@pylonsync/create-pylon 0.3.332 → 0.3.334

package/templates/shop/AGENTS.md

@@ -1,47 +1,47 @@
-# AGENTS.md — working in a Pylon project
+# AGENTS.md: working in a Pylon project
 
-Operating rules for a coding agent in this Pylon app. You — the agent — are a first-class user of Pylon: one Rust binary (`pylon`) serves the API, auth, sync, WebSocket, SSE, and native React 19 SSR from one process on one port, so you build, run, and ship a whole app without wiring services together or leaving the codebase. This is production infrastructure, not a sandbox — real auth, SQLite or Postgres, row-level policies, jobs, search, and one-command deploy — so build like it ships. You declare entities, policies, and server functions in TypeScript; the binary does the rest. The full API reference is the **llms-full.txt** at https://docs.pylonsync.com/llms-full.txt — read it before guessing an API name.
+Pylon serves the API, auth, sync, WebSocket, SSE, and native React 19 SSR from one Rust process on one port. Treat this app as production infrastructure: it supports real auth, SQLite or Postgres, row-level policies, jobs, search, and one-command deploys. Declare entities, policies, and server functions in TypeScript; the binary handles the runtime. Read the full API reference at https://docs.pylonsync.com/llms-full.txt before guessing an API name.
 
 ## Directory conventions
 
 **Unified SSR app:**
-- `app.ts` — data model + manifest (`entity()` + `field.*`, queries/actions/policies, `routes: await discoverAppRoutes()`). Ends with `console.log(JSON.stringify(manifest))`.
-- `app/` — file-based SSR routes. `app/page.tsx` → `/`, `app/about/page.tsx` → `/about`, `app/blog/[slug]/page.tsx` → `/blog/:slug`. `app/layout.tsx` is the shell; `app/error.tsx` / `app/not-found.tsx` are boundaries.
-- `app/globals.css` — Tailwind v4 entrypoint (auto-compiled and injected).
-- `functions/` — server functions, one per file, `default`-exported.
-- `.pylon/` — local dev state (sqlite, jobs, sessions, uploads). Created by `pylon dev`. Do not commit.
+- `app.ts`: data model + manifest (`entity()` + `field.*`, queries/actions/policies, `routes: await discoverAppRoutes()`). Ends with `console.log(JSON.stringify(manifest))`.
+- `app/`: file-based SSR routes. `app/page.tsx` → `/`, `app/about/page.tsx` → `/about`, `app/blog/[slug]/page.tsx` → `/blog/:slug`. `app/layout.tsx` is the shell; `app/error.tsx` / `app/not-found.tsx` are boundaries.
+- `app/globals.css`: Tailwind v4 entrypoint (auto-compiled and injected).
+- `functions/`: server functions, one per file, `default`-exported.
+- `.pylon/`: local dev state (SQLite, jobs, sessions, uploads). Created by `pylon dev`. Do not commit.
 
-**Monorepo app:** backend is `apps/api/` (entry `apps/api/schema.ts`, handlers in `apps/api/functions/`); frontend in `apps/web/`. `pylon.manifest.json` / `pylon.client.ts` are generated — do not hand-edit.
+**Monorepo app:** backend is `apps/api/` (entry `apps/api/schema.ts`, handlers in `apps/api/functions/`); frontend in `apps/web/`. `pylon.manifest.json` / `pylon.client.ts` are generated; do not hand-edit.
 
 ## The core authoring loop
 
-1. **Define an entity** — `entity("Thing", { name: field.string(), done: field.boolean().default(false) })`. Modifiers: `.optional()`, `.unique()`, `.readonly()` (settable on insert, rejected on client update — use for `authorId`/`orgId`), `.serverOnly()` (never in HTTP responses), `.encrypted()` (AEAD at rest, needs `PYLON_ENCRYPTION_KEY`), `.crdt("text")` (collaborative).
-2. **Write a policy** — `policy({ entity: "Thing", allowRead, allowInsert, allowUpdate, allowDelete })` with CEL-like expressions over `auth.*` / `data.*` (e.g. `"auth.userId == data.authorId"`). **Omitted actions DENY by default.** Wide-open dev policies (`allow*: "true"`) are flagged by `pylon lint` — tighten before shipping.
-3. **Author a function** in `functions/<name>.ts` — `query` (read-only), `mutation` (transactional read+write), or `action` (external I/O, no direct `ctx.db`). Import `{ query, mutation, action, v }` from `@pylonsync/functions`. `auth` defaults to `"user"` (secure-by-default); set `"public"` explicitly for unauthenticated access. Use `ctx.db.*`, `ctx.auth.userId`, `ctx.error(code, msg)`.
-4. **Read it on the client** — `db.useQuery("Thing")` (live, re-renders on any write) or `db.useQueryOne("Thing", id)`. Call functions with `db.fn(name, args)` / `callFn`. On SSR pages, read via `use(serverData.list("Thing"))` inside `<Suspense>`.
+1. **Define an entity:** `entity("Thing", { name: field.string(), done: field.boolean().default(false) })`. Modifiers: `.optional()`, `.unique()`, `.readonly()` (settable on insert, rejected on client update; use for `authorId`/`orgId`), `.serverOnly()` (never in HTTP responses), `.encrypted()` (AEAD at rest, needs `PYLON_ENCRYPTION_KEY`), `.crdt("text")` (collaborative).
+2. **Write a policy:** `policy({ entity: "Thing", allowRead, allowInsert, allowUpdate, allowDelete })` with CEL-like expressions over `auth.*` / `data.*` (e.g. `"auth.userId == data.authorId"`). Omitted actions deny by default. `pylon lint` flags wide-open development policies such as `allow*: "true"`; tighten them before shipping.
+3. **Author a function** in `functions/<name>.ts`: `query` (read-only), `mutation` (transactional read+write), or `action` (external I/O, no direct `ctx.db`). Import `{ query, mutation, action, v }` from `@pylonsync/functions`. `auth` defaults to `"user"` (secure-by-default); set `"public"` explicitly for unauthenticated access. Use `ctx.db.*`, `ctx.auth.userId`, `ctx.error(code, msg)`.
+4. **Read it on the client:** `db.useQuery("Thing")` (live, re-renders on any write) or `db.useQueryOne("Thing", id)`. Call functions with `db.fn(name, args)` / `callFn`. On SSR pages, read via `use(serverData.list("Thing"))` inside `<Suspense>`.
 
