Research Skill Guide

For ancestry and genealogy research — Claude Code CLI and claude.ai web

---

1. What This Is

The Research Skill is a structured way to get Claude to do serious research for you. You fill in a short form describing what you want to know, and Claude searches the web, reads the sources, and writes a cited report — complete with confidence ratings so you know how solid each finding is.

For genealogy and family history work, this is particularly useful: you get sourced findings (not just Claude’s memory), honest confidence ratings that tell you when a fact is rock-solid vs. needs more digging, and a built-in Next Steps section pointing you toward the specific archives or records most likely to advance your research.

Two ways to use it:

	Claude Code CLI	claude.ai Web
How	Fill in a template file → type /research	Paste a mini-form into a chat → Claude replies with the report
Automation	Fully automatic: parallel research agents, report appended to file	Manual: Claude researches inline, you copy the result
Best for	Repeated use, organized file-based workflow	Quick questions, one-off research, no setup required

Both are legitimate approaches — the web interface isn’t a fallback, it’s a different workflow.

---

2. What You Get

Every research report follows the same structure:

1. Executive Summary — 4–6 sentences covering the most important finding or conclusion
2. Research sections — organized by what you asked for (see goal types in Section 6)
3. Next Steps — 4–6 specific, actionable items: which archives to check, what records likely exist, where to look next
4. Sources — every source labeled and numbered, so you can check the original

Source Labels (in Genealogy Terms)

Label	What It Means
[Primary]	The actual historical document — full text read and cited. Official records, legislation, peer-reviewed papers. The gold standard.
[Secondary]	Established publication with full text — newspaper article, professional journal, library guide. Reliable, but one step removed.
[Partial]	Claude found a reference but couldn’t read the full document — abstract only, or paywalled. Noted honestly; not summarized.
[Unverified]	Source located but couldn’t be fetched at all. Listed so you know it exists.

Confidence Ratings

Each section opens with a confidence rating. This is a feature, not a failure.

Rating	What It Means
High	Backed by primary and secondary sources with full text read
Medium	Mix of full-text and partial sources — solid but incomplete picture
Low	Mostly references and abstracts — the finding is directional, not confirmed
⚠ Speculative	No full-text sources found — Claude is pointing you toward where to look, not reporting facts

For book research, this matters enormously. A “Speculative” rating doesn’t mean the finding is wrong — it means you now know exactly where to go to confirm it. That’s useful intelligence, not a failure.

Example Executive Summary

Confidence: Medium (62)

County Mayo experienced severe population loss during the Great Famine (1845–1852), with mortality and emigration reducing some parishes by 40–60%. Daily life before the famine centred on small-scale potato cultivation, seasonal labour migration to England, and a strong oral tradition in Irish. The Catholic parish system was the primary social institution, with records beginning in the 1820s for most parishes. Civil registration did not begin until 1864, making Catholic baptism and marriage registers the key genealogical source for this period. Griffith’s Valuation (1847–1864) provides the most comprehensive surviving property survey and is heavily indexed online.

---

3. Quick Start — Your First Research Task (~5 minutes)

Using Claude Code CLI

1. Copy _Research-Template.md (in the Research folder) → rename it YYYY-MM-DD_Topic.md
   • Example: 2026-03-14_Mayo-Famine-Daily-Life.md
2. Open the file and fill in the Topic section with one or two sentences describing what you want to know
3. Leave all other settings at their defaults — they’re already set to sensible values
4. Open a Claude Code terminal, type /research, press Enter
5. Wait a few minutes — Claude is running parallel searches. Open the file in Obsidian when it finishes: the report is appended below a --- separator.

