Genealogy Research Recipes
One fact-checked skill that gives AI agents sourced recipes for DNA evidence and record searches across countries and archives.
- Source repo
- Kinprove/genealogy-research-recipes
- Stars
- ★ 17
- Last updated
- 1d ago
- License
- NOASSERTION
- Primary language
- Python
- FA score
- 60/100 · Some gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code · Claude.ai
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A hobbyist facing an unknown-paternity lead wants DNA matches organized along WATO lines before contacting anyone.
- Not a fit if
- Teams that need a standalone service or API, not an agent plugin
- Institutions that must stay fully offline, since recipes direct web searches
- Users limited to a plain API client with no plugin marketplace
- Source review
- 60/100 · Some gaps
What does this agent do, and when should you use it?
Genealogy Research Recipes is a single Agent Skill published by Kinprove for Claude, Codex, Hermes Agent and OpenClaw. It packages a large catalog of Markdown "recipes" that walk an agent through evidence and proof judgment, DNA reasoning (WATO, Y-DNA/mtDNA, X-DNA and phasing), record-site retrieval, national church and civil registers, migration, military, probate and land records. Each recipe lives under skills/genealogy-research-recipes/recipes/ and has its sources listed in the parallel sources/ directory. The plugin is plain text plus one offline helper, scripts/lookup_beider_name.py; it registers no tools or hooks, makes no network calls of its own, and needs no Kinprove account or API key. Actual lookups happen through the host agent's own web tools when you ask for research, so names, places and dates travel through that agent to public sites such as FamilySearch or national archives. Installation goes through each host's plugin marketplace, and the repository is itself the kinprove marketplace for Claude. An optional Kinprove MCP connector (https://api.kinprove.io/mcp/v1) lets the agent read your own Kinprove matches and projects.
After install, the agent loads the skill with skill_view(name="genealogy-research-recipes:genealogy-research-recipes") and then targets a specific file via file_path="recipes/<name>.md". The recipe text directs the agent through a task: test evidence against proof standards (evidence-proof-judgment.md, genealogy-standards-audit.md), reason about unknown parentage with WATO (unknown-parentage-wato.md), search matches and records on Ancestry, FamilySearch, MyHeritage, 23andMe, GEDmatch, FamilyTreeDNA or DNA Painter (for example ancestry-dna-match-investigation.md, gedmatch-practical-dna-workflows.md), route by country and confession to church or civil registers (scottish-parish-statutory-record-recovery.md, german-register-image-recovery.md), or chase immigration, probate, land and military files. The bundled script scripts/lookup_beider_name.py resolves Beider given-name variants locally and offline. Output is whatever research the agent performs from the recipe — findings, search paths and source citations. The plugin itself stores and transmits nothing.
- A hobbyist facing an unknown-paternity lead wants DNA matches organized along WATO lines before contacting anyone.
- Someone researching a Jewish family in the Russian Empire needs the right order for name-change, registration and metrical-book searches.
- A researcher who keeps failing to find an ancestor in U.S. census or arrival indexes wants site-specific search tactics rather than retrying keywords.
- A writer or client needs a proof-standard audit of conclusions before publishing a family history.
- A tester comparing the same match across several DNA companies wants the cross-vendor comparison controls spelled out.
- A team already using Claude Code or Codex wants genealogy method captured as a reusable, updatable plugin.
How do you install or deploy this agent?
Claude Code:
/plugin marketplace add Kinprove/genealogy-research-recipes
/plugin install genealogy-research-recipes@kinproveClaude.ai and the Claude desktop app: open Customize → Plugins → Add → Add marketplace, enter Kinprove/genealogy-research-recipes, then add genealogy-research-recipes.
Codex:
codex plugin marketplace add Kinprove/genealogy-research-recipes
codex plugin add genealogy-research-recipes@kinproveHermes Agent:
hermes plugins install Kinprove/genealogy-research-recipes --no-enable
hermes plugins enable genealogy-research-recipesOpenClaw:
openclaw plugins install git:github.com/Kinprove/genealogy-research-recipesRestart the Gateway after installing on OpenClaw. Updates use claude plugin update genealogy-research-recipes@kinprove, codex plugin marketplace upgrade kinprove, hermes plugins update genealogy-research-recipes or openclaw plugins update genealogy-research-recipes.
How do you use this agent?
Once installed, load the skill and point it at a recipe:
skill_view(name="genealogy-research-recipes:genealogy-research-recipes")
skill_view(name="genealogy-research-recipes:genealogy-research-recipes", file_path="recipes/<name>.md")Then state the research question in natural language along with the names, places and dates you already know; the agent follows the recipe's steps. To let it read your own Kinprove data, add the MCP connector:
https://api.kinprove.io/mcp/v1The skill states its release date, and when the installed copy is more than a week old the agent mentions once per conversation that a newer version may exist.
What are this agent's strengths and limitations?
- One plugin carries dozens of recipes across DNA reasoning, record sites, national registers and archives, instead of a separate plugin per site.
- Every recipe is fact-checked with sources collected under skills/genealogy-research-recipes/sources/, so claims can be traced.
