Research Assistant
/install zcx-research-assistant
Research Assistant — 深度研究助手
Conduct multi-source deep research with cross-verification, structured report generation, and full source attribution.
Research Workflow
1. Clarify Scope → What's the research question? What depth is needed?
2. Multi-Source → Search across at least 3 independent sources
3. Cross-Verify → Identify conflicts, confirm consensus
4. Synthesize → Build structured report
5. Cite Sources → Every claim gets a source URL
Always follow this order. Do not skip cross-verification.
Step 1 — Clarify Scope
Ask yourself:
- Topic: What exactly is being researched?
- Depth: Quick brief (3-5 key points) vs deep report (SWOT, data, trends)?
- Timeframe: Current snapshot, historical trend, or future outlook?
- Geography: Global, regional (APAC, EU, US), or specific country (China)?
- Audience: Technical (data-heavy) vs executive (summary + recommendations)?
Output: A 1-2 sentence research objective statement.
Step 2 — Multi-Source Retrieval
Method A: Web Search (Broad Discovery)
web_search(query="\x3Ctopic> \x3Ccontext>", count=5-10)
Search strategy — use 3-5 search queries per research topic:
- Overview:
"\x3Ctopic>" overview 2026 - Data/statistics:
"\x3Ctopic>" statistics data market size - News/recent:
"\x3Ctopic>" latest news 2026 - Competitors:
"\x3Ctopic>" competitors leading companies - Analysis:
"\x3Ctopic>" analysis report SWOT
Method B: Deep Fetch (Page-Level Detail)
For each promising search result, fetch the full page:
web_fetch(url, extractMode="markdown", maxChars=15000)
Prioritize sources in this order:
- Primary sources — Official company reports, government data, regulatory filings
- Industry sources — Specialized trade publications, analyst reports
- Major media — Established financial/industry news outlets
- Secondary sources — Aggregators, summaries, blogs (lowest priority)
Minimum source count: At least 3 independent sources per key claim.
Method C: Targeted Queries (Gap Filling)
When initial results leave gaps (missing data, conflicting info, too old):
web_search(query="\x3Crefined query>", count=3)
Repeat until each major section of the report has supporting sources.
Step 3 — Cross-Verification
Conflict Detection
When sources disagree:
| Conflict Type | Resolution |
|---|---|
| Different numbers (e.g., market size) | Check dates (older may be out of date), cite the most recent or authoritative |
| Contradictory claims | Find a 3rd/4th source to break the tie |
| One strong + one weak | Trust the stronger (more authoritative, more recent, cited data) |
| Both weak | Flag as "conflicting reports" in the output |
Verification Checklist
For each key fact, confirm:
- At least 2 independent sources agree
- Data includes a clear date or timeframe
- Source is authoritative (official > news > blog)
- Numbers are internally consistent (percentages add up, totals match)
- No obvious bias or agenda in the source
Quality Scoring
| Score | Criteria |
|---|---|
| ✅ Confirmed | 3+ independent reliable sources agree |
| ⚠️ Likely | 2 sources agree, no strong counter-evidence |
| ❓ Uncertain | Only 1 source, or sources conflict |
| ❌ Contradicted | Multiple sources disagree with the claim |
Mark each major finding with its confidence level.
Step 4 — Report Templates
Template A: Research Brief (Quick — 3-5 key points)
# Research Brief: [Topic] (Date)
## Objective
[1-line research question]
## Key Findings
1. **[Finding 1]** — Source: [URL]
2. **[Finding 2]** — Source: [URL]
3. **[Finding 3]** — Source: [URL]
## Summary
[3-5 sentence synthesis of all findings]
## Sources
- [Source 1 Title](URL)
- [Source 2 Title](URL)
Template B: SWOT Analysis
# SWOT: [Company/Topic] (Date)
## Strengths (Internal Positive)
- [Strength 1] — Source: [URL]
- [Strength 2] — Source: [URL]
## Weaknesses (Internal Negative)
- [Weakness 1] — Source: [URL]
- [Weakness 2] — Source: [URL]
## Opportunities (External Positive)
- [Opportunity 1] — Source: [URL]
- [Opportunity 2] — Source: [URL]
## Threats (External Negative)
- [Threat 1] — Source: [URL]
- [Threat 2] — Source: [URL]
## Strategic Implications
[2-3 sentence synthesis of what the SWOT means]
Template C: Competitor Analysis
# Competitive Analysis: [Industry/Segment] (Date)
## Market Overview
- Total addressable market: [data + source]
- Growth rate: [data + source]
- Key trends: [3-5 trends + sources]
## Competitor Profiles
### [Competitor A]
- **Position:** [Market leader / Challenger / Niche]
- **Revenue/Size:** [Data + source]
- **Key Strengths:** [...]
- **Key Weaknesses:** [...]
- **Recent Moves:** [1-2 notable events + source]
### [Competitor B]
...
## Competitive Map
| Dimension | Competitor A | Competitor B | Competitor C |
|:----------|:------------:|:------------:|:------------:|
| Market Share | X% | Y% | Z% |
| Pricing | High/Medium/Low | ... | ... |
| Feature Set | Full/Basic | ... | ... |
## Recommendations
[Based on the analysis]
Template D: Full Industry Research Report
# Industry Report: [Industry Name] (Date)
## 1. Executive Summary
[3-5 sentence overview with key numbers]
## 2. Market Size & Growth
- Current market size: [data + year + source]
- CAGR: [X% + source]
- Forecast: [projected size + year + source]
## 3. Key Players
[Table of top 5-10 companies with market share, revenue, focus areas]
## 4. Trends & Drivers
1. **[Trend 1]** — [1-2 sentence explanation + source]
2. **[Trend 2]** — [1-2 sentence explanation + source]
## 5. Challenges & Risks
1. **[Risk 1]** — [1-2 sentence explanation + source]
2. **[Risk 2]** — [1-2 sentence explanation + source]
## 6. Regulatory Environment
[Key regulations, policies, compliance requirements + sources]
## 7. Outlook & Recommendations
[3-5 specific, actionable recommendations based on data]
## 8. Sources
For each source: [Title](URL) — confidence level
Source Citation Rules
Every factual claim in the output must include a source URL.
Format:
[Claim text] — Source: [Source Name](URL)
For reports with many sources, include a Sources section at the end with all URLs.
Blacklisting
Do not use these source types as primary sources:
- Anonymous forums (unless specifically investigating sentiment)
- Self-published/promotional content
- Unedited AI-generated content
- Clickbait aggregators with no original reporting
If a user asks for information from a blacklisted source type, warn them and find better sources.
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install zcx-research-assistant - 安装完成后,直接呼叫该 Skill 的名称或使用
/zcx-research-assistant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
Research Assistant 是什么?
Conduct multi-source research with cross-verification to produce structured reports like SWOT, competitor analyses, or industry briefs with full source citat... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 46 次。
如何安装 Research Assistant?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install zcx-research-assistant」即可一键安装,无需额外配置。
Research Assistant 是免费的吗?
是的,Research Assistant 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Research Assistant 支持哪些平台?
Research Assistant 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 Research Assistant?
由 zhaocaixia888(@zhaocaixia888)开发并维护,当前版本 v1.0.2。