The Elections Group · portable expertise
A library of AI skills that encode how an election expert actually handles a task — grounded in the governing law, honest about uncertainty, and built for human review rather than blind trust. Not an app. Not a hosted service. Portable judgment that drops into the agent an official already has.
Pilot jurisdiction: New Jersey. Every skill is dated to the law it was checked against.
The spine of every skill
A vanilla model can sound authoritative and cite sources — which makes a wrong legal fact more damaging, not less. So every skill separates how to think about the task from the facts it must be true to. The judgment is AI-drafted and expert-reviewed; the authority is sourced from real law and expert-verified.
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
Why a skill, not a prompt
Copy-paste prompt pages make the official do the assembly and give them nothing beyond the words on the page. These skills carry the discipline itself — the caveats, the jurisdiction-awareness, the instinct to flag uncertainty — and travel with the official into their own tool.
The library
No tiers. No skill ranked above another. Each addresses real friction in an official’s day where a grounded skill earns its place over a vanilla model.
Drafts a clear, correct notice telling a voter their mail ballot needs curing — what the defect is, exactly how to fix it, and the real deadline for the voter's jurisdiction, cited to statute.
Rewrites a dense notice, form instruction, or legal paragraph into plain language a voter can actually act on — without changing what it legally means or dropping a required detail.
Turns a procedure or a day-of question into a clear training outline or a calm, correct answer a poll worker can act on — with a built-in instinct to escalate rather than improvise on the law.
Drafts a calm, accurate holding statement for a developing situation — says what is known, what isn't yet, and what the office is doing, without speculating or over-promising.
Drafts a complete, correctly-classified response to a records request — every item addressed, each release/withhold/redact decision cited to the governing exemption.
Distribution, kept light
Copy the skill’s contents into Claude, ChatGPT, Copilot, or Gemini, then ask your question. Works everywhere, no setup.
Install it as a Claude skill, a Custom GPT, or a Copilot agent so it’s always on for your team.
Grab the whole folder — SKILL.md, references, checks, examples — versioned and dated. Yours to keep and review.
The non-negotiable
No legal fact ships unverified against a real source. Uncertainty is flagged, not smoothed. Every skill is tested by an expert against real questions before it’s published — and dated to the law it was checked against.
Read the method & trust model