About
A skill is the portable vehicle that carries that judgment into whatever AI tool an election official already uses — so any official, in any tool, gets responses that reflect how an election expert would actually handle the task.
The Election Skills Library is a collection of AI skills — folders of structured instructions plus supporting reference material — that encode The Elections Group’s election-administration expertise. A skill is not an app, not a hosted service, and not a model. It’s portable expertise that drops into an agent the user already has, and it travels with them: into Claude, ChatGPT, Copilot, or Gemini.
Each skill targets a language-heavy, judgment-driven task that consumes officials’ time and carries real friction — somewhere a grounded skill plus an LLM delivers justifiable value over a vanilla model. If a task wants deterministic correctness — ballot proofing, tabulation, chain-of-custody — it wants software, and we leave it to software.
An existing prompt library is a webpage of prompts to copy and paste. It makes the official do the assembly and hands them no expertise beyond the prompt text. These skills encode the expertise itself — the caveats, the discipline, the jurisdiction-awareness — and carry it into the official’s own tool. The difference is the difference between a recipe card and a cook.
Every skill separates the judgment from the facts it must be true to — and holds each to a different standard. It’s the spine of the whole project.
How to approach the task, the way an election expert would.
AI-drafted · expert-reviewed
The actual law — the facts the judgment grounds itself in.
Sourced from real law · expert-verified
The work is phased so that depth comes before breadth, and subject-matter-expert verification is never the bottleneck that gets skipped to move faster.
Phase 0 — Validate the formatunderway / done
Hand-author one method-dominant agnostic skill and one fact-dominant grounded skill, and test both in a real tool. An afternoon, no infrastructure.
Phase 1 — The portfoliounderway / done
Several agnostic skills plus one or two grounded, self-verifying skills, all grounded and tested against New Jersey. Demo to a friendly official.
Phase 2 — Productionize the pipeline
Document drafting → sourcing → review → testing → versioning. Light tooling for the drafting and sourcing; review and testing stay human.
Phase 3 — Expand
More task categories, then jurisdictions beyond New Jersey — gated by subject-matter-expert verification capacity, never ahead of it.
Phase 4 — Distribution & adoption
Kept deliberately light: structured content that pastes, drops into a wrapper, or downloads. No platform to build or maintain.
A standing commitment
Skills are drafts for human review, never auto-sent. No legal fact ships unverified against a real source. Uncertainty is flagged, not smoothed. Subject-matter experts are the trust gate — and the plan cannot skip them.