Brain

livro_comer_sem_culpa nutrition macros — only in Firebase RTDB (rules expired 2021); needs live Firebase or is lost

adrianacarioba Firestore `receitas` — likely overlaps sendo_magro; import only if it adds recipes

Optional: migrate the 158 committed sendo JPEGs to GCS too (shed ~30 MB git bloat) — new uploads already go to GCS as of #670

adriana-recipes: extractors (PDF/gdoc/text → candidates → recipes) (#618)

adriana-recipes: load archived app repos (Firebase/JSON dumps) (#619)

Shipped dashboard one-click merge, not-a-duplicate, and prefer-source actions for recipe pairs (#621).

Applied recipe de-dupe to prod GCS and landed the admin duplicates UX (#763).

Added an in-repo runbook for the receitas subdomain domain mappings and DNS covering both the com and com br domains (#729).

Added an "Área restrita" footer link to the CMS and made the public site work on both adrianacarioba.com and .com.br via an apex redirect stub (#707).

Moved the public recipe site to its own subdomain receitas.adrianacarioba.com and fixed the hero and listing edge-margin bug (#702).

Scoped the public /receitas mount to the adrianacarioba host only, so other hosts fall through to 404 (#687).

Shipped the ungated public recipe site at /receitas with auto-derived tempo and difficulty tags, and dropped the origem source pill and filter from the admin UI (#681).

Dashboard photo uploads now persist to GCS and stream back via the app's credentials, surviving redeploys without a public bucket; a data exposure was flagged (#670)

Production: GCS durable overrides + dashboard live

Spot-check the 32 reconstructed weeks against the original PDFs on the Mac (provenance: data/reconstructed-weeks.json → gmailMessageId per week)

Optional admin gate on /barbacena UI chrome only

Reconstructed the 32 quarantined weeks from the weekly e-mails and re-dated two misfiled PDFs, restoring the out/2023–mai/2024 herd series (#840)

Accuracy audit fixes: 32 corrupt weeks quarantined for manual review, annual Excel sheets excluded from weekly sums, revenue extraction corrected (#840)

Dashboard usability overhaul: interactive charts, annual reports tab with CSV/print, richer week detail, form prefill (#838)

Barbacena Excel history 2015-2022 into dashboard (#836)

Barbacena full history ingest to production (#834)

Durable overrides on Cloud Run via GCS (#830)

GCS durable overrides for prod form saves (#830)

Server API + auth, death/sales charts, estimate empty PDFs (#828)

Sign in + write mode on prod (#816)

Update brain button via xAI (#818)

Make project tags clickable to their repo URL

Consider a `/brain/api/projects/:slug` endpoint returning parsed JSON

Fix scanner nesting depth so subsites/*/BRAIN.md are discovered (needs slug handling for slashed names)

Unsure what the "update brain" prompt is all about - it should be about reading the repo and seeing what has been done and update the main brain

Update brain button xAI (#818)

Brain sign in write mode (#816)

Sign in / read-only write mode for Brain viewer (#816)

Open-issues 403: banner, smoke script, App Issues permission (#817)

Portfolio closeout after ownership hardening (#792)

Path map, checkoff, open-issues error banner (#785)

Union-merge driver for concurrent BRAIN ships (#779)

**Amuseum (Casa Bola visitation)** — Calculadora Interativa upgraded to the full venture planner: corporate structure + tax, seven toggleable revenue streams (ticket tiers, membership tiers), payroll/opex/startup budget, 5-year projection with payback, scen...

**Mídia archive** — paginated feed viewer with load-more and better previews for pulled social posts; Phase 2 press feeds + API keys (#208).

Stand up a dedicated GCP project + OAuth client for eduardolongo.com, then swap the bridge for per-host login

Polish migrated project copy and images where placeholders remain

Doc audit wrote down the editable-overlay content precedence and the live data file hiding in the prototypes folder (#582)

Editable content and media now write through to GCS and restore at boot, so admin edits survive restarts (#562)

Amuseum: simple-editor aplicado a todas as páginas; /new inteiro atrás do login de admin; raiz legada e /amuseum continuam públicos (refs #421).

/new: link Amuseum corrigido no site inteiro; mockup live ganhou fotos reais da categoria Bolas via passthrough sanitizado (refs #419).

Amuseum: mockup navegável do site live criado em /amuseum/live atrás do login — interface apenas, sem backend (refs #417).

Amuseum: site público lançado em /amuseum no lugar do teaser; plano de negócios movido para /amuseum/plano com assets do planner gateados (refs #415).

