Distribution + queues locked: 3-store browser-ext + dual VS Code marketplace + PWA-only + CF Queues + Hookdeck

decision Sat Jun 20 2026 00:00:00 GMT+0000 (Coordinated Universal Time) decisionsarchitecturedistributionextensionsvscodepwaqueuehookdeckbatch-13

Distribution + queues locked

Single consolidated Batch 13 decision covering four parallel distribution / reliability picks. All four locks were originally proposed at wider scope and walked back to the floors below — the family chose simplicity over coverage every time.

Decision

1. Browser extensions: Chrome + Firefox + Edge

Every oriz-*-ext repo publishes to all three browser stores:

Chrome Web Store — $5 one-time dev fee (sunk cost, not a subscription)
Firefox Add-ons (AMO) — free, unlimited
Microsoft Edge Add-ons — free, unlimited

CI flow: build once via web-ext; submit in parallel to all three.

2. VS Code extensions: VS Code Marketplace + Open VSX

Every oriz-*-vsc-ext repo dual-publishes:

VS Code Marketplace — Microsoft's official, free
Open VSX Registry — Eclipse Foundation, required for VSCodium / Cursor / Theia / Gitpod / code-server

CI flow: build once into .vsix; vsce publish + ovsx publish.

3. PWA-only on every site

Every site in the family ships as an installable PWA via @vite-pwa/astro. No native wrapper. Workbox-generated service worker, manifest, install prompt.

4. Webhook reliability: Cloudflare Queues + Hookdeck

The reliability stack is two layers:

Hookdeck (webhook ingress) — 50K events/mo free, exponential-backoff retries, replay UI
Cloudflare Queues (fan-out queue) — 1M ops/mo free, native Worker binding

Producers POST to Hookdeck ? Hookdeck retries to api.oriz.in Worker ? Worker enqueues onto Cloudflare Queues ? consumer Workers drain.

Why

All four picks share the same logic: simplicity over coverage, built on the family's already-locked stack, and all free or one-time-fee.

Three browser stores reach >99% of extension users; adding a fourth (Safari, Opera-native) would gate on Apple Developer Program ($99/yr — fights no-subscriptions-anywhere) or duplicate Chromium reach.
Two VS Code marketplaces reach every VS Code-compatible editor (VSCodium, Cursor, Theia, Gitpod, code-server). JetBrains is a different IDE family — different artifact format, different build, different audience. The family has no JetBrains plugin in plan; pre-emptively wiring the publish step would be over-coverage.
PWA-only means every site is installable today with no Apple Developer / Google Play tax, no signing complexity, no per-platform binary build. Capacitor and Tauri stay walked back unless a hardware / native-only feature need lands.
CF Queues + Hookdeck stacks two reliability layers from services the family already pays $0 for. Trigger.dev's durable-workflow model is more powerful but redundant for current webhook volume, and adds a new account / credentials surface.

Implications

Every extension repo (-ext and -vsc-ext per branding/repo-naming-suffixes.md) carries one CI workflow that publishes to all the stores in its flavour's row of the matrix below. Per decisions/process/per-repo-ci-workflows.md.
Every site repo enables @vite-pwa/astro via the preset; the only per-site override is name / icons.
Every external-webhook producer (Razorpay first; Lemon Squeezy, GitHub, others as added) targets a Hookdeck connection URL, not api.oriz.in directly.
All credentials originate at Doppler and mirror to GitHub Secrets per security/secrets-management-doppler.md.
The earlier infrastructure/extensions-cross-store-publish.md (Batch 1) and infrastructure/hookdeck-for-webhook-reliability.md (Batch 4) and decisions/architecture/queue-cloudflare-native.md (Batch 8) are NOT superseded — they remain in force; this Batch 13 lock consolidates them with the new VS Code + PWA + Hookdeck-as-ingress facets and records the walked-back alternatives.

Publish + reliability matrix

Surface	Targets (active)	Walked back
Browser extension (`oriz-*-ext`)	Chrome Web Store + Firefox AMO + Edge Add-ons	Safari (Apple Developer fee), standalone Opera
VS Code extension (`oriz-*-vsc-ext`)	VS Code Marketplace + Open VSX	JetBrains Marketplace
Site (`oriz-*-site`)	PWA via `@vite-pwa/astro`	Capacitor, Tauri, TWA
Webhook reliability	Hookdeck (ingress) ? Cloudflare Queues (fan-out)	Trigger.dev (durable workflows)

Walked back — why simplicity over coverage

Walked back	Originally proposed for	Why walked back
JetBrains Marketplace	VS Code distribution trio	Different IDE family entirely (Gradle build, JetBrains Platform SDK, separate audience). Family has no JetBrains plugin in plan; adding the publish target pre-emptively is dead code. Re-open if a JetBrains plugin is ever greenlit — likely under a `-jb-ext` suffix added to `repo-naming-suffixes.md`.
Capacitor	Native PWA wrapper, parallel to vite-pwa	iOS publish requires Apple Developer Program at $99/yr, conflicts with `no-subscriptions-anywhere`. Android publish doable free but adds a per-app binary, signing, Play Store review. PWA install on Android already covers the install-icon use case.
Tauri	Native PWA wrapper, parallel to vite-pwa	Adds a Rust toolchain + WebView dependency for zero gain over PWA on the surface the family targets (web). Re-open only if a desktop-native feature lands (system tray, native-menubar, native-FS) that PWA APIs can't express.
Trigger.dev	Durable-workflow queue alternative to CF Queues	Powerful programming model (code-defined multi-step workflows with checkpointed state) but overkill for current webhook volume. Adds a separate account, credentials surface, and hosting dependency. CF Queues + Hookdeck covers the same reliability surface inside the family's existing stack. Re-open if multi-step durable workflows become a real need (the `queue/inngest.md` entry covers the same swap-target shape).

The shape of every walked-back item is the same: marginal coverage gain at the cost of a new credentials surface, a paid tier, or a build-toolchain dependency. The family stays small, the stack stays on Cloudflare, and re-opening any of these is a one-decision flip when a real need surfaces.