Using claude.ai Web

1. Go to claude.ai, start a new chat
2. Make sure web search is enabled (the globe icon in the message bar)
3. Paste the mini-form from Section 5, fill in your topic and context
4. Claude researches inline and returns the full report in the chat
5. Select all the report text → copy → paste into an Obsidian note or Word document
   • Suggested filename: YYYY-MM-DD_Topic.md

---

4. Using Claude Code CLI (Full Automated Skill)

How It Works

When you run /research, the skill:

1. Finds your file — auto-detects the most recently modified .md file in the Research folder (excluding template files starting with _)
2. Reads your parameters — topic, geography, depth, goal, and context from the form
3. Runs a scout scan (recommended default) — a lightweight agent identifies 4–6 research angles most likely to answer your question
4. Dispatches parallel specialist agents — one agent per angle, all searching simultaneously. At Deep depth, these use the most capable model available.
5. Synthesizes — a final agent assembles all findings into one coherent report, renumbers citations globally, calculates confidence scores
6. Appends the report — the completed report is added to your file below ---; your original form is preserved above it

The Template

Location:

• WSL: /mnt/d/FSS/KB/Business/_WorkingOn/Research/_Research-Template.md
• Windows: D:\FSS\KB\Business\_WorkingOn\Research\_Research-Template.md

Important: never edit the template file itself — copy it for each new research task. The skill auto-detects files by modification time; the template’s _ prefix excludes it from detection.

Naming convention: YYYY-MM-DD_Brief-Topic.md

• 2026-03-14_Irish-Famine-Context.md
• 2026-03-14_German-Pennsylvania-Records.md
• 2026-03-14_Ancestry-vs-FamilySearch-Scotland.md

Filling In the Form

The only required field is Topic. Everything else has a sensible default.

Topic guidance:

• Include specific names, places, and dates where you have them
• State what you’re trying to find out, not just a subject area
• Weak: Irish history
• Strong: Daily life in rural County Mayo, Ireland in the 1840s — I'm writing a chapter on what my ancestors' lives would have been like before they emigrated during the Famine

For all other parameters, see Section 6.

Running the Skill

/research

Auto-detects the most recently modified research file.

/research /mnt/d/FSS/KB/Business/_WorkingOn/Research/2026-03-14_My-Topic.md

Run on a specific file.

After It Finishes

• Report appears below the --- separator in your file
• Your original form (topic, parameters) is preserved above the separator
• To re-run or research a different angle: make a new file from the template. Don’t delete the report.

Re-Run Guard

The skill checks for an existing report before starting. If it finds one, it stops and tells you:

Report already exists in [filename]. Delete the # Research Report section (everything from --- onward) to re-run, or create a new research task file.

This prevents accidentally overwriting a completed report.

---

5. Using claude.ai Web Interface

How It’s Different

On the web interface, Claude researches inline within the conversation — no file I/O, no parallel agents. The quality is comparable for most genealogy questions; the main difference is manual steps on your end (paste the request, copy the result).

Requirement: web search must be enabled. Look for the globe icon in the message input area — it should be active (not greyed out). Without it, Claude will draw only on its training knowledge, not current sources.

The Mini-Form

Copy and paste this into a new chat, filling in the bracketed fields:

Research request:

Topic: [Describe what you want researched — specific names, places, dates where known]
Goal: Understand the landscape
Depth: Deep (search thoroughly and cite sources)
Geography: [Global — or specify: County Mayo, Ireland / Pennsylvania, USA / etc.]

Additional context: [What you already know. What you're trying to confirm or find.
Book audience if relevant. What you're trying to decide.
"Focus on free/public sources" if applicable.]

Please structure your report with:
- Executive Summary (with confidence rating)
- Research sections appropriate to the goal
- Next Steps
- Sources labeled Primary / Secondary / Partial with confidence ratings per section

Fill in Topic and Geography at minimum. Additional context significantly improves relevance — see Section 7b.

Setting Up a Genealogy Research Project (Optional)

If you use claude.ai frequently for genealogy, you can add condensed instructions as custom instructions so every chat automatically uses the report format — you won’t need to paste the form each time.

Go to: claude.ai → Settings → Custom Instructions, and paste:

When I ask you to research something, structure your response as a formal research report:

1. Executive Summary (4–6 sentences, confidence rating)
2. Research sections with headings appropriate to what I asked
3. Next Steps (4–6 specific, actionable items)
4. Sources section — label each source: [Primary] for official documents/peer-reviewed papers with full text, [Secondary] for established publications with full text, [Partial] for abstracts or snippets only

Open each section with: *Confidence: High / Medium / Low / Speculative* based on source quality.