 ## Key gotchas
 
-- **Policies deny by default; server functions BYPASS them.** Direct client CRUD (`/api/entities/*`) and sync are policy-checked. Functions run with full DB access — enforce trust with `ctx.auth` checks inside the handler, not policies.
-- **Type page props from the SDK, don't hand-roll them.** `import type { PageProps, Metadata } from "@pylonsync/react"`. Every page/layout gets `{ url, params, searchParams, auth, response, serverData }`; `PageProps<{ slug: string }>` types a `[slug]` route's params. Request headers/cookies are intentionally NOT on `PageProps` — they're server-only and stripped from hydration, so reading them in the render would mismatch.
-- **Anonymous output caching is opt-in + earned.** `export const revalidate = 60` (seconds) on a page makes it CDN-cacheable (`public, s-maxage=60`) — but ONLY if the render is auth-INDEPENDENT: it must NOT read `props.auth` (reading it at all opts out, even for anonymous), set no cookie, and the app must not run strict per-caller policies (`PYLON_STRICT_FN_POLICIES`). `export const dynamic = "force-static"` caches until the next deploy; `"force-dynamic"` never caches. Fail-closed: without the opt-in (or if any condition fails) the page is `no-cache`. A page that reads `auth` or sets a cookie is never shared. The SAME earned render is also kept in an **origin disk cache** (`.pylon/.cache/ssr`): a cookie-less GET with no query string is served straight off disk for the TTL — skipping the render entirely — then re-rendered live when stale. The disk cache is namespaced per deploy (wiped on each new build) and OFF in `pylon dev` (so an edit is never masked by a stale entry); invalidation is by the `revalidate` TTL or the next deploy.
-- **No-JS forms use `route.ts` + `<Form>`.** Drop `app/.../route.ts` exporting `export const POST: RouteHandler = async ({ form, db, response, auth }) => { await db.insert("X", {...}); response.redirect("/x?ok=1"); }` (303 POST-redirect-GET by default). Render `<Form action="/x">` (from @pylonsync/react) with plain `<input name=...>` — works with JS off (native POST→handler→redirect) and is enhanced to no-reload when JS is on. The handler's `db` is read+write (mutation trust model — gate on `auth`); CSRF is automatic (Origin gate + SameSite=Lax). Multipart/file uploads aren't supported yet — use urlencoded forms + `/api/files`.
-- **`loading.tsx` streams a skeleton while the page's data resolves.** Drop `app/.../loading.tsx` (default export, page props) and the nearest one becomes a route-level Suspense fallback: Pylon flushes the shell + skeleton immediately, then reveals the real page when its top-level `use(serverData…)` resolves (no blank page). It only shows when the PAGE suspends — a page that wraps its own `<Suspense>` around a child (like `/dashboard` in this template) handles that itself. The skeleton is SERVER-ONLY: don't read `serverData` in it. A page with no `loading.tsx` is buffered (unchanged).
-- **`export const streaming = true` streams a page's OWN inner `<Suspense>` boundaries.** Without it (and without a `loading.tsx`), a page is BUFFERED — the whole document, including suspended children, resolves before the first byte. Opt in and the shell + each inner `<Suspense>` fallback flush immediately, then each boundary's real content streams in as its data resolves (multi-boundary progressive streaming). It's opt-in because it changes the response timing contract: a streaming render commits its HTTP head BEFORE suspended subtrees finish, so (a) it's never CDN/disk cacheable — don't combine with `export const revalidate`; (b) `response.setStatus/setCookie/redirect/notFound` only take effect from the SYNCHRONOUS shell render — a call from inside a suspended subtree is dropped (the runtime logs a loud warning naming what was lost); (c) a `throw` from a deep `<Suspense>` child resolves via its nearest `error.tsx` at HTTP 200, not a 5xx. Hydration is clean for any number of boundaries (the data blob ships before hydration runs). Type the config with `import type { RouteSegmentConfig } from "@pylonsync/react"`.
-- **`error.tsx` / `not-found.tsx` boundaries are HYDRATED (interactive).** `app/.../error.tsx` catches a throw below it (HTTP 500) and receives `{ error: { message, digest }, reset }` (`import type { ErrorBoundaryProps }`) — `reset()` re-attempts the route; the stack NEVER reaches the client (dev overlay + logs only). `app/.../not-found.tsx` renders at 404 (also for `response.notFound()`) and gets the page props (`NotFoundProps`), no `reset`. Both run useState/onClick/hooks.
-- **Client navigation hooks live in @pylonsync/react.** `useRouter()` → `{ push, replace, back, forward, refresh, prefetch }`; `useSearchParams()` → reactive `URLSearchParams`; `usePathname()` → reactive pathname. The hooks are CLIENT-reactive — during SSR they return defaults (empty params / "/"); for server-side URL values read the `url` / `searchParams` page props.
-- **Dynamic + catch-all routes follow Next conventions.** `app/blog/[slug]/page.tsx` → `params.slug`. `app/docs/[...path]/page.tsx` is a catch-all (matches `/docs/a/b/c`; `params.path === "a/b/c"` — `.split("/")` for segments). `app/shop/[[...filters]]/page.tsx` is an optional catch-all (also matches the bare `/shop`, with `params.filters === ""`). A catch-all must be the last segment; static beats dynamic beats catch-all on overlap.
+- **Policies deny by default; server functions BYPASS them.** Direct client CRUD (`/api/entities/*`) and sync are policy-checked. Functions run with full DB access; enforce trust with `ctx.auth` checks inside the handler, not policies.
+- **Type page props from the SDK, don't hand-roll them.** `import type { PageProps, Metadata } from "@pylonsync/react"`. Every page/layout gets `{ url, params, searchParams, auth, response, serverData }`; `PageProps<{ slug: string }>` types a `[slug]` route's params. Request headers/cookies are intentionally NOT on `PageProps`; they're server-only and stripped from hydration, so reading them in the render would mismatch.