Amuseum: planner reorganizado e relatório reescrito com os números atuais dos três cenários (refs #413).

Amuseum: widened the page container to 1140px so the planner tables read on one line, with report prose capped at 76ch for readability (refs #411).

**Publish pipeline** — `writeFeed` lands at `new/data/` under the site tree; live feed served at `/new/data/items.json` from the DB. Instagram images are re-hosted to durable GCS at import (scontent URLs expire + login-wall, so that's the only fresh window)...

**Persistence** — DB restores from `gs://lucaslongo-com-media-scraper-db` on cold start and pushes back after every change. The remote is **canonical**: an instance that can't load it freezes uploads instead of clobbering it (#318); bucket versioning is the...

Confirm the queue survives a real Cloud Run recycle (make a change, recycle, check it persists)

Phase 2 continued: curate a real Brazilian-press `config.crawlFeeds` list; set `SERPAPI_API_KEY` / `YOUTUBE_API_KEY` on Cloud Run to light up those rails (#208)

Phase 3: generalize the adriana Instagram ingest into an export adapter

Phase 4: Instagram API adapter (Creator/Business + Meta app review)

Evaluate licensed social-listening tools vs DIY maintenance

Documented a permanent DB-sync operator runbook so any agent can pull or push between local and prod on request, with the usual guardrails (refs #724)

Added a local-to-prod DB sync helper with pull and push npm scripts that back up before copying and force a Cloud Run restart on push (refs #724)

Fixed the duplicate-queue root cause: Google News redirect shells are now re-resolved to real publisher URLs at crawl time and retried with a time-boxed negative cache (refs #709)

Fixed the queue showing the same article twice by adding a title-based sweep that collapses same-article candidates each crawl when redirect shells stay unresolved (refs #704)

Local dev CSS broke because relative asset links only resolve under a trailing-slash mount, so added a guarded trailing-slash redirect to the local dev server (#688)

Fixed the broken dev and test npm scripts: they invoked ts-node, which was never a dependency; tsx runs them now (#643)

Crawl stores publisher URLs for Google News shells #578

Instagram thumbnails are now re-hosted to durable storage at import and expiring CDN links are re-derived or dropped, so the live feed never 403s (refs #363).

#594: point longo.app.br at the hub — Cloudflare zone, Registro.br NS switch, Cloud Run domain mappings (manual)

Fill in the Workspace domain so the mail/calendar quick links deep-link into the JWL account (Documentos already points at the shared drive folder)

Link de Documentos agora abre direto a pasta compartilhada da JWL no Drive (#607)

Hub ganhou tema verde e a marca passou a ser grafada JWL (#605)

Projeto jwl criado com hub de onboarding da família (#593)

#108 — upgrade the blog admin editor when prioritized

Published the interactive programming-languages genealogy chart page with SwiftUI lineage tracing, search and filters (#888)

Published project spotlight blog posts + re-seeded blog.db on persist bucket (#882)

Removed the dead PHP-era editor scripts and the Old site footer links that 404ed in production (#648)

Replaced the goo.gl Sailing Trip photos link with the resolved photos.google.com share URL in the seeds and the prod in-page save, verified live (refs #585)

Documented that content/*.json is a seed only: GCS overwrites it at boot, so prod edits go through the in-page editor (#582)

Defer Turnstile until contact/chat field engage (#557)

Slimmed chat Turnstile so Success chrome no longer crowds the message box, refs #517

Stabilized chat Turnstile managed flow after postMessage origin errors, refs #515

Pin/maquete confirmados por satélite e galeria em grade de categorias (#573)

Trocar o quadrado ilustrativo da maquete pelo perímetro real se a matrícula/levantamento aparecer (pedir ao cartório de Campo Limpo Paulista ou à JL)

Bairro corrigido para Figueira Branca em todos os endereços da página, mapas e memória #599

Pin movido para a casa real via satélite, casa e piscina reorientadas com relevo re-baked, e galeria virou grade de categorias com capas #573

Brain refresh; #573 open for pin/lot/house polish

Maquete inicia em satélite bem zoomada, chalé redesenhado e galeria por categorias #570

Galeria real do anúncio JL e fatos/preço alinhados ao CH00256 #566

Versão em inglês com bandeiras PT/EN e idioma inicial pelo país do visitante (#555)

Página de venda estreia com maquete 3D do terreno, mapa/Earth por endereço e pipeline local de fotos (#547)