I'm writing a family history book. When context includes genealogical research, prioritize: historical archives, civil registration records, church records, census records, land records, emigration/immigration records, and finding aids from national archives.

Saving Results

1. Select all the report text in the chat (Ctrl+A won’t work — click at the start of the report, Shift+click at the end)
2. Copy (Ctrl+C)
3. Paste into Obsidian or Word
4. Save as YYYY-MM-DD_Topic.md

Honest Comparison

	CLI	Web
Setup	Requires Claude Code installed	None — open browser
Steps per research task	Fill form → type /research → open file	Paste form → wait → copy result → save
Parallel research agents	Yes — faster, more thorough	No — single-pass
File organization	Automatic	Manual
Quality for typical genealogy questions	Excellent	Comparable
Best for	Regular use, building a research archive	Quick questions, sharing with others, no install

---

6. Parameter Reference

Geography

Default: Global

What it does: Narrows Claude’s search to sources relevant to a specific region.

When to change: Almost always for genealogy. Local history sources, regional archives, and county-level records are dramatically more relevant than global results when you’re researching a specific place.

Genealogy examples:

• County Mayo, Ireland for Famine-era Irish research
• Pennsylvania, USA for German immigrant records
• Scottish Highlands, Scotland for pre-1855 records
• Global for comparing genealogy databases or researching migration patterns

---

Depth

Default: Deep

What it does: Controls how thoroughly Claude searches. Deep = 3–4 parallel searches per angle, full-text source fetching, comprehensive report. Moderate = 2 searches per angle, top 2 sources only, shorter report.

When to change: Use Moderate for quick fact-checks (“when did civil registration start in Ireland?”) or preliminary scoping before a deeper dive. Use Deep for anything going into the book.

---

Search Strategy

Default: Scout first (recommended)

What it does: Scout first runs a quick angle-identification pass before the deep research — Claude identifies the 4–6 most useful angles to research, then pursues each one in depth. Direct dive skips the scout and goes straight to research.

When to change: Leave at Scout first unless the topic is extremely specific and focused. The scout phase is fast and significantly improves the relevance of the deep dive, especially for vague or broad genealogy questions.

---

Goal

Default: Understand the landscape

What it does: Shapes what sections appear in the report and what Claude focuses on. See the decision table in Section 7c.

---

Sources

Default: Cited in report

What it does: When checked, every finding is cited with a source label and confidence rating.

When to change: Never, for book research. You need provenance — the ability to say where every fact came from. “No citations needed” is only appropriate for background reading where you don’t care about sourcing.

---

AI Sources

Default: Claude only

What it does: Adds parallel queries to ChatGPT, Gemini, or Grok, merging their findings into the report. Requires API keys to be configured in your environment.

When to change: Leave at default. Multi-AI requires API key setup and is more complex to configure. Claude with web search is sufficient for genealogy research.

---

Output

Default: Append to task file (always active)

What it does: Report is appended to the same file as your research form. “Also export to Google Doc” creates a shareable Google Doc — requires Google service account credentials.

When to change: Leave at default. Google Doc export requires credentials setup. Sharing is easier by copying the file.

---

7. Tips for Ancestry Research

7a. Writing a Good Topic

Specific beats vague — every time. Claude produces much better research when it has names, places, and dates to anchor the search.

Weak topics:

• Irish history
• German immigrants
• Scottish records

Strong topics:

• Daily life in rural County Mayo, Ireland in the 1840s — I'm writing a chapter on what life was like before the Famine emigration. Focus on farming, community, religion, and what the landscape looked like.
• What records exist for German immigrants who arrived in Pennsylvania between 1860 and 1880? I'm looking for naturalization records, church records, and land records specifically.
• Which genealogy database — Ancestry, FindMyPast, or FamilySearch — has the best coverage for Scottish Highlands records before civil registration began in 1855?

The extra sentences cost you nothing but dramatically change what Claude looks for.