+- **Anonymous output caching is opt-in and conditional.** `export const revalidate = 60` makes a page CDN-cacheable (`public, s-maxage=60`) only when the render is auth-independent: it must not read `props.auth`, set a cookie, or run with strict per-caller policies (`PYLON_STRICT_FN_POLICIES`). `export const dynamic = "force-static"` caches until the next deploy; `"force-dynamic"` never caches. When any condition fails, the page is `no-cache`. Eligible renders also use the origin disk cache at `.pylon/.cache/ssr`: a cookie-less GET without a query string is served from disk for the TTL and rerendered when stale. The cache is namespaced per deploy, cleared by each build, disabled in `pylon dev`, and invalidated by the `revalidate` TTL or the next deploy.
+- **No-JS forms use `route.ts` and `<Form>`.** Add `app/.../route.ts` exporting `export const POST: RouteHandler = async ({ form, db, response, auth }) => { await db.insert("X", {...}); response.redirect("/x?ok=1"); }` (303 POST-redirect-GET by default). Render `<Form action="/x">` from `@pylonsync/react` with plain `<input name=...>`. It uses native POST → handler → redirect without JavaScript and no-reload enhancement with JavaScript. The handler's `db` is read+write under the mutation trust model, so gate it on `auth`. CSRF protection is automatic through the Origin gate and SameSite=Lax. Multipart uploads are not supported yet; use URL-encoded forms and `/api/files`.
+- **`loading.tsx` streams a skeleton while the page's data resolves.** Drop `app/.../loading.tsx` (default export, page props) and the nearest one becomes a route-level Suspense fallback: Pylon flushes the shell + skeleton immediately, then reveals the real page when its top-level `use(serverData…)` resolves (no blank page). It only shows when the PAGE suspends; a page that wraps its own `<Suspense>` around a child (like `/dashboard` in this template) handles that itself. The skeleton is SERVER-ONLY: don't read `serverData` in it. A page with no `loading.tsx` is buffered (unchanged).
+- **`export const streaming = true` streams a page's inner `<Suspense>` boundaries.** Without it or a `loading.tsx`, the page is buffered until all suspended children resolve. With it, the shell and fallbacks flush immediately, then each boundary streams its content. Streaming commits the HTTP head before suspended subtrees finish, so the page is never CDN- or disk-cacheable; do not combine it with `export const revalidate`. Calls to `response.setStatus`, `setCookie`, `redirect`, or `notFound` only take effect during the synchronous shell render. A call from a suspended subtree is dropped and logged. An error from a deep `<Suspense>` child resolves through the nearest `error.tsx` at HTTP 200 rather than 5xx. Type the config with `import type { RouteSegmentConfig } from "@pylonsync/react"`.
+- **`error.tsx` / `not-found.tsx` boundaries are HYDRATED (interactive).** `app/.../error.tsx` catches a throw below it (HTTP 500) and receives `{ error: { message, digest }, reset }` (`import type { ErrorBoundaryProps }`); `reset()` re-attempts the route; the stack NEVER reaches the client (dev overlay + logs only). `app/.../not-found.tsx` renders at 404 (also for `response.notFound()`) and gets the page props (`NotFoundProps`), no `reset`. Both run useState/onClick/hooks.
+- **Client navigation hooks live in @pylonsync/react.** `useRouter()` → `{ push, replace, back, forward, refresh, prefetch }`; `useSearchParams()` → reactive `URLSearchParams`; `usePathname()` → reactive pathname. The hooks are CLIENT-reactive; during SSR they return defaults (empty params / "/"); for server-side URL values read the `url` / `searchParams` page props.
+- **Dynamic + catch-all routes follow Next conventions.** `app/blog/[slug]/page.tsx` → `params.slug`. `app/docs/[...path]/page.tsx` is a catch-all (matches `/docs/a/b/c`; `params.path === "a/b/c"`; `.split("/")` for segments). `app/shop/[[...filters]]/page.tsx` is an optional catch-all (also matches the bare `/shop`, with `params.filters === ""`). A catch-all must be the last segment; static beats dynamic beats catch-all on overlap.
 - **`serverData` (SSR) is READ-ONLY.** No write methods; the runtime rejects write frames (`SSR_WRITE_FORBIDDEN`). Mutations belong in actions/functions, never in a page render.
-- **`response.*` / `response.redirect()` / `response.notFound()` must fire in the synchronous shell render**, before any `await` / `<Suspense>`. The HTTP head commits when the shell is ready — status/headers/cookies set from a suspended subtree are lost, and `redirect`/`notFound` thrown below a Suspense boundary are swallowed.
-- **`ctx.llm` and `ctx.connections` are on mutation + action only, NOT query** (reactive purity). `action` has no direct `ctx.db` — use `ctx.runQuery` / `ctx.runMutation`.
+- **`response.*` / `response.redirect()` / `response.notFound()` must fire in the synchronous shell render**, before any `await` / `<Suspense>`. The HTTP head commits when the shell is ready; status/headers/cookies set from a suspended subtree are lost, and `redirect`/`notFound` thrown below a Suspense boundary are swallowed.
+- **`ctx.llm` and `ctx.connections` are on mutation + action only, NOT query** (reactive purity). `action` has no direct `ctx.db`; use `ctx.runQuery` / `ctx.runMutation`.
 - **It's `db.useQueryOne`, not `useOne`.** Validators and field types have aliases: `v.bool`/`v.boolean`, `v.float`/`v.number`.
-- **There is no `ctx.files` or `defineWorkflow`/`defineJob`.** Files go through `<FileUpload>` + `/api/files/*`. Deferred (one-shot) execution is `ctx.scheduler.runAfter/runAt/cancel`. Recurring work is a **cron**: `cron("0 * * * *", "fnName")` in `buildManifest({ crons: [...] })` (import `cron` from `@pylonsync/sdk`) — it fires the named function (make it `internal: true`) on the schedule; the function runs with anonymous auth — its own `ctx.db.*` is server-side (not policy-gated), so write directly; only `ctx.auth.elevate({ admin: true, reason: "..." })` (reason mandatory) to chain an `internal: true` function via `ctx.scheduler`.
+- **Use the supported file and scheduling APIs.** Files go through `<FileUpload>` and `/api/files/*`; there is no `ctx.files`. One-shot work uses `ctx.scheduler.runAfter`, `runAt`, or `cancel`; there is no `defineWorkflow` or `defineJob`. Recurring work uses `cron("0 * * * *", "fnName")` in `buildManifest({ crons: [...] })`, imported from `@pylonsync/sdk`. Make the target function `internal: true`. It runs with anonymous auth, but its own `ctx.db.*` calls are server-side and bypass policies. Use `ctx.auth.elevate({ admin: true, reason: "..." })`, with a mandatory reason, only when chaining another internal function through `ctx.scheduler`.
 
