matiasportugau-ui

Turn a repeated agent or prod failure into a permanent harness improvement (test/hook/script + optional AGENTS line + RULE-PROVENANCE). Use when user says harness-ratchet, ratchet this failure, never again, or after a live-fix that should become structural.

Complete Shopify store connector for Calculadora BMC: OAuth/PKCE, Admin GraphQL, Storefront API, webhooks, catalog sync, products/variants/inventory/prices, orders, customers, discounts, metafields, theme/Liquid/design, Checkout Extensibility, Shopify Functions, app embeds, and Partner Dashboard options. Use when the user wants to develop Shopify design, functionality, catalog, storefront, theme, app extensions, or any Shopify option from this repo; also for shopify connector, conectar Shopify, tienda Shopify, theme Liquid, Online Store 2.0, or full-store Shopify integration (broader than shopify-integration-v4 questions/quotes).

Implements and maintains the Shopify questions/quotes flow that replaces Mercado Libre: OAuth 2.0/PKCE, webhooks→Google Sheet sync, admin question list, LLM/rule-based suggested replies, one-click approve & send, auto-reply scheduler (UTC-3). Use when working on Shopify backend, Cloud Run, Firestore, Sheets API, HMAC validation, or LATAM Spanish reply formatting for Uruguay.

Orchestrates local Agent Review Issue and Fix: Bugbot review (optional), bmc-issue-fix-reviewer fixes, optional security-review, expert-debug cleanup, and gate:local. Use when user says /issue-and-fix, issue and fix, revisar y arreglar, agent review local, or Cursor Agent Review insufficient funds.

Local alternative to Cursor Agent Review (Issue and Fix): reviews branch or uncommitted diffs against project rules, finds bugs/security/logic issues, and applies fixes. Runs gate:local after fixes. Use when Agent Review fails (insufficient funds), user says issue and fix, revisar y arreglar, or review and fix changes.

Remotion video creation expertise — compositions, useCurrentFrame, interpolate, Sequences, assets via staticFile, render CLI, captions, audio, FFmpeg. Use when working with .tsx files importing from "remotion" or "@remotion/*", building video compositions in React, configuring Root.tsx, rendering frames/stills, adding audio/video/captions, or when the user mentions Remotion, Remotion Studio, or video composition. Do NOT use for general React UI work, non-video animation, or pure ffmpeg pipelines without React.

Run promptfoo evals against BMC chat prompts (Panelin agent, autolearn, ML responses). Use when the user wants to test, compare, or regression-check LLM prompts/models in this repo.

Apply the Anthropic "Applied AI" frontend visual language to BMC components — calm warm palette, generous whitespace, serif display + clean sans body, restrained color, content-first layout. Use when redesigning a screen, creating a new panel, or auditing existing JSX/CSS for visual consistency. Trigger when user mentions "Applied AI design", "Anthropic style", "claude.ai look", or asks to redesign/restyle a BMC component.

Maintains and evolves the BMC team's Universal Knowledge Base (UKB) — a persistent, self-structuring database of all validated team knowledge, pristine agent conclusions, architectural decisions, and conversation history. Every agent reads from and contributes to the UKB as their collective long-term memory. Use when recording validated insights, querying past conclusions, evolving the knowledge schema, or ensuring continuity across runs.

Manages the BMC team's permanent memory: the Conversation Log Database (full history of every agent conversation) and the Universal Knowledge Base (distilled universal truth nourished by agent conclusions). Governs how agents read/write both systems, how the schema auto-evolves, and how knowledge is promoted from raw log entries to confirmed universal truths. Use when any agent needs to record a conversation, extract a conclusion, query team memory, or evolve the knowledge schema.

Enables the BMC agent team to run in parallel windows with live, continuous inter-agent communication. Manages a shared Live Log Center where every agent continuously appends structured log entries; any team member can read, filter, and act on live logs from all other agents without stopping their own work. Use when enabling parallel execution, live agent dialogue, or when a central log hub for the full team is needed.

Specialist for the Panelin Calculadora (port 5173): BOM, pricing, panels, Drive integration, PDF, WhatsApp export. Knows constants, calculations, helpers. Use when working on the quote builder, pricing logic, or Calculadora-Cotizaciones flow.

Evalúa según mejores desempeños en distintas áreas y tareas. Sabe desde cero qué procesos conviene ejecutar en paralelo o en serie. Muy orientado a objetivos; prevé según scores y contexto la mejor combinación de agentes. Use when deciding parallel vs serial execution, agent combination, or orchestration strategy.

Maps both planillas (Google Sheets/templates) and the dashboard interface so the agent knows where each element lives. Produces a plan and proposal before implementation. Use when mapping planillas and dashboard UI together, defining where templates vs interface components live, or generating the mapping plan and proposal for sheet + dashboard.

Syncs project state across all BMC/Panelin areas. Reads PROJECT-STATE.md, runs propagation when changes affect multiple agents, updates state and pendientes. Use when user says sync project state, actualizar equipo, full team run, or keep everyone updated after a change.

Runs a comprehensive BMC Dashboard health audit in fixed order: log anomaly analysis, system state, git history, config, dependencies, API endpoints, security, Google Sheets/scopes, viewer and ngrok. Produces a single "REPORTE SUPER AGENTE BMC — ESTADO COMPLETO". Use when the user asks for super agente bmc, reporte estado completo, analisis anomalias logs bmc, auditoria dashboard bmc, or full system health check.