---

7b. Using Additional Context Effectively

The Additional Context field is where you give Claude the lens through which to interpret your research. Things that help:

• Your book audience — “writing for general readers, not genealogists — assume no prior knowledge of Irish record systems”
• What you already know — “I know Thomas Brennan emigrated from Mayo in 1847 and arrived in Boston. I’m trying to understand what life was like in the parish he came from.”
• What you’re trying to confirm — “Family story says he was a blacksmith. I want to know how common that was in this region and period, and what records might confirm an occupation.”
• Source constraints — “Focus on free and public sources — I don’t have subscriptions to pay databases.”
• What to avoid — “Skip general Irish history — I know the broad strokes. Focus specifically on County Mayo and local parish records.”

---

7c. Goal Decision Table

What you want to know	Goal
Historical period/place background for book context	Understand the landscape
What records exist for a person, place, or period	Identify opportunities
Verify or fact-check a family story	Understand the landscape
Evaluate which genealogy database to use	Decision support
How reliable a particular record type is	Assess risks
A specific historical event (famine, migration wave, war)	Understand the landscape
Whether a research strategy makes sense	Decision support

You can check multiple goals — the report will cover all of them in sequence.

---

7d. What the Skill Can and Can’t Do

Can do:

• Search public web sources, archive finding aids, historical society pages
• Find Wikipedia, academic articles, library guides, digital collections
• Tell you what record types exist and where they’re held
• Research historical context — social history, migration patterns, daily life
• Compare databases and explain coverage gaps
• Find digitization status of specific record sets

Can’t do:

• Log into Ancestry, FindMyPast, FamilySearch, or any subscription service
• Access private family trees
• Transcribe or interpret documents you upload
• Search within databases (it can tell you the database exists and what’s in it, not search it for you)

Think of it as doing the reconnaissance — when you’re done, you know exactly what to look for, where to look, and what questions to bring to the archive or database.

---

8. Example Research Tasks

Three complete examples showing filled-in template blocks with parameter notes.

---

Example 1: Historical Context for a Book Chapter

Goal: Write an immersive chapter on daily life in County Mayo before and during the Famine.

## Topic
Daily life in rural County Mayo, Ireland in the 1840s — specifically: what did people eat,
how did they farm, what did their homes look like, what was the role of the Catholic church,
and how did the community function socially before and during the Great Famine (1845-1852)?
I'm writing a family history book and need accurate background for a chapter set in this
period and place.

## Parameters

Geography: Specify: County Mayo, Ireland
Depth: [x] Deep
Search Strategy: [x] Scout first
Goal: [x] Understand the landscape
Sources: [x] Cited in report

## Additional Context
Audience is general readers — no prior knowledge of Irish history assumed.
Ancestors were small tenant farmers, likely in a western parish.
I need enough detail to write vivid, accurate narrative, not a textbook summary.
Focus on what daily life felt like, not political history.

Parameter notes: Geography narrowed to county level for maximum local relevance. Deep depth for book research — this goes in print. Understand the landscape is the right goal for background/context chapters.

---

Example 2: Records Discovery

Goal: Find out what records exist before visiting the archive.

## Topic
What genealogical records exist for German immigrants who settled in Pennsylvania
between 1860 and 1880? Specifically looking for: naturalization records, church records
(Lutheran and Catholic), land records, and census records that might record birthplace
at the county or village level in Germany.

## Parameters

Geography: Specify: Pennsylvania, USA
Depth: [x] Moderate
Search Strategy: [x] Scout first
Goal: [x] Identify opportunities
Sources: [x] Cited in report

## Additional Context
Ancestor arrived approximately 1865-1870, settled in Berks or Lancaster County area.
Occupation unknown. I want to know: which record sets are digitized and searchable online
vs. held in physical archives, and which archives in Pennsylvania hold German immigrant records
for this period.

Parameter notes: Moderate depth for a recon pass — I want a map of what exists before investing time. Identify opportunities is the right goal when the question is “what records exist and where?” Geography set to Pennsylvania rather than Global because I want PA-specific archive information.

---

Example 3: Database Comparison

Goal: Decide which subscription to buy for a specific research need.

## Topic
Comparing genealogy databases for Scottish Highlands research before 1855
(before civil registration). Which of Ancestry, FindMyPast, or FamilySearch has
the best coverage for: Old Parish Registers (baptisms, marriages, burials),
heritors' records, estate papers, and sasine registers for Inverness-shire
and Ross-shire?