 ## Testing
 
-`pylon test` discovers every `*.test.ts` / `*.test.tsx` file under `tests/` (or `functions/`) and runs it with **Bun's test runner** (`import { test, expect } from "bun:test"`). Run the suite with `pylon test` (or `npm test`); filter with `pylon test <substring>`. This template ships `bunfig.toml` + `tests/setup.ts` (registers happy-dom) so component tests render out of the box, plus starter tests under `tests/` — replace them with your own.
+`pylon test` discovers every `*.test.ts` / `*.test.tsx` file under `tests/` (or `functions/`) and runs it with **Bun's test runner** (`import { test, expect } from "bun:test"`). Run the suite with `pylon test` (or `npm test`); filter with `pylon test <substring>`. This template ships `bunfig.toml` + `tests/setup.ts` (registers happy-dom) so component tests render out of the box, plus starter tests under `tests/`; replace them with your own.
 
-**Tier 1 — pure logic (reach for this first).** Keep the decisions that matter — access/plan gating, pricing, credit math, validation, formatting — in pure functions in `lib/`, and test those exhaustively. No server, instant, and it's where the real bugs live. Keep your `query`/`mutation`/`action` handlers as thin wrappers around them, so the logic is testable without a running app.
+**Tier 1: pure logic (start here).** Keep access and plan gating, pricing, credit math, validation, and formatting in pure functions under `lib/`, and test them exhaustively. These tests need no server and run instantly. Keep `query`, `mutation`, and `action` handlers as thin wrappers so their decision logic remains testable without a running app.
 
 ```ts
 import { expect, test } from "bun:test";
@@ -52,7 +52,7 @@ test("unknown slug → undefined", () => {
 });
 ```
 
-**Tier 2 — React components.** `@testing-library/react` + happy-dom are already wired (`tests/setup.ts`). Render and assert. The template uses the classic JSX transform, so add `import React from "react"` in `.tsx` tests. For a component that reads Pylon data hooks, **mock the boundary**, then dynamic-`import` the component so the mock is in place first:
+**Tier 2: React components.** `@testing-library/react` and happy-dom are already wired through `tests/setup.ts`. Render and assert. The template uses the classic JSX transform, so add `import React from "react"` in `.tsx` tests. For a component that reads Pylon data hooks, mock the boundary, then dynamically import the component so the mock is in place first:
 
 ```tsx
 import { test, expect, mock } from "bun:test";
@@ -70,7 +70,7 @@ test("renders orgs from the query", () => {
 });
 ```
 
-**Tier 3 — functions over HTTP (only when Tier 1 can't cover it).** A handler's full behavior (policies, `ctx.db`, auth) lives in the running app. Start `pylon dev` in another terminal and call the API; `resetDb()` from `@pylonsync/functions` clears the in-memory DB between cases (no-ops if the server's down, refuses production).
+**Tier 3: functions over HTTP.** Use this tier when pure logic tests cannot cover the behavior. A handler's policies, `ctx.db` calls, and auth run in the app. Start `pylon dev` in another terminal and call the API. `resetDb()` from `@pylonsync/functions` clears the in-memory database between cases; it does nothing when the server is down and refuses to run in production.
 
 ```ts
 import { afterEach, expect, test } from "bun:test";
@@ -90,9 +90,9 @@ test("createThing then read it back", async () => {
 });
 ```
 
-`pylon test:security` is a separate adversarial probe — it hits a running app and reports auth/policy holes (run `pylon dev`, then `pylon test:security`).
+`pylon test:security` is a separate adversarial probe; it hits a running app and reports auth/policy holes (run `pylon dev`, then `pylon test:security`).
 
-## Use the CLI — don't guess
+## Use the CLI
 
 | Need | Command |
 |---|---|

package/templates/shop/CLAUDE.md

@@ -1,7 +1,7 @@
 # CLAUDE.md
 
-Agent guidance for this Pylon project lives in **AGENTS.md** — the cross-editor
-standard that Claude Code, Cursor, Codex, and others all read. It's imported
-below so Claude Code picks it up automatically:
+Agent guidance for this Pylon project lives in **AGENTS.md**, the cross-editor
+standard read by Claude Code, Cursor, Codex, and other agents. The import below
+makes it available to Claude Code:
 
 @AGENTS.md

package/templates/shop/README.md

@@ -1,12 +1,11 @@
 # __APP_NAME__
 
-A small DTC store built with [Pylon](https://pylonsync.com) — a server-rendered
-storefront with a **cart, real Stripe checkout, and live inventory**, plus a
-private owner dashboard, all from one binary on one port. No Next.js, no
-separate API server.
+A small DTC store built with [Pylon](https://pylonsync.com), with a
+server-rendered storefront, cart, Stripe Checkout, live inventory, and a
+private owner dashboard.
 
