase 2: 原始来源
│   ├── filtered-sources.json # Phase 3: 筛选后来源
│   └── excluded-sources.json # 排除的来源及原因
├── analysis/
│   ├── insights.md           # Phase 4: 深度分析
│   └── validation-matrix.md   # Phase 5: 交叉验证
└── reports/
    ├── final-report.md        # Phase 6: 最终报告
    └── revised-report.md      # Phase 7: 修订报告（如有）

工具依赖

工具	用途	备选方案
OpenAlex API	学术论文检索	PubMed/Google Scholar
jq	JSON处理	Python json模块
bc	数学计算	Python计算

质量评估标准

学术论文评分卡（0-10分）

维度	评分项	分值
时效性	近3年	2.5
时效性	3-5年	2
时效性	5年+	1
权威性	Nature/Science/Lancet	3
权威性	其他顶刊	2.5
权威性	PubMed索引	2
方法论	样本量>1000	1
方法论	有对照组	1
方法论	多中心	1
可复现	数据/代码公开	1
透明度	利益冲突声明	0.5

证据等级：A(9-10) / B(7-8) / C(5-6) / D(\x3C5)

报告模板引用

详细报告模板请参考：templates/report-template.md 来源卡片模板请参考：templates/source-card.md

Skill版本：2.0 | 最后更新：2026-03-19

Usage Guidance

This skill appears to implement a reasonable multi‑phase research workflow and includes a small Python scoring script and report templates, but you should do the following before installing or relying on it: - Verify scoring consistency: the included QUALITY_CRITERIA and templates use a 0–1.0 or normalized quality scale, while scripts/quality-score.py computes totals on a 0–10 scale. Confirm how scores are normalized in the final report to avoid misleading grades. - Ensure runtime environment: the script expects Python3 and that the agent environment can read/write the described report directory structure and JSON source files. Test the script locally with sample source JSON to verify behavior. - Review web access implications: the skill will fetch many external sources (OpenAlex, PubMed, Google Scholar, company sites). Make sure the agent is allowed to make outbound web requests, and be aware of rate limits, robots.txt and copyright considerations (Google Scholar may require manual scraping or special handling). - Validate data handling: because the skill aggregates many sources and generates reports, check that no sensitive/internal documents will be pulled or published inadvertently when run in an environment with access to private intranets or proprietary APIs. - Test with non‑sensitive topics first: run the pipeline on public topics to confirm outputs (source lists, filtered/ excluded reasoning, final report) and that references/links are correctly formatted and accessible. Overall: coherent and low technical risk, but the scoring/template mismatch and heavy sourcing requirements merit review/testing before trusting outputs for decision making.

Capability Analysis

Type: OpenClaw Skill Name: deep-research-v2 Version: 2.0.0 The skill bundle provides a comprehensive and professional framework for conducting multi-phase deep research, including planning, multi-source retrieval, and quality validation. It includes a Python utility (scripts/quality-score.py) for calculating research source credibility based on objective criteria like publication year and journal impact. No indicators of data exfiltration, malicious execution, or harmful prompt injection were found; all instructions and code are strictly aligned with the stated purpose of academic and market analysis.

Capability Assessment

ℹ Purpose & Capability

Name/description (deep multi‑source research) align with the files and instructions: planning, multi‑source retrieval, screening, analysis, validation, and report templates are all present. No unexpected credentials, binaries, or external installers are requested. Minor coherence issue: different files use different quality-score scales (QUALITY_CRITERIA and templates reference a 0-1.0/percentage-style score while scripts/quality-score.py computes a 0-10 total), which is likely a logic/normalization bug rather than malicious behavior.

✓ Instruction Scope

SKILL.md describes a bounded 7‑phase research pipeline and references legitimate public sources (OpenAlex, PubMed, Google Scholar) and local processing (Python/jq/bc alternatives). The instructions do not ask the agent to read unrelated system files or environment variables, nor to post results to unknown endpoints. The skill states it auto-activates when a user requests deep research — that is consistent with its purpose and agent invocation defaults.

✓ Install Mechanism

No install spec; the skill is instruction-only with one small included Python script. Nothing is downloaded from external/unknown URLs, no archives are extracted, and no package installs are declared.

✓ Credentials

The skill requests no environment variables, no credentials, and no config paths. It references OpenAlex and other public sources but does not require keys in the manifest. This is proportionate to its stated function.

✓ Persistence & Privilege

always:false and normal model invocation settings (disable-model-invocation:false). The skill does not request persistent privileges or modify system/global skill configurations. Autonomous invocation is allowed by default and appropriate for this type of skill.

Version History

v2.0.0

Deep Research Pro 2.0.0 introduces a comprehensive, multi-phase framework for high-quality deep research with robust validation and analysis. - Adds a strict 7-phase research workflow: from planning and multi-source data gathering to deep analysis, cross-validation, and feedback-driven iteration. - Enforces standardized quality gates and reporting formats, ensuring source transparency and evidence-based conclusions. - Supports flexible research depth (shallow/medium/deep) and source selection (academic, industry, or all). - Implements detailed quality scoring and documentation for all outputs, including synthesis and revised reports. - Integrates tool dependencies and structured output directory for improved traceability and usability.

Metadata

Slug deep-research-v2

Version 2.0.0

License MIT-0

All-time Installs 0

Active Installs 0

Total Versions 1

Frequently Asked Questions

What is Deep Research Pro v2?

提供完整多阶段深度研究，覆盖规划、跨源检索、质量筛选、深入分析、交叉验证及结构化报告生成。 It is an AI Agent Skill for Claude Code / OpenClaw, with 210 downloads so far.

How do I install Deep Research Pro v2?

Run "/install deep-research-v2" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Deep Research Pro v2 free?

Yes, Deep Research Pro v2 is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Deep Research Pro v2 support?

Deep Research Pro v2 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Deep Research Pro v2?

It is built and maintained by xueylee-dotcom (@xueylee-dotcom); the current version is v2.0.0.

More Skills

Deep Research Pro v2