Live path-mount at `lucaslongo.com/ocainteriors` on the main Cloud Run image

Confirm `info@ocainteriors.com` Cloudflare Email Routing catch-all is active

Own domain + standalone Cloud Run service when ready

Confirm photo usage rights for the Live Oak Residence photography (© Frank Garnica) before any external/production use

Doc fix: the contact form has had a real Turnstile-guarded backend since #534; removed the stale UI-only Next-up item (#582)

Audit: documented path-mount as live deploy model, refs #497

generalized the editor to support mixed data-edit patterns across sites including amuseum (refs #396 #398).

fixed missing Login/Edit buttons in the footer: editor init could run before React mounted the auth slot; the editor now re-renders it with retry nudges (refs #398).

added Elena's real headshot to the About page, replacing the interior-photo placeholder

fixed two deploy bugs that kept the site 404ing after launch: the Cloud Run path trigger and the root Dockerfile both missed the ocainteriors directory

added a real About/Team page matching the live production content that the Claude Design export didn't include; nav About link now points to it instead of Home

implemented the real site from the Claude Design export with React and no build step: Home, Services, Projects, Gallery, Contact, served as static assets under /ocainteriors

Terrain-glb bake (`terrain/<sha256>/package-v1.glb`, martini TIN + KTX2 drape) → fills `pose.y` + `manifest.terrain`; then load it on the web via GLTFLoader/KTX2 instead of the current client-side `buildTerrain` fallback

Server-side re-bake of retained raw uploads as the fusion engine improves (raw is stored per ride; the recorder path already bakes server-side)

Kill-gate review: does the "my ride became a video game" wow land? (blocks Phases 1–3)

Decide go/no-go on Phase 1 (server fuse+bake pipeline) after the kill-gate

Phase 2: R1 device build + gates on physical iPhone (simulator compile #808 shipped)

Phase 3: auto 9:16 highlight reel (HUD = watermark) + freemium gating

Full-screen dial details with live charts (#880)

START RIDE above Health (#878)

Instrument-cluster home UI and app icon (#876)

Calibration live HUD and failure copy (#874)

Relaxed calibration brake gate (#872)

XcodeGen keeps HealthKit entitlements (#868)

HealthKit entitlements restored (#866)

START crash GNSS clock fix (#864)

Stable on `lucaslongo-site` host routing + Cloudflare Email auto-reply Worker

Audit: dropped stale “move into public/” next-up; site already ships from project root via host static routing, refs #497

tidied the auto-reply Worker setup; repo now ignores local Wrangler output (#61)

email auto-reply Worker live: contact@ replies + forwards to textures.sb@gmail.com (#61)

retired the per-project Dockerfile/workflow; served by lucaslongo-site host routing

live on GCP + Cloudflare catch-all mail; Namecheap hosting retired

Add fixtures when live Gmail emails fail to parse

Validate new matcher on a full production scan pass

AI chain drops defunded gemini/openai; Claude shortcuts resolve to current models instead of retired ones (#640)

Verified GMAIL_REFRESH_TOKEN: OAuth refresh + Gmail list Amazon OK; root .env synced from Secret Manager (#551)

Doc audit: README caught up to the merged Scan Gmail UI and BRAIN.md as canonical memory; sentinel-string and lockfile invariants written down (#582)

Safer global Gmail–Amazon charge matching #578

Re-authorized Gmail OAuth; local .env + Secret Manager + Cloud Run rebound (#551)

Hardened transaction_review_state: zombie cleanup on sync/reset/review load, mutation matrix documented, tests, refs #509

Stopped the activity popover from being wiped after a Gmail scan: the post-scan reload now uses a quiet mode so the summary stays until dismissed (refs #375).

Fixed the Gmail scan silently failing on production: the NDJSON parser swallowed server-sent errors; failures like an expired refresh token now surface in the activity log (refs #375).

Keep git ↔ Hermes sync honest (memories + tracked skills only)

multi-model routing skill versioned and pushable (xAI-only default)

Brain card = status only, not skill bodies

Optional: Brain pane listing tracked skills (later)

Canonical agent worktree folder _worktrees inside main checkout

macos-disk-cleanup skill tracked in agent-profile for Mac disk sweeps

Escalation alert procedure for complex work / Claude / second eye (#849)

xAI-only multi-model routing: Grok plan/review + Composer implement (#847)

Added the receitas-dns Hermes skill so Hermes can finish wiring the recipe subdomain mappings and DNS (#735).

Pulled Hermes memories + tracked skills into git #578

Surfaced agent-profile as a Brain card and scoped skill sync (#519)