-The realtime point: each product shows its stock, and the moment someone buys
-the last unit it flips to "Sold out" for everyone with the page open.
+Each product shows current stock. When someone buys the last unit, it changes
+to "Sold out" on every open page.
 
 ## Develop
 
@@ -14,9 +13,8 @@ the last unit it flips to "Sold out" for everyone with the page open.
 __RUN_DEV__
 ```
 
-Open http://localhost:4321. Then **open a second tab**, add the last of an item
-to your cart and check out in one, and watch it flip to "Sold out" in the
-other — with no refresh.
+Open http://localhost:4321. Add the last unit of an item to your cart and check
+out in one tab, then watch it change to "Sold out" in another.
 
 > Checkout works with **zero config**: with no Stripe keys set, it holds the
 > stock and records a "reserved" order for you to follow up on. Add
@@ -36,7 +34,7 @@ other — with no refresh.
 - `functions/seedProducts.ts` loads the catalog from config into the DB on first
   visit (idempotent).
 
-## Privacy — read this
+## Privacy
 
 The `Order` entity holds the customer's name + email (PII), so its policy in
 `app.ts` **denies every client read and write**. The public page only reads
@@ -71,9 +69,9 @@ listen` setup.
 
 ## Rebrand it
 
-Everything lives in **`lib/site.config.ts`** — brand, colors, the product list
-(with starting stock), value props, reviews, policies. Edit that one file and
-the whole store re-themes; the products re-seed on a fresh database.
+Brand, colors, products and starting stock, value props, reviews, and policies
+live in **`lib/site.config.ts`**. Editing that file updates the store, and a
+fresh database seeds products from it.
 
 ## Layout
 

package/templates/shop/lib/site.config.ts

@@ -1,10 +1,9 @@
-// THE single source of truth for everything business-specific. Rebrand the
-// whole store by editing this ONE file — the landing page and layout read from
-// here. The product list (incl. starting stock) seeds the Product table on
+// Business-specific copy and settings live here. The landing page and layout
+// read this file. The product list, including starting stock, seeds the Product table on
 // first visit; after that, stock lives in the database and updates live.
 //
 // Colors live here (applied as CSS variables on <html> in app/layout.tsx).
-// Fictional demo copy — replace the values, keep the shape.
+// Fictional demo copy. Replace the values and keep the shape.
 
 /* ----------------------------- types ----------------------------- */
 
@@ -71,7 +70,7 @@ export const siteConfig: ShopConfig = {
   seo: {
     title: "Ember Goods — hand-poured candles, made in Dallas.",
     description:
-      "Small-batch candles and home goods, hand-poured in Dallas. Live stock — what you see is what's in the studio. Free shipping over $50.",
+      "Small-batch candles and home goods, hand-poured in Dallas. Stock counts match the studio shelf. Free shipping over $50.",
   },
 
   hero: {

package/templates/todo/AGENTS.md

@@ -1,47 +1,47 @@
-# AGENTS.md — working in a Pylon project
+# AGENTS.md: working in a Pylon project
 
-Operating rules for a coding agent in this Pylon app. You — the agent — are a first-class user of Pylon: one Rust binary (`pylon`) serves the API, auth, sync, WebSocket, SSE, and native React 19 SSR from one process on one port, so you build, run, and ship a whole app without wiring services together or leaving the codebase. This is production infrastructure, not a sandbox — real auth, SQLite or Postgres, row-level policies, jobs, search, and one-command deploy — so build like it ships. You declare entities, policies, and server functions in TypeScript; the binary does the rest. The full API reference is the **llms-full.txt** at https://docs.pylonsync.com/llms-full.txt — read it before guessing an API name.
+Pylon serves the API, auth, sync, WebSocket, SSE, and native React 19 SSR from one Rust process on one port. Treat this app as production infrastructure: it supports real auth, SQLite or Postgres, row-level policies, jobs, search, and one-command deploys. Declare entities, policies, and server functions in TypeScript; the binary handles the runtime. Read the full API reference at https://docs.pylonsync.com/llms-full.txt before guessing an API name.
 
 ## Directory conventions
 
 **Unified SSR app:**
-- `app.ts` — data model + manifest (`entity()` + `field.*`, queries/actions/policies, `routes: await discoverAppRoutes()`). Ends with `console.log(JSON.stringify(manifest))`.
-- `app/` — file-based SSR routes. `app/page.tsx` → `/`, `app/about/page.tsx` → `/about`, `app/blog/[slug]/page.tsx` → `/blog/:slug`. `app/layout.tsx` is the shell; `app/error.tsx` / `app/not-found.tsx` are boundaries.
-- `app/globals.css` — Tailwind v4 entrypoint (auto-compiled and injected).
-- `functions/` — server functions, one per file, `default`-exported.
-- `.pylon/` — local dev state (sqlite, jobs, sessions, uploads). Created by `pylon dev`. Do not commit.
+- `app.ts`: data model + manifest (`entity()` + `field.*`, queries/actions/policies, `routes: await discoverAppRoutes()`). Ends with `console.log(JSON.stringify(manifest))`.
+- `app/`: file-based SSR routes. `app/page.tsx` → `/`, `app/about/page.tsx` → `/about`, `app/blog/[slug]/page.tsx` → `/blog/:slug`. `app/layout.tsx` is the shell; `app/error.tsx` / `app/not-found.tsx` are boundaries.
+- `app/globals.css`: Tailwind v4 entrypoint (auto-compiled and injected).
+- `functions/`: server functions, one per file, `default`-exported.
+- `.pylon/`: local dev state (SQLite, jobs, sessions, uploads). Created by `pylon dev`. Do not commit.
 
-**Monorepo app:** backend is `apps/api/` (entry `apps/api/schema.ts`, handlers in `apps/api/functions/`); frontend in `apps/web/`. `pylon.manifest.json` / `pylon.client.ts` are generated — do not hand-edit.
