research-save

name	research-save
description	Use this skill when the user asks to "research and save [topic]", "deep dive [topic] and save", "research [topic] to vault", "save research on [topic]", "研究并保存 [主题]", "深度研究 [主题]", "研究 [主题] 到笔记库", "保存关于 [主题] 的研究", "note this down", "记下来", "save to vault", "保存到笔记库", "turn this into a note", "make a research note", or mentions keywords like "研究保存", "深度研究并记录", "research to notes", "save findings to vault", "research-save". This skill conducts comprehensive web research (via a general-purpose subagent, or inline web tools as fallback) and creates properly integrated notes in the vault, saving 30-60 minutes of manual research and formatting work. Supports auto-merge for duplicate topics and handles low-quality results with incomplete status tags.
version	0.1.0

Research Save Skill

Conduct deep web research on academic/professional topics and create properly integrated notes in the Obsidian vault. Saves 30-60 minutes of manual research and formatting work.

Arguments

Parse from $ARGUMENTS:

• <主题>: Required. The research topic (Chinese or English)
• --folder <文件夹>: Optional. Override default folder (default: 6_研究/)
• --link [[笔记]]: Optional. Explicitly link to specified note(s) (can be used multiple times)

Workflow

Phase 1: Pre-validation

1. Confirm current date/time:

   date "+%Y-%m-%d %H:%M"
   

2. Check for existing notes on topic (duplicate detection):

   • Search vault for notes matching the topic using Glob and Grep
   • Search locations: 6_研究/, 2_AI/, 3_背景/
   • Exclude _index.md and CLAUDE.md from matches
   • If duplicate found with >70% topic overlap, switch to Auto-merge Mode

3. Read research context from 3_背景/:

   • Check 3_背景/Research Status.md or similar for current research priorities
   • This helps contextualize the research within the user's academic focus

Phase 2: Web Research

This phase must work on any machine, so it does not depend on a custom subagent type. Pick the first option that is available:

Option A — Delegate to a research subagent (preferred). Use the general-purpose subagent type (always available in Claude Code). If a dedicated research agent such as web-researcher exists on this machine, you may use it instead — but never assume it exists.

Task(
  subagent_type: "general-purpose",   # portable default; swap only if a research agent is known to exist
  prompt: """
  Think hard and conduct deep web research on the topic below.

  Research topic: [topic]

  Use whatever web tools are available (Tavily MCP, Metaso MCP, WebSearch,
  WebFetch, etc.). Do NOT rely on any specific tool being present — fall back
  to whatever this environment provides.

  Requirements:
  - Find 10+ authoritative sources
  - Prioritize: academic papers, official reports, reputable institutions
  - Cross-verify key claims across multiple sources
  - Structure findings with clear sections
  - Include full citations with URLs

  Save your full findings to a file and return only the file path plus a short
  status summary (do not return the entire report to the caller).

  Output format (in the saved file):
  1. Executive summary (200-300 words)
  2. Main findings (organized by 3-5 subtopics)
  3. Key data points, statistics, and quotes
  4. Source list with credibility notes
  5. Suggested tags and keywords
  """
)

Option B — Research directly (fallback when subagents are unavailable). If you cannot spawn a subagent, do the research yourself in the main loop using whatever web tools exist (WebSearch / WebFetch are always available; prefer Tavily/Metaso MCP if configured). Aim for the same 10+ sources and cross-verified findings, then proceed to Phase 3.

Phase 3: Related Note Discovery

1. Search vault folders for related concepts:

   • Use Grep to search 6_研究/, 2_AI/, 3_背景/ for topic keywords
   • Read _index.md files to identify related notes by description

2. Compile related notes list:

   • Identify at least 3 notes for wiki linking
   • Only include notes that actually exist (no speculative links)

3. Check for MoC candidates:

   • If 5+ related notes found, suggest creating a Map of Content

Phase 4: Note Generation

Create note following vault standards:

---
created: [YYYY-MM-DD]
tags:
  - type/research
  - status/draft
  - topic/[main-topic]
aliases: [alternative names, English name, 中文别名]
---

# [Research Title]

> **Summary**: [200-300 word standalone executive summary of key findings]

---

## Research Background

[Why this topic matters, context, relevance to user's research]

## Main Findings

### [Subtopic 1]
[Comprehensive findings with data, quotes, analysis]

### [Subtopic 2]
[Comprehensive findings with data, quotes, analysis]

### [Subtopic 3]
[Additional subtopics as needed]

## Key Data

| Metric | Data | Source |
|--------|------|--------|
| ... | ... | ... |

## Core Conclusions

[Synthesized insights and implications]

---

## Related

- [[Related Note 1]]
- [[Related Note 2]]
- [[Related Note 3]]
[Include --link specified notes here]

## References

1. [Source Title 1](URL1)
2. [Source Title 2](URL2)
... (10+ sources)

---

*Research Date: [YYYY-MM-DD]*

Quality Requirements:

• Word count: 1500+ words (comprehensive coverage)
• Sources: 10+ authoritative sources with full citations
• Standalone summary at beginning: 200-300 words
• Wiki links: 3+ links to existing vault notes

Phase 5: Vault Integration

1. Determine target folder:

   • Use --folder if specified
   • Default: 6_研究/
   • Ask user if uncertain (per vault rules)

2. Generate filename:

   • Use topic as filename (Chinese topics use Chinese filenames)
   • Sanitize special characters
   • If duplicate name exists, add date suffix: Topic_2026-01-05.md

3. Save note:

   • Write to: <vault-root>/[folder]/[filename].md (vault root = the current project directory, the folder containing CLAUDE.md)

4. Update folder index:

   • Append to [folder]/_index.md:

   - [[New Note Name]]: One-sentence description of research findings.
   

5. Report results:

   • File path created
   • Word count
   • Number of sources cited
   • Wiki links created
   • Suggested follow-up actions

Auto-Merge Mode

When duplicate note is detected:

1. Show existing note summary to user:

   • Display filename and first 500 characters
   • "Found existing note: [[note name]]. Auto-merging new research..."

2. Merge strategy:

   • Preserve original frontmatter, update modified date
   • Add new research sections under "## Research Update [Date]"
   • Append new sources to References section
   • Add new related notes to Related section
   • Update summary if new findings are significant

3. Report merge results:

   • Original file path
   • Sections added/updated
   • New sources integrated
   • Word count increase

Low Quality Fallback

If research yields insufficient quality results (< 5 authoritative sources):

1. Save note with incomplete status:

   • Add #status/incomplete tag instead of #status/draft
   • Document gaps in note body

2. Add limitations section:

   ## Limitations
   
   > **Note**: This research yielded fewer authoritative sources than expected.
   > Consider: [specific gaps identified]
   > Suggested follow-up: [recommendations for manual research]
   

3. Report incomplete status to user with recommendations

Quick Reference

Scenario	Command Example
Basic research	research and save quantum computing
Chinese topic	研究并保存 行为经济学
Custom folder	research AI ethics --folder 2_AI/
Explicit link	deep dive game theory --link [[博弈论MOC]]
Multiple links	研究保存 实验经济学 --link [[Research Portfolio]] --link [[博弈论]]

Vault Path

The vault root is the current project directory (the folder containing CLAUDE.md). Use paths relative to it; never hardcode an absolute path.

Related Documentation

• [[references/research-quality]] - Source evaluation criteria
• [[references/merge-patterns]] - Duplicate handling details
• [[references/vault-integration]] - Note format standards
• [[examples/basic-usage]] - Simple examples
• [[examples/advanced-usage]] - Complex scenarios