- No tools or hooks are registered and no network calls originate from the plugin, so the privacy boundary is explicit and no account is needed.
- Documented install and update commands exist for Claude Code, Claude.ai/desktop, Codex, Hermes Agent and OpenClaw, matching each host's release rhythm.
- The bundled scripts/lookup_beider_name.py resolves given-name variants offline on your machine.
- The catalog is large and spread across many Markdown files, so a newcomer must hunt for the recipe that fits.
- Real searching depends on the host agent's web tools; result quality hinges on that agent and on site accessibility, since the plugin ships no retrieval backend.
- The Hermes plugin catalog pins a reviewed commit and can lag the latest release, and ClawHub installs require separately checking publisher, version and source commit.
- ClawHub, Hermes catalog and GitHub installs are separate distribution paths, which invites version drift.
- Licensed FSL-1.1-MIT: each version stays under FSL terms for two years before becoming MIT, which constrains commercial redistribution.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Genealogy Research Recipes This agent | 60 · Some gaps | Agent plugin / skillFree | ★ 17 | 1d ago | Python | Codex · Claude Code · Claude.ai |
| ex-USSR Kinship Research | 56 · Major gaps | Agent plugin / skillFree + model costs | ★ 2k | 1d ago | Python | Codex · Claude Code |
| AutoResearchClaw | 52 · Major gaps | CLIFree + model costs | ★ 15k | 2mo ago | Python | Codex · Claude Code · OpenAI API |
| MetaClaw | 50 · Major gaps | CLIFree + model costs | ★ 3.5k | 5mo ago | Python | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
README explicitly states the plugin collects, stores and sends nothing, ships only one offline script, and registers no tools or hooks, giving good least-privilege and external-effect control (least_privilege=2, external_effects=2). However it handles highly sensitive DNA and family-privacy data, yet only points to a Kinprove privacy policy with no concrete local retention, redaction or consent mechanism, so sensitive_data_handling=1. User confirmation is only implied (agent uses its own web tools for public sites) with no pre-execution confirmation flow, user_confirmation=1. Rollback relies solely on plugin update commands with no version pinning or data recovery, rollback=1. Dependency security: CI pins the gitleaks image by sha256 and the Hermes runtime by commit, but the gitleaks config depends on a not-yet-exported .gitleaks.toml, an unfinished item, so dependency_security=2. Source attribution is reflected in README and the sources/ structure, source_attribution=2.
Install commands, package name and version in README match package.json, and CI covers multiple runtimes, self_consistency=2. Dependency availability depends on external marketplace/CLI versions ([email protected], [email protected]) and the missing gitleaks config, dependency_availability=1. Failure messages: tests show the lookup script emits structured statuses like manifest_invalid and not_in_index, failure_messages=2.
README covers Claude, Codex, Hermes and OpenClaw installs and a very broad set of audiences/scenarios (DNA, national archives, multilingual records), audience_and_scenarios=3. Capability boundaries are stated (no network, no tools, optional MCP) but unsuited scenarios are not systematically listed, capability_boundaries=2. Trigger precision: as a single skill with many recipes, README does not explain how the agent selects a specific recipe, trigger_precision=1. Environment fit has per-platform steps but depends on specific CLI versions, environment_fit=2.
README organizes many recipes by theme with clear information architecture, information_architecture=3. Install notes are detailed per platform, install_notes=3. Naming stability: package and skill names align, but the large and growing recipe set makes naming consistency hard to confirm statically, naming_stability=2. Examples and FAQ: only install examples, no usage examples or FAQ, examples_and_faq=1. Known limitations are only scattered (e.g., Hermes catalog may lag), not consolidated, known_limitations=1. License is FSL-1.1-MIT, complete but non-standard open source, license=2. Version exists in package.json but no CHANGELOG, versioning_changelog=1. Maintenance responsibility points to MAINTAINERS.md and Kinprove, but publisher identity is unverified, maintenance_responsibility=2.
Recipes target concrete research tasks with moderate output usability, but no output templates or structured result examples are shown, output_usability=2. Fact-checked domain recipes add marginal value over generic LLM knowledge, marginal_value=2. Cost is low (plain text, no network calls), cost_benefit=2.
README claims each recipe is fact-checked with sources listed, and the sources/ structure supports claim_traceability=2. No concrete cross-source corroboration mechanism is described, cross_source_corroboration=1. Tests show the lookup output distinguishes exact/fuzzy and labels results as not evidence, reflecting fact/inference separation, fact_inference_separation=2.
- Handles sensitive DNA and family-privacy data, but README gives no concrete local retention, redaction or consent mechanism; assess privacy risk before use.
- The gitleaks config depends on a not-yet-exported .gitleaks.toml, so CI secret scanning may currently fail; this is an unfinished item.
- Publisher identity is unverified; verify publisher, version and source commit on the marketplace/ClawHub before installing.
- As a single skill with many recipes, it does not explain how the agent precisely triggers a specific recipe, risking selection bias.
- License is FSL-1.1-MIT with competing-use restrictions; confirm compliance before commercial use.