+**Monorepo app:** backend is `apps/api/` (entry `apps/api/schema.ts`, handlers in `apps/api/functions/`); frontend in `apps/web/`. `pylon.manifest.json` / `pylon.client.ts` are generated; do not hand-edit.
 
 ## The core authoring loop
 
-1. **Define an entity** — `entity("Thing", { name: field.string(), done: field.boolean().default(false) })`. Modifiers: `.optional()`, `.unique()`, `.readonly()` (settable on insert, rejected on client update — use for `authorId`/`orgId`), `.serverOnly()` (never in HTTP responses), `.encrypted()` (AEAD at rest, needs `PYLON_ENCRYPTION_KEY`), `.crdt("text")` (collaborative).
-2. **Write a policy** — `policy({ entity: "Thing", allowRead, allowInsert, allowUpdate, allowDelete })` with CEL-like expressions over `auth.*` / `data.*` (e.g. `"auth.userId == data.authorId"`). **Omitted actions DENY by default.** Wide-open dev policies (`allow*: "true"`) are flagged by `pylon lint` — tighten before shipping.
-3. **Author a function** in `functions/<name>.ts` — `query` (read-only), `mutation` (transactional read+write), or `action` (external I/O, no direct `ctx.db`). Import `{ query, mutation, action, v }` from `@pylonsync/functions`. `auth` defaults to `"user"` (secure-by-default); set `"public"` explicitly for unauthenticated access. Use `ctx.db.*`, `ctx.auth.userId`, `ctx.error(code, msg)`.
-4. **Read it on the client** — `db.useQuery("Thing")` (live, re-renders on any write) or `db.useQueryOne("Thing", id)`. Call functions with `db.fn(name, args)` / `callFn`. On SSR pages, read via `use(serverData.list("Thing"))` inside `<Suspense>`.
+1. **Define an entity:** `entity("Thing", { name: field.string(), done: field.boolean().default(false) })`. Modifiers: `.optional()`, `.unique()`, `.readonly()` (settable on insert, rejected on client update; use for `authorId`/`orgId`), `.serverOnly()` (never in HTTP responses), `.encrypted()` (AEAD at rest, needs `PYLON_ENCRYPTION_KEY`), `.crdt("text")` (collaborative).
+2. **Write a policy:** `policy({ entity: "Thing", allowRead, allowInsert, allowUpdate, allowDelete })` with CEL-like expressions over `auth.*` / `data.*` (e.g. `"auth.userId == data.authorId"`). Omitted actions deny by default. `pylon lint` flags wide-open development policies such as `allow*: "true"`; tighten them before shipping.
+3. **Author a function** in `functions/<name>.ts`: `query` (read-only), `mutation` (transactional read+write), or `action` (external I/O, no direct `ctx.db`). Import `{ query, mutation, action, v }` from `@pylonsync/functions`. `auth` defaults to `"user"` (secure-by-default); set `"public"` explicitly for unauthenticated access. Use `ctx.db.*`, `ctx.auth.userId`, `ctx.error(code, msg)`.
+4. **Read it on the client:** `db.useQuery("Thing")` (live, re-renders on any write) or `db.useQueryOne("Thing", id)`. Call functions with `db.fn(name, args)` / `callFn`. On SSR pages, read via `use(serverData.list("Thing"))` inside `<Suspense>`.
 
 ## Key gotchas
 
-- **Policies deny by default; server functions BYPASS them.** Direct client CRUD (`/api/entities/*`) and sync are policy-checked. Functions run with full DB access — enforce trust with `ctx.auth` checks inside the handler, not policies.
-- **Type page props from the SDK, don't hand-roll them.** `import type { PageProps, Metadata } from "@pylonsync/react"`. Every page/layout gets `{ url, params, searchParams, auth, response, serverData }`; `PageProps<{ slug: string }>` types a `[slug]` route's params. Request headers/cookies are intentionally NOT on `PageProps` — they're server-only and stripped from hydration, so reading them in the render would mismatch.
-- **Anonymous output caching is opt-in + earned.** `export const revalidate = 60` (seconds) on a page makes it CDN-cacheable (`public, s-maxage=60`) — but ONLY if the render is auth-INDEPENDENT: it must NOT read `props.auth` (reading it at all opts out, even for anonymous), set no cookie, and the app must not run strict per-caller policies (`PYLON_STRICT_FN_POLICIES`). `export const dynamic = "force-static"` caches until the next deploy; `"force-dynamic"` never caches. Fail-closed: without the opt-in (or if any condition fails) the page is `no-cache`. A page that reads `auth` or sets a cookie is never shared. The SAME earned render is also kept in an **origin disk cache** (`.pylon/.cache/ssr`): a cookie-less GET with no query string is served straight off disk for the TTL — skipping the render entirely — then re-rendered live when stale. The disk cache is namespaced per deploy (wiped on each new build) and OFF in `pylon dev` (so an edit is never masked by a stale entry); invalidation is by the `revalidate` TTL or the next deploy.
-- **No-JS forms use `route.ts` + `<Form>`.** Drop `app/.../route.ts` exporting `export const POST: RouteHandler = async ({ form, db, response, auth }) => { await db.insert("X", {...}); response.redirect("/x?ok=1"); }` (303 POST-redirect-GET by default). Render `<Form action="/x">` (from @pylonsync/react) with plain `<input name=...>` — works with JS off (native POST→handler→redirect) and is enhanced to no-reload when JS is on. The handler's `db` is read+write (mutation trust model — gate on `auth`); CSRF is automatic (Origin gate + SameSite=Lax). Multipart/file uploads aren't supported yet — use urlencoded forms + `/api/files`.
-- **`loading.tsx` streams a skeleton while the page's data resolves.** Drop `app/.../loading.tsx` (default export, page props) and the nearest one becomes a route-level Suspense fallback: Pylon flushes the shell + skeleton immediately, then reveals the real page when its top-level `use(serverData…)` resolves (no blank page). It only shows when the PAGE suspends — a page that wraps its own `<Suspense>` around a child (like `/dashboard` in this template) handles that itself. The skeleton is SERVER-ONLY: don't read `serverData` in it. A page with no `loading.tsx` is buffered (unchanged).
