PolicyEngine

PolicyEngine/policyengine-claude

Agent Claude Code

Stage-6 agent. Given a deviation signature (where the microsim disagrees with the prior), identifies the most likely calibration targets / imputed variables driving the gap. Reads Microcosm calibration targets and known-issues to produce a ranked diagnostic checklist. The single highest-expertise stage — converts…

PolicyEngine/policyengine-claude

Agent Claude Code

Runs a PolicyEngine microsimulation given a reform-dict and jurisdiction. Returns budgetary, poverty, distributional, and geographic impacts. Stripped of any tracker-specific DB-write logic — pure computation. Generalizes the legislative-tracker impact-calculator.

PolicyEngine/policyengine-claude

Agent Claude Code

Stage 5.5 of /analyze-policy. When the original reform has no exact-shape external comparator, runs the closest-shape externally-scored reforms through the PE model as "mirror-shape" validation runs. If the model reproduces the published external scores within ±25%, it corroborates the data calibration and lifts…

PolicyEngine/policyengine-claude

Agent Claude Code

Independent reviewer for reform analyses. Mode 1 (predict) reads ONLY the reform provisions and predicts expected outcomes from first principles — before any PolicyEngine numbers exist. Mode 2 (interrogate) compares those blind predictions against the actual microsim results and classifies every divergence as a…

PolicyEngine/policyengine-claude

Agent Claude Code

Given a reform provision, locates the corresponding PolicyEngine parameter YAML path(s), checks current values, and emits a reform-dict snippet ready for the PolicyEngine API. Federal + state coverage. Generalizes the legislative-tracker param-mapper.

PolicyEngine/policyengine-claude

Agent Claude Code

Fetches and analyzes policy text (legislative bills, executive orders, agency rules, ARPA-style proposals, white papers) and extracts the specific numeric/structural changes to tax and benefit programs. Generalizes the legislative-tracker bill-researcher to handle federal + state, draft + enacted, US + UK + Canada.

PolicyEngine/policyengine-claude

Agent Claude Code needs its repo

Finds prior scored reforms to anchor a new analysis — PolicyEngine's own published scores first, then official fiscal notes (state legislative fiscal offices, JCT, CBO, CRFB), then think-tank analyses (Tax Foundation, ITEP, CBPP, TPC). Generalizes the legislative-tracker fiscal-finder by adding the PolicyEngine…

PolicyEngine/policyengine-claude

Agent Claude Code

Validates that all parameters have proper references that corroborate the values.

PolicyEngine/policyengine-claude

Agent Claude Code

Classifies a reform as PARAMETRIC (existing parameter edit), STRUCTURAL (needs new variable/formula logic), or NOT-POSSIBLE (out of model scope). The Stage-2 gate that decides whether to proceed with microsim or stop and write a "not modelable" note.

PolicyEngine/policyengine-claude

Agent Claude Code

Stage-5 comparator. Takes the microsim output and the prior-scores anchor, returns PASS (within expected band, write up the result) or INVESTIGATE (mismatch, trigger calibration-diagnostics). Normalizes single-year vs 10-year scores and accounts for dataset-version drift.

PolicyEngine/policyengine-claude

Agent Claude Code

Generates a mechanical, neutral, non-advocacy description of a reform — both a 1-2 sentence summary and a structured provisions array grouped by program. Used for blog posts, dashboard descriptions, PR write-ups. Generalizes the legislative-tracker model-describer.

PolicyEngine/policyengine-claude

Command Claude Code

End-to-end policy analysis pipeline. Takes a reform (bill number, URL, or description), classifies whether it's modelable, finds prior PE scores, runs the microsim, compares to priors, and — if the run mismatches — diagnoses which calibration target is likely driving the gap.

PolicyEngine/policyengine-claude

Command Claude Code

Audit a web app's SEO in a single read-only pass — meta tags, Open Graph, canonical, crawlability, plus the PolicyEngine embed/canonical gotchas.

PolicyEngine/policyengine-claude

Command Claude Code

Audit a state income tax PR's parameter values against official PDF sources (read-only, no code changes).

PolicyEngine/policyengine-claude

Command Claude Code

Orchestrates multi-agent workflow to backdate and quality-improve existing state program parameters.

PolicyEngine/policyengine-claude

Command Claude Code needs its repo

Build new UI components for @policyengine/ui-kit — with design token validation, tests, visual preview, and PR creation.

PolicyEngine/policyengine-claude

Command Claude Code

Create PR as draft, wait for CI to pass, then mark ready for review (solves the "I'll check back later" problem).

PolicyEngine/policyengine-claude

Command Claude Code

Deploys a PolicyEngine dashboard to Vercel (and optionally Modal) and registers it in the app.

PolicyEngine/policyengine-claude

Command Claude Code

Orchestrates multi-agent workflow to implement contributed policy reforms (proposed bills, policy experiments).

PolicyEngine/policyengine-claude

Command Claude Code

Generate social images and posts from a blog post or announcement.

PolicyEngine/policyengine-claude

Command Claude Code

Calculate benefits, taxes, and net income for a single household under baseline or a reform — the standalone entry for household-level questions.

PolicyEngine/policyengine-claude

Command Claude Code

Implements the model extension behind a STRUCTURAL /analyze-policy verdict — baseline change for enacted law, gov/contrib reform for proposals — opens the policyengine-{country} PR, and records the re-run that completes the original analysis.

PolicyEngine/policyengine-claude

Command Claude Code needs its repo

Scaffold a single-purpose interactive PolicyEngine tool (calculator or widget) — Next.js 14 + Tailwind 4 + ui-kit + embedding boilerplate. For multi-page policy analysis dashboards, use /create-dashboard.

PolicyEngine/policyengine-claude

Command Claude Code

Search the local analyses archive for prior /analyze-policy runs — did we already score this? What did we conclude?