omni

Use when manually triggering a bulk run of meeting briefings — dispatching a cohort of meeting_briefing agent jobs against prod via the imminence-gated endpoint, monitoring them to completion, and producing the outcome/cost/runtime analysis and CSV export. Irreversible prod spend (~$6 per dispatched briefing); requires gp-admin AWS access and this repo's packages/gp-api.

Open a PR following GoodParty conventions and drive the delegate-reviewer bot to approval autonomously. Use when the user says "open a PR", "ship this", "create a PR", "create a PR and address delegate", or otherwise wants changes turned into an approved PR in omni.

Write a GoodParty Technical Design Doc (TDD) for a big-rock feature: a human-reviewable design of the system changes, posted to the ClickUp TDD folder, with deeper detail split into an Implementation Notes subpage. Use when the user wants to design a feature, write a tech design / design doc / TDD, or turn a scoped problem into a reviewable technical design. Not for small features or bugs.

Use when you want to visually review a cohort of meeting_briefing artifacts rendered in the real webapp UI — "pop the gallery", "view the briefings", "let me page through these briefings". Pulls artifact JSON from S3 into a local dev-only gallery at http://localhost:4000/dev/briefings that renders each briefing through the production briefing components with prev/next paging. Dev-only, read-only, no prod spend.

Use when you want to spot-check a cohort of community-issue artifacts (top_community_issues + trending_issues) locally — "pop the issue gallery", "view the issues", "let me eyeball these issue artifacts". Pulls artifact JSON from S3 into a local dev-only viewer at http://localhost:4000/dev/issues that renders ranked issues with category/priority chips, overview text, and sources. Dev-only, read-only, no prod spend.

Use when you need to analyze the cost of CAP agent experiment runs or cohorts (meeting_briefing, community_issues, etc.) — what a job costs, where in a run the spend happens, which runs drive the bulk of spend, and how cost scales with district size. Resolves a scope from plain English, extracts per-turn cost-weighted data to a cached parquet, and produces cost-distribution tables and plots. Population-not-sample, distributions-not-point-estimates, and dollars always traced to the invoice-validated DB costUsd.

Write an analytics event's governance metadata (purpose, status, supersession lineage, product tag) into Amplitude — when adding a new client event, retiring or superseding one, or enriching/backfilling an existing event that is missing metadata. Called by instrument-analytics-event after it registers a new event, and runnable standalone for governance updates after an audit. Client (Amplitude) events only.

Enter prototype mode to build, extend, or explore a UI prototype in packages/prototypes. Use when the user says "prototype", "prototype mode", "build a mockup", "build a prototype UI", "work on a prototype", "design mode", or wants to iterate on a visual concept without production constraints.

Build/author a CAP agent experiment (manifest + instruction) from a runbook and test it in the cloud. Use when turning a locally-validated runbook into a deployed PMF experiment (experiments/<slug>/manifest.json + instruction.md), authoring the broker-quirk CRITICAL RULES, sizing scope/output_schema, or publishing and dispatching it in dev to verify the artifact.

Add a Segment/Amplitude analytics event when building a feature — in the webapp (frontend, via trackEvent) or in gp-api (backend, via AnalyticsService). Decide whether it earns an event, which side fires it, name it per the standard, register it, and fire it. For frontend events it then hands off to the event-metadata skill to record governance metadata, and it also handles client events a change removes (routing them to retirement).

Add a new HTTP endpoint to a feature module following the controller/services/schemas convention, PrismaBase pattern, and Zod response validation. Use when adding any new route to gp-api.

Create an Amplitude feature flag (feature toggle) for GoodParty in both dev and prod via the Amplitude MCP. Use when the user wants to make a new feature flag, add a toggle, set up a flag to roll out a feature, or asks to turn on / enable / release a flag in prod (the skill creates the flag and explains that releasing in prod is a manual Amplitude UI step). Covers GoodParty's dev and prod Amplitude projects.

Use when you need a cohort of test users in dev that mirrors the current PRODUCTION user population's distribution across district size (total registered voters), for either the Win (candidate) or Serve (elected-official) product. Creates real dev users bound to real ICP offices (so downstream ICP-gated agent dispatch fires), and emits an org-slug manifest to hand off to a dispatch skill (e.g. bulk-community-issues-cohort). Analysis is Databricks-only; creation is headless (no Playwright). Does NOT dispatch — it stops at the manifest.

Scaffold a new prototype in packages/prototypes. Use when the user says "new prototype", "create a prototype", "scaffold a prototype", "make a shell with X tabs", "make a prototype with X and Y screens", or wants a fresh prototype folder generated from a tab list.

Use when manually dispatching a cohort of community-issue agent jobs (top_community_issues + trending_issues) against prod via the AdminOrM2MGuard-protected endpoint POST /v1/community-issues/dispatch, monitoring them to completion, and producing the outcome/cost/runtime analysis and CSV export. Lower spend than briefings (~$1-2 per org pair of runs); requires gp-admin AWS access and this repo's packages/gp-api.