-- **`export const streaming = true` streams a page's OWN inner `<Suspense>` boundaries.** Without it (and without a `loading.tsx`), a page is BUFFERED — the whole document, including suspended children, resolves before the first byte. Opt in and the shell + each inner `<Suspense>` fallback flush immediately, then each boundary's real content streams in as its data resolves (multi-boundary progressive streaming). It's opt-in because it changes the response timing contract: a streaming render commits its HTTP head BEFORE suspended subtrees finish, so (a) it's never CDN/disk cacheable — don't combine with `export const revalidate`; (b) `response.setStatus/setCookie/redirect/notFound` only take effect from the SYNCHRONOUS shell render — a call from inside a suspended subtree is dropped (the runtime logs a loud warning naming what was lost); (c) a `throw` from a deep `<Suspense>` child resolves via its nearest `error.tsx` at HTTP 200, not a 5xx. Hydration is clean for any number of boundaries (the data blob ships before hydration runs). Type the config with `import type { RouteSegmentConfig } from "@pylonsync/react"`.
-- **`error.tsx` / `not-found.tsx` boundaries are HYDRATED (interactive).** `app/.../error.tsx` catches a throw below it (HTTP 500) and receives `{ error: { message, digest }, reset }` (`import type { ErrorBoundaryProps }`) — `reset()` re-attempts the route; the stack NEVER reaches the client (dev overlay + logs only). `app/.../not-found.tsx` renders at 404 (also for `response.notFound()`) and gets the page props (`NotFoundProps`), no `reset`. Both run useState/onClick/hooks.
-- **Client navigation hooks live in @pylonsync/react.** `useRouter()` → `{ push, replace, back, forward, refresh, prefetch }`; `useSearchParams()` → reactive `URLSearchParams`; `usePathname()` → reactive pathname. The hooks are CLIENT-reactive — during SSR they return defaults (empty params / "/"); for server-side URL values read the `url` / `searchParams` page props.
-- **Dynamic + catch-all routes follow Next conventions.** `app/blog/[slug]/page.tsx` → `params.slug`. `app/docs/[...path]/page.tsx` is a catch-all (matches `/docs/a/b/c`; `params.path === "a/b/c"` — `.split("/")` for segments). `app/shop/[[...filters]]/page.tsx` is an optional catch-all (also matches the bare `/shop`, with `params.filters === ""`). A catch-all must be the last segment; static beats dynamic beats catch-all on overlap.
+- **Policies deny by default; server functions BYPASS them.** Direct client CRUD (`/api/entities/*`) and sync are policy-checked. Functions run with full DB access; enforce trust with `ctx.auth` checks inside the handler, not policies.
+- **Type page props from the SDK, don't hand-roll them.** `import type { PageProps, Metadata } from "@pylonsync/react"`. Every page/layout gets `{ url, params, searchParams, auth, response, serverData }`; `PageProps<{ slug: string }>` types a `[slug]` route's params. Request headers/cookies are intentionally NOT on `PageProps`; they're server-only and stripped from hydration, so reading them in the render would mismatch.
+- **Anonymous output caching is opt-in and conditional.** `export const revalidate = 60` makes a page CDN-cacheable (`public, s-maxage=60`) only when the render is auth-independent: it must not read `props.auth`, set a cookie, or run with strict per-caller policies (`PYLON_STRICT_FN_POLICIES`). `export const dynamic = "force-static"` caches until the next deploy; `"force-dynamic"` never caches. When any condition fails, the page is `no-cache`. Eligible renders also use the origin disk cache at `.pylon/.cache/ssr`: a cookie-less GET without a query string is served from disk for the TTL and rerendered when stale. The cache is namespaced per deploy, cleared by each build, disabled in `pylon dev`, and invalidated by the `revalidate` TTL or the next deploy.
+- **No-JS forms use `route.ts` and `<Form>`.** Add `app/.../route.ts` exporting `export const POST: RouteHandler = async ({ form, db, response, auth }) => { await db.insert("X", {...}); response.redirect("/x?ok=1"); }` (303 POST-redirect-GET by default). Render `<Form action="/x">` from `@pylonsync/react` with plain `<input name=...>`. It uses native POST → handler → redirect without JavaScript and no-reload enhancement with JavaScript. The handler's `db` is read+write under the mutation trust model, so gate it on `auth`. CSRF protection is automatic through the Origin gate and SameSite=Lax. Multipart uploads are not supported yet; use URL-encoded forms and `/api/files`.
+- **`loading.tsx` streams a skeleton while the page's data resolves.** Drop `app/.../loading.tsx` (default export, page props) and the nearest one becomes a route-level Suspense fallback: Pylon flushes the shell + skeleton immediately, then reveals the real page when its top-level `use(serverData…)` resolves (no blank page). It only shows when the PAGE suspends; a page that wraps its own `<Suspense>` around a child (like `/dashboard` in this template) handles that itself. The skeleton is SERVER-ONLY: don't read `serverData` in it. A page with no `loading.tsx` is buffered (unchanged).
+- **`export const streaming = true` streams a page's inner `<Suspense>` boundaries.** Without it or a `loading.tsx`, the page is buffered until all suspended children resolve. With it, the shell and fallbacks flush immediately, then each boundary streams its content. Streaming commits the HTTP head before suspended subtrees finish, so the page is never CDN- or disk-cacheable; do not combine it with `export const revalidate`. Calls to `response.setStatus`, `setCookie`, `redirect`, or `notFound` only take effect during the synchronous shell render. A call from a suspended subtree is dropped and logged. An error from a deep `<Suspense>` child resolves through the nearest `error.tsx` at HTTP 200 rather than 5xx. Type the config with `import type { RouteSegmentConfig } from "@pylonsync/react"`.
