The plugin ships one always-on skill, three slash commands, and two subagents. Answers end with the card that fits the result.
/cicada-guide:research-legislation <bill number or topic> [state] [year]
A sourced brief on one bill: identification (number, state, session, title, and status with its date), what the bill does, sponsors, legislative history with roll-call counts, how members voted when you ask, and sources. It closes with what the brief could not establish. Give it a topic instead and it lists one line per matching bill, then offers to brief one.
Ends with the bill card (show_bill) and a plain-language headline and summary. A topic list skips the card until you pick a bill.
/cicada-guide:voting-record <legislator name> [state] [bill] [session or date range]
Answers three kinds of question: one legislator's votes over time, one roll call broken down by party, or one legislator's vote on one bill. Every answer states the window covered and the filters applied. Absent and not voting are counted separately, not treated as positions.
Ends with the legislator record (show_person_record) for a voting history, or the bill card for a question about one bill or roll call.
/cicada-guide:contact-legislator <name> [state]
Who a legislator is and how to reach them: the seat (office, state, chamber, and district) as recorded, and every contact detail on record, naming each one that isn't. It never contacts anyone and never guesses an address or number. Scope is state legislators only, not members of Congress, governors, or local officials.
Ends with the contact card (show_official).
get-legislation loads by itself on any question about state bills, legislators, roll calls, or sessions. It carries the rules that keep answers accurate: which tool fits which question, how to page through results, how to turn vote records into legislator names, and how to apply project settings.
It declines requests to rank, grade, score, or rate legislators, or to predict how they will vote. Only recorded votes are reported.
Claude dispatches these on its own when a request matches. Each returns one consolidated report instead of filling the conversation with intermediate tool calls.
Assembles a full sourced brief on one bill: record, text, sponsors, roll calls, and the vote breakdown. It covers the same ground as /cicada-guide:research-legislation, run autonomously. Because it can't ask questions mid-run, it returns the candidates instead of picking when the bill is ambiguous.
Resolves an ambiguous legislator name to one person by probing each candidate's vote history for the state they serve and reading their recorded seat. It returns RESOLVED, AMBIGUOUS, or NOT FOUND, and never guesses.
A subagent's output isn't shown as a card, so each one names the card that fits its result, and Claude shows it in the conversation.
In hosts that render MCP Apps, answers end with the fitting card without being asked. Hosts that can't render cards get a shorter text version, so the written answer is complete on its own.
The written answer doesn't repeat what the card shows. The headline and summary on a bill card are Claude's plain-language reading of the bill, not an official summary, and never predict passage.
