Genealogy becomes an agent skill

A genealogy research workflow for former Soviet records is now packaged as a skill for AI agents, including Codex and Claude Code. It’s built to work across church books, civil records, military files, deportation and repression records, and unindexed archive scans. The project showed 745 stars, a striking level of interest for such a specialist tool. You can read the project page.

The practical idea is to make the agent follow a research process, not just produce a family story. It keeps records separate from interpretations, tracks where each finding came from, and labels what is proven, uncertain or contradicted. It also checks that conclusions match the evidence register. The workflow covers all 15 former Soviet republics, and is designed to answer in the user’s language while quoting original records with translations.

For a business owner, the wider point is that agents can be given specialist methods and guardrails for a narrow job, rather than being asked to improvise from a broad prompt. This example is about family history, but the same principle matters wherever work depends on traceable evidence and careful handling of uncertainty.