Weekly PolitiClaw digest style: what the user's reps did this week against the stances they declared. One message, readable in ~60 seconds, facts not cheerleading, built on the tool's tier-1/tier-2/tail bundling and a mandatory "things you might be surprised by" section.
How to map each contest on the user's ballot against the stances they declared — showing where candidates and incumbents align with their values and where they don't — without prescribing votes. Enforces no prescriptive recommendations and strict guardrails on when LLM-search-derived narrative is allowed.
How to drive `politiclaw_configure` end-to-end — a single staged tool that walks the user through address → top issues → monitoring mode → accountability → final monitoring contract. The tool re-derives the current stage from DB state on every call; you just relay the prompt and collect the next answer.
How to surface bills, votes, and committee events that touch the stances the user declared — so they can see when their reps are (or aren't) representing them — without drifting into advocacy. Decides when to alert, when to summarize, and when to stay silent. Enforces the anti-echo-chamber rule that every substantial summary must include a dissenting or complicating view when one exists, and the four-class alert shape that every proactive message follows (headline, why-it-matters, what-happened, optional next).
How to help the user close the loop on a stance-gap the monitoring loop surfaced: put an accountability question in front of the rep in the user's own words. Covers when to offer a letter or call draft, how to use politiclaw_draft_outreach, and the firm rule that PolitiClaw never sends mail or dials — the user sends from their own client.