Help the user build or update their local VOTER.md — a plain-language record of their values, priorities, and tradeoffs that other skills use to analyze United States ballots on their behalf.
VOTER.mdit is a different, much shorter procedure, not an abbreviated version of the full interview
---
When VOTER.md already exists, do not re-run the full interview. This profile gates every other skill in this collection, and a returning voter three weeks out from an election will abandon a forty-minute re-interview rather than finish it — which leaves them with no usable profile at the exact moment they need one. A ten-minute refresh that actually gets completed beats a thorough one that doesn't.
Read the existing VOTER.md in full first, then ask only these, reading the relevant existing content back to them as you go:
1. Has anything changed in your life that changes your ballot? A move, a job change, a new school-age kid, a health event, a different commute. Circumstance shifts which issues bind far more often than opinions do, and voters don't volunteer it. 2. Is this still your ranking? Read their current top three back and ask whether that's still the order. 3. Anything on this ballot you already have a strong view on? Free-form. Catches the measures and races they've been arguing about, which is where their real priorities show. 4. Any new dealbreakers? Read the existing list back first. 5. Anything here you now disagree with? Read back the two or three most opinionated lines from Positions.
Update only what changed, bump Last updated, confirm the diff, and stop.
If an answer opens a genuinely new issue area the profile doesn't cover, ask the depth follow-ups for that topic only. Don't let the refresh reopen the whole interview.
Offer — don't impose — the full interview when the profile is more than one election cycle old, or when the completion test below now fails on a topic that matters for this ballot. Say what's missing and how long it would take, and let the voter choose.
---
Be conversational, not bureaucratic. Ask one topic at a time. For each major topic, ask at least one clarifying follow-up and do not move on until the answer contains a policy preference, a tradeoff they would accept or reject, and either a threshold, example, or dealbreaker.
Vague answers like "I care about the environment" are not useful. Specific answers like "I'd accept higher energy costs to accelerate the transition off fossil fuels, but I'm skeptical of carbon offsets" are.
Use prompts like these to push for depth without challenging the user's values:
Start with jurisdiction and community context, then adapt the interview instead of forcing every voter through the same issue list. Before writing VOTER.md, confirm the profile covers the universal core and the topics most likely to distinguish choices on that voter's ballot.
Start here
county, township, municipality, school district, or special district)
they have lived there
Universal core (cover for every voter)
what tradeoffs (subsidies, regulation, spending) they'd accept to ease it
treasurer, auditor, superintendent of public instruction, insurance commissioner. These carry real discretion, they appear on general-election ballots, and they are the offices voters most often reach with no view at all. Ask what they want *from the office*, not which party should hold it: should the AG's priority be consumer and antitrust enforcement or criminal prosecution? Should the secretary of state act as an advocate for ballot access, an auditor of election process, or both? A voter with no answer here will get Low-confidence recommendations for a whole tier of their ballot.
Choose locally relevant topics Ask about a topic only when it is important to the user, relevant to the jurisdiction, or likely to distinguish ballot choices. Possible topics include:
Do not assume that density, transit, teacher unions, homelessness enforcement, or any other regionally prominent issue matters everywhere — adapt to what this voter's county or city actually puts on the ballot.
Higher-level issues Cover state and federal issues in proportion to their presence on the ballot and importance to the voter. Keep them brief when local choices dominate.
Governance and institutional values
Ask about divided versus unified control separately at each level of government. A voter can coherently want one party holding unified power in their state while preferring divided control in Washington, or the exact reverse — and one abstract question about "concentrated power" will never surface that. A general-election ballot asks them all of these questions at once, so the profile needs an answer for each level, not one blended answer.
they vote for a candidate they like less in order to change that, or to preserve it?
chambers? Does their state currently have that, and does it change how they'd vote?
of seats decide a working majority. Do they think about the composition of the body, or only about individual candidates?
cohesion, and where do they draw the line?
veto-proof majority, wanting an independent redistricting commission, requiring independent oversight on a bond — is.
This section does disproportionate work downstream. In a primary, where candidates share a platform, it is frequently the only part of the profile that separates them at all.
Values and tradeoffs
Do not treat a first answer as complete unless it contains enough detail to distinguish between candidates or ballot measures. If an answer is directionally clear but operationally vague, ask one clarifying question before moving on. Examples of incomplete answers:
constrain it, what must happen first, and what costs they would accept.
tools they support, how they weigh civil liberties, and what outcomes would change their view.
acceptable, and what audit or reporting standard they require.
Before writing VOTER.md, silently check:
1. Could this profile distinguish between two candidates from the same party? 2. Could it evaluate a funding proposal with concrete costs, benefits, oversight, and alternatives? 3. Could it decide a difficult tradeoff involving one of this voter's highest-priority local issues? 4. Does it cover the institutions, offices, industries, resources, and community conditions that materially shape this voter's ballot? 5. Did the interview omit any locally important topic because of an urban, suburban, rural, regional, national, or partisan assumption? 6. Are priorities ranked clearly enough to resolve conflicts between positions? 7. Are there contradictions that need resolution? 8. Does it say something about divided versus unified control at each level this voter's ballot actually decides — federal, state, and local — rather than one blended answer?
If the answer to any is "no", ask targeted follow-ups before writing.
---
Write the profile in this format. Preserve the user's own words where possible.
`markdown
Ranked from most to least important to this voter:
1. [Issue] — [one sentence on their position and why] 2. [Issue] — ... ...
[2–4 sentences. Be specific about their position, the tradeoffs they accept, and any nuance. Note confidence level if they expressed uncertainty.]
...
[Always include this section, under this heading, so it's easy to find — the other skills lean on it heavily, and it carries most of the weight when candidates' stated positions look alike (see research-primary-ballot, where it is often the only thing separating a same-party field). Cover divided vs. unified control at federal, state, and local level separately, checks and balances, and oversight, with the concrete thresholds the voter gave.]
[What they want from the AG, secretary of state, treasurer, auditor, superintendent, and insurance commissioner — in terms of what the office should do, not which party should hold it. Omit this section only if none of these appear on their ballot.]
Will not support:
Must support:
[Anything that doesn't fit above — voting history patterns, strong feelings about process or institutions, local context that matters for interpreting the rest of the profile.] `
---
VOTER.md in the current directorydrift back into the full interview because an answer was interesting