## Parameters

Geography: Specify: Scottish Highlands, Scotland
Depth: [x] Moderate
Search Strategy: [x] Scout first
Goal: [x] Decision support
Sources: [x] Cited in report

## Additional Context
I have a FamilySearch free account. Deciding whether Ancestry or FindMyPast is worth
paying for. Main focus is the period 1750-1855 in Inverness-shire. ScotlandsPeople is
a known option — include it in the comparison if relevant.

Parameter notes: Decision support goal produces an Options Analysis and a directional Recommendation, not just a summary. Moderate depth is sufficient for a “which database?” question. Geography narrowed to the specific region to get coverage information relevant to those counties, not Scotland generally.

---

9. Troubleshooting / Common Questions

The skill can’t find my file / picks the wrong file

The skill auto-detects the most recently modified .md file in the Research folder (excluding files starting with _). If it picks the wrong file, specify the path explicitly:

/research /mnt/d/FSS/KB/Business/_WorkingOn/Research/2026-03-14_My-Topic.md

Or open the file in Obsidian and save it (touching the modification time) before running /research.

---

I got “Confidence: Speculative” — did something go wrong?

No. Speculative means Claude couldn’t find full-text sources for that section — only references and abstracts. It’s being honest with you rather than making things up. Check the Next Steps section: it will tell you where to look to find the actual sources. For genealogy, Speculative often appears for very local or pre-digitization records — exactly the stuff that’s in physical archives, not online.

---

I want to research the same topic again with a different angle

Create a new file from the template. Don’t try to delete the report section — make a fresh task file with a refined topic. The old file with its report is a permanent record of what you found; keep it.

---

The report is too short or too general

Two common causes:

1. Topic too vague — add specific names, places, dates, and what you’re trying to find out
2. Geography set to Global — narrow it to the specific region (county, state, country)

Also check Depth is set to Deep, not Moderate.

---

The web interface report isn’t structured the way I want

Make sure you pasted the full mini-form from Section 5, including the “Please structure your report with:” instruction at the bottom. If you’ve set up custom instructions (Section 5), those will apply automatically to every chat. If the output still isn’t structured, ask Claude directly: “Please reformat that as a research report with Executive Summary, confidence ratings per section, Next Steps, and a labeled Sources list.”

---

Can I share research files with others?

Yes — the .md files are plain text and open in any markdown editor or text editor. For non-Obsidian users, share via Google Docs (copy and paste the contents) or export to PDF from Obsidian.

---

10. File Organization

Where Everything Lives

File	WSL Path	Windows Path
Template	/mnt/d/FSS/KB/Business/_WorkingOn/Research/_Research-Template.md	D:\FSS\KB\Business\_WorkingOn\Research\_Research-Template.md
Research files	/mnt/d/FSS/KB/Business/_WorkingOn/Research/	D:\FSS\KB\Business\_WorkingOn\Research\

Naming Convention

YYYY-MM-DD_Brief-Topic.md

Examples:

• 2026-03-14_Mayo-Famine-Daily-Life.md
• 2026-03-14_German-Pennsylvania-Records-1860-1880.md
• 2026-03-14_ScottishHighlands-Database-Comparison.md

The date prefix keeps files sorted chronologically in Obsidian’s file explorer.

Archiving Completed Research

Once you’ve used a research file and incorporated the findings into your book, move it to a subfolder:

Research/
  Archive/
    2026-03-14_Mayo-Famine-Daily-Life.md
    2026-03-14_German-Pennsylvania-Records.md

This keeps the active Research folder clean so auto-detection works reliably.

Auto-Detection Behaviour

When you run /research without specifying a file, the skill:

1. Scans all .md files in the Research folder (top level only — not subfolders)
2. Excludes any file starting with _ (template files)
3. Selects the most recently modified file

This means: open your task file in Obsidian and make any edit (even a space) to ensure it’s selected, or specify the path explicitly if there’s any ambiguity.

---

Guide written for Claude Code CLI (skill version: /research) and claude.ai web interface. Template location: D:\FSS\KB\Business\_WorkingOn\Research\_Research-Template.md