Research - Multi-Source Research & Synthesis Skill

Research - Multi-Source Research & Synthesis Skill

Argument Parsing

• No argument → ask the user for a research topic or URLs
• [topic] → use WebSearch to find relevant sources, then synthesize
• [url1 url2 ...] → fetch and analyze the provided URLs directly
• [topic] [url1 url2 ...] → combine user-provided URLs with additional web research

Argument: $ARGUMENTS

---

Difference from /oh-my-study-with-me:study

	research	study
Scope	Multiple sources, broad coverage	Single source, deep dive
Depth	Wide (cross-source synthesis)	Deep (Why? dialogue)
Interaction	Mostly autonomous gather + synthesize	Interactive Socratic dialogue
Output	Research summary with source attribution	Study memo with verification
Best for	Surveying a topic, comparing approaches, finding consensus	Mastering one book/article in depth

The two skills are complementary:

1. Use /oh-my-study-with-me:research to survey the landscape of a topic.
2. Use /oh-my-study-with-me:study to deep dive into the most important source.
3. Use /oh-my-study-with-me:blog to write about your findings.

---

Phase 0: Source Collection

1. Parse user input to detect source types:

   • URLs (http/https) → categorize as web article, GitHub repo, or YouTube
   • Plain text topic → will need WebSearch in Phase 1
   • Mixed (topic + URLs) → use both provided URLs and search for more

2. Classify each source:

   Pattern	Type	Fetch Method
   github.com/{owner}/{repo}	GitHub repo	WebFetch for README + gh CLI for repo structure
   youtube.com/watch or youtu.be/	YouTube	WebFetch (best-effort page info, description)
   Other http(s)://	Web article	WebFetch
   No URL	Topic search	WebSearch → collect top results

3. Show the source plan to the user:

   Research Plan:
   - Topic: [topic]
   - Sources to fetch: [list with types]
   - Additional web search: yes/no
   

   Ask the user to confirm or add/remove sources.

---

Phase 1: Parallel Fetch

1. Web search (if topic provided without sufficient URLs):

   • Use WebSearch with the topic to find 3–5 relevant, high-quality sources.
   • Prefer official documentation, reputable tech blogs, and peer-reviewed content.
   • Show found sources to the user and confirm before fetching.

2. Fetch all sources using appropriate tools:

   • Web articles: WebFetch → extract main content, strip navigation/ads
   • GitHub repos: WebFetch for README; use gh api or WebFetch for repo tree, key source files
   • YouTube: WebFetch for page content (title, description, any available transcript info)
   • Local files: Read tool if user provides file paths

3. Extract key content from each source:

   • Title, author, date (if available)
   • Main arguments or key points
   • Code examples or configurations (if relevant)
   • Data, benchmarks, or measurements cited

4. Report progress to the user as each source is fetched.

---

Phase 2: Cross-Source Analysis

1. Identify common themes across all sources:

   • What concepts appear in multiple sources?
   • Where do sources agree? (consensus points)
   • Where do sources disagree or contradict? (debate points)

2. Find unique insights per source:

   • What does each source contribute that others don't?
   • Are there novel approaches or perspectives?

3. Assess source quality:

   • Is the information current?
   • Does the source cite evidence or provide benchmarks?
   • Is there potential bias? (e.g., vendor documentation)

4. Identify knowledge gaps:

   • What questions remain unanswered across all sources?
   • What aspects need deeper investigation?

---

Phase 3: Research Summary

Produce a structured research output and present it to the user:

# [Topic] Research Summary

## Key Findings
[3–7 bullet points with source attribution]
- Finding 1 [Source A, Source B]
- Finding 2 [Source C]
- ...

## Areas of Consensus
[What most sources agree on]

## Areas of Debate
[Where sources disagree, with each side's argument]

## Unique Insights
[Notable points from individual sources]
- Source A contributes: ...
- Source B contributes: ...

## Knowledge Gaps
[What remains unclear or needs further investigation]

## Recommended Next Steps
- Deep dive: "[specific topic]" with `/swm:study`
- Write about: "[angle]" with `/swm:blog`
- Hands-on: "[technology]" with `/swm:lab`

---

Phase 4: Save & Connect

Output Directory Structure

research-notes/{topic}/
├── summary.md          # Synthesized findings (Phase 3 output)
├── sources.md          # Source list with key excerpts
└── raw/                # Individual source notes (optional, for large research)
    ├── {source-1}.md
    └── {source-2}.md

summary.md

The Phase 3 research summary output.

sources.md

# [Topic] Sources

## Source Index
| # | Title | Type | URL | Key Contribution |
|---|-------|------|-----|-----------------|
| 1 | [title] | Web article | [url] | [one-line summary] |
| 2 | [title] | GitHub repo | [url] | [one-line summary] |

## Detailed Excerpts

### Source 1: [Title]
- **URL**: [url]
- **Author**: [if known]
- **Date**: [if known]
- **Key Points**:
  - ...
- **Notable Quotes/Data**:
  - ...

After saving:

1. Show the saved file paths.
2. Suggest next steps:
   • "Study deeper with /swm:study [url-or-topic]"
   • "Write about it with /swm:blog [topic]"
   • "Set up a lab with /swm:lab [technology]"

---

Important Notes

• Always attribute findings to their sources. Never present information without citation.
• When sources conflict, present both sides fairly. Do not pick a winner without evidence.
• For GitHub repos, focus on README, architecture, and key design decisions — do not dump entire codebases.
• YouTube fetching is best-effort. If transcript is unavailable, work with title, description, and any page content.
• Report progress as you fetch. Research can take time with many sources.
• If a source fails to fetch, note the failure and continue with remaining sources.
• Freeform text input (no URLs, no clear topic) → ask the user to clarify before proceeding.