Deep Research Pro v2.1

Conducts stepwise in-depth research by generating detailed source cards with data, quotes, quality scores, cross-analyzing findings, and producing fully cite...

Security Scan

The skill's stated purpose (deep research and source-card generation) matches most of its files and instructions, but there are multiple internal inconsistencies (file-format vs. scanner script, differing source-count requirements, and strict quoting requirements) that make the package error-prone or prone to unexpected behavior — not clearly malicious, but not fully coherent.

Purpose & Capability

Name/description (in-depth research) aligns with required actions: querying OpenAlex, web browsing, generating source cards, cross-analysis, and a final report. Tools mentioned (OpenAlex, web browsing, local quality-score script, filesystem write/read) are appropriate for the stated purpose.

Instruction Scope

SKILL.md enforces heavy file I/O (creating many directories/files) and demands long direct quotes (>=50 words) and storing them in source cards — this may cause copyright/exposure concerns and is a broad data-collection behavior. There are concrete mismatches: the quality-score.py script expects a JSON input (python3 quality-score.py <source_json>) but SKILL.md and templates produce markdown card files; other files disagree on minimum source counts (SKILL.md: deep mode >=15 cards, QUALITY_CRITERIA.md and RESEARCH_PROTOCOL.md: total sources >=20). These inconsistencies could cause runtime failures or unexpected skipping of steps.

Install Mechanism

Instruction-only skill with no install spec and no external downloads; included Python script is local. This is low install risk — nothing will be fetched from arbitrary URLs or installed system-wide.

Credentials

The skill requests no environment variables, no credentials, and declares no config paths. Network usage (OpenAlex API and web browsing) is consistent with research purpose and doesn't require secret access. No unexplained credential requests found.

Persistence & Privilege

Skill writes numerous files to the agent workspace (research/, sources/, analysis/, reports/) and preserves generated cards across runs (per SKILL.md). always:false and default autonomous invocation are in place; autonomous invocation combined with persistent filesystem writes is expected here but means the skill can create/retain sizable data on disk — users should be comfortable with that.

What to consider before installing

This skill broadly does what it says (deep research and producing source cards), but several internal inconsistencies and operational risks should be resolved before use: - Fix format mismatch: scripts/quality-score.py expects JSON input but templates and SKILL.md produce markdown source cards. Either convert card files to JSON before scoring, adjust the script to parse the markdown frontmatter, or change the templates to emit JSON. Without this, the quality-check step will fail. - Reconcile source-count requirements: different files mention 15 vs 20 required sources. Decide on a single requirement and update SKILL.md, QUALITY_CRITERIA.md, and RESEARCH_PROTOCOL.md to match; otherwise the self-checks may block report generation. - Be aware of data/exposure and copyright: the skill requires copying at least one direct quote of >=50 words from each source into local cards. That may reproduce copyrighted text; consider extracting paraphrases or storing citation metadata rather than long verbatim quotes. - File-system impact: the skill will create and retain many files. Confirm you trust the workspace and are OK with persistent local data (including possibly scraped content from the web). - Operational testing: run a dry test on non-sensitive topics to confirm the workflow and script behavior. If you rely on automatic quality scoring, ensure the scoring script's assumptions about input format and fields (publication_year, sample_size, etc.) match your card data. If these issues are fixed (script/template alignment, unified requirements, and a clear policy about quoting/copyright), the skill appears coherent and usable. As-is, it's more likely to fail or behave unexpectedly, so proceed only after remediation or with caution.

Skill: Deep Research Pro (v2.1 - Enforced Depth)

版本：2.1.0 描述：世界领先的深度研究技能，强制来源卡片化、禁止跳过中间步骤

核心原则

没有来源卡片，就没有报告

触发命令

/research <主题> [--depth deep]

---

🔴 强制执行流程（不可跳过）

Step 1: 研究规划 (必须输出文件)

• 生成 research/plan.md
• 列出至少 5 个具体检索查询式 (Query Strings)
• 用户确认后才能继续

Step 2: 来源检索与卡片化 (核心深度环节)

• 禁止直接写报告
• 每找到 1 个有效来源，必须立即生成 sources/card-{{001..020}}.md
• 最低数量要求：deep 模式至少 15 个有效卡片
• 每个卡片必须包含：
  • 完整标题、作者、年份、URL/DOI
  • 核心数据提取：具体数值、样本量、P值、置信区间
  • 原文引用：至少 1 段直接引用原文 (Quote)
  • 质量评分：使用 scripts/quality-score.py 计算

Step 3: 交叉分析 (必须输出文件)

• 生成 analysis/synthesis.md
• 强制对比：至少找出 3 组相互矛盾或差异的数据
• 强制归因：解释差异原因 (样本？方法？地域？)
• 每个观点必须标注来源卡片编号 (如 [card-003])

Step 4: 报告生成 (最后一步)

• 生成 reports/final-report.md
• 引用检查：每个数据点后必须标注 [[card-xxx]]
• 完整性检查：如果来源卡片 < 15 个，拒绝生成报告并报错

---

🔴 质量门禁 (Quality Gates)

自检脚本使用方法

# 检查来源卡片数量
COUNT=$(ls sources/card-*.md 2>/dev/null | wc -l)
if [ $COUNT -lt 15 ]; then
 echo "❌ 错误：来源卡片不足 15 个，当前 $COUNT 个。请继续检索。"
 exit 1
fi

# 检查引用链接
if ! grep -q "\[\[card-" reports/final-report.md; then
 echo "❌ 错误：报告未包含来源卡片引用。"
 exit 1
fi

# 运行质量评分
for card in sources/card-*.md; do
 python3 scripts/quality-score.py "$card"
done

---

📋 分步执行命令

Step 1: 规划

生成研究计划：创建 research/plan.md
列出检索式：至少 5 个具体查询
用户确认：等待用户确认后再继续

Step 2: 检索与卡片化（循环执行）

使用 OpenAlex API 检索学术论文
使用网页浏览工具获取行业报告
每找到一个有效来源，立即生成卡片：
 - sources/card-001.md
 - sources/card-002.md
 ...
直到卡片数量 >= 15

Step 3: 分析与报告

生成 analysis/synthesis.md（交叉分析）
生成 reports/final-report.md（最终报告）
每个数据点必须标注 [[card-xxx]]

---

📄 来源卡片模板

详细模板见：templates/source-card.md

关键字段：

• source_id: card-001 格式
• core_data: 具体数值（样本量、结果、p值）
• quote: 原文引用（至少50字）
• quality_score: 0-10分

---

🔧 工具依赖

工具	用途	状态
OpenAlex API	学术论文检索	✅ 可用
网页浏览工具	行业报告获取	✅ 可用
quality-score.py	质量评分	✅ 已部署
文件系统	读写权限	✅ 可用

---

⚠️ 异常处理

• 如果无法获取最新数据 (2024-2026)，必须在报告首页标注"数据截止于模型训练时间"
• 如果来源质量评分普遍 < 6 分，必须停止并告知用户"缺乏高质量证据"
• 如果用户中断流程，已生成的卡片文件保留供下次使用

---

📊 与v2.0对比

维度	v2.0	v2.1
卡片化	无	强制
数据提取	无	必须具体数值
原文引用	无	必须50字+
分步执行	建议	强制
引用标注	无	必须[[card-xxx]]
质量门禁	建议	强制检查

---

Skill版本：2.1.0 | 最后更新：2026-03-19