+- **`error.tsx` / `not-found.tsx` boundaries are HYDRATED (interactive).** `app/.../error.tsx` catches a throw below it (HTTP 500) and receives `{ error: { message, digest }, reset }` (`import type { ErrorBoundaryProps }`); `reset()` re-attempts the route; the stack NEVER reaches the client (dev overlay + logs only). `app/.../not-found.tsx` renders at 404 (also for `response.notFound()`) and gets the page props (`NotFoundProps`), no `reset`. Both run useState/onClick/hooks.
+- **Client navigation hooks live in @pylonsync/react.** `useRouter()` → `{ push, replace, back, forward, refresh, prefetch }`; `useSearchParams()` → reactive `URLSearchParams`; `usePathname()` → reactive pathname. The hooks are CLIENT-reactive; during SSR they return defaults (empty params / "/"); for server-side URL values read the `url` / `searchParams` page props.
+- **Dynamic + catch-all routes follow Next conventions.** `app/blog/[slug]/page.tsx` → `params.slug`. `app/docs/[...path]/page.tsx` is a catch-all (matches `/docs/a/b/c`; `params.path === "a/b/c"`; `.split("/")` for segments). `app/shop/[[...filters]]/page.tsx` is an optional catch-all (also matches the bare `/shop`, with `params.filters === ""`). A catch-all must be the last segment; static beats dynamic beats catch-all on overlap.
 - **`serverData` (SSR) is READ-ONLY.** No write methods; the runtime rejects write frames (`SSR_WRITE_FORBIDDEN`). Mutations belong in actions/functions, never in a page render.
-- **`response.*` / `response.redirect()` / `response.notFound()` must fire in the synchronous shell render**, before any `await` / `<Suspense>`. The HTTP head commits when the shell is ready — status/headers/cookies set from a suspended subtree are lost, and `redirect`/`notFound` thrown below a Suspense boundary are swallowed.
-- **`ctx.llm` and `ctx.connections` are on mutation + action only, NOT query** (reactive purity). `action` has no direct `ctx.db` — use `ctx.runQuery` / `ctx.runMutation`.
+- **`response.*` / `response.redirect()` / `response.notFound()` must fire in the synchronous shell render**, before any `await` / `<Suspense>`. The HTTP head commits when the shell is ready; status/headers/cookies set from a suspended subtree are lost, and `redirect`/`notFound` thrown below a Suspense boundary are swallowed.
+- **`ctx.llm` and `ctx.connections` are on mutation + action only, NOT query** (reactive purity). `action` has no direct `ctx.db`; use `ctx.runQuery` / `ctx.runMutation`.
 - **It's `db.useQueryOne`, not `useOne`.** Validators and field types have aliases: `v.bool`/`v.boolean`, `v.float`/`v.number`.
-- **There is no `ctx.files` or `defineWorkflow`/`defineJob`.** Files go through `<FileUpload>` + `/api/files/*`. Deferred (one-shot) execution is `ctx.scheduler.runAfter/runAt/cancel`. Recurring work is a **cron**: `cron("0 * * * *", "fnName")` in `buildManifest({ crons: [...] })` (import `cron` from `@pylonsync/sdk`) — it fires the named function (make it `internal: true`) on the schedule; the function runs with anonymous auth — its own `ctx.db.*` is server-side (not policy-gated), so write directly; only `ctx.auth.elevate({ admin: true, reason: "..." })` (reason mandatory) to chain an `internal: true` function via `ctx.scheduler`.
+- **Use the supported file and scheduling APIs.** Files go through `<FileUpload>` and `/api/files/*`; there is no `ctx.files`. One-shot work uses `ctx.scheduler.runAfter`, `runAt`, or `cancel`; there is no `defineWorkflow` or `defineJob`. Recurring work uses `cron("0 * * * *", "fnName")` in `buildManifest({ crons: [...] })`, imported from `@pylonsync/sdk`. Make the target function `internal: true`. It runs with anonymous auth, but its own `ctx.db.*` calls are server-side and bypass policies. Use `ctx.auth.elevate({ admin: true, reason: "..." })`, with a mandatory reason, only when chaining another internal function through `ctx.scheduler`.
 
 ## Testing
 
-`pylon test` discovers every `*.test.ts` / `*.test.tsx` file under `tests/` (or `functions/`) and runs it with **Bun's test runner** (`import { test, expect } from "bun:test"`). Run the suite with `pylon test` (or `npm test`); filter with `pylon test <substring>`. This template ships `bunfig.toml` + `tests/setup.ts` (registers happy-dom) so component tests render out of the box, plus starter tests under `tests/` — replace them with your own.
+`pylon test` discovers every `*.test.ts` / `*.test.tsx` file under `tests/` (or `functions/`) and runs it with **Bun's test runner** (`import { test, expect } from "bun:test"`). Run the suite with `pylon test` (or `npm test`); filter with `pylon test <substring>`. This template ships `bunfig.toml` + `tests/setup.ts` (registers happy-dom) so component tests render out of the box, plus starter tests under `tests/`; replace them with your own.
 
-**Tier 1 — pure logic (reach for this first).** Keep the decisions that matter — access/plan gating, pricing, credit math, validation, formatting — in pure functions in `lib/`, and test those exhaustively. No server, instant, and it's where the real bugs live. Keep your `query`/`mutation`/`action` handlers as thin wrappers around them, so the logic is testable without a running app.
+**Tier 1: pure logic (start here).** Keep access and plan gating, pricing, credit math, validation, and formatting in pure functions under `lib/`, and test them exhaustively. These tests need no server and run instantly. Keep `query`, `mutation`, and `action` handlers as thin wrappers so their decision logic remains testable without a running app.
 
 ```ts
 import { expect, test } from "bun:test";
@@ -52,7 +52,7 @@ test("unknown slug → undefined", () => {
 });
 ```
 
-**Tier 2 — React components.** `@testing-library/react` + happy-dom are already wired (`tests/setup.ts`). Render and assert. The template uses the classic JSX transform, so add `import React from "react"` in `.tsx` tests. For a component that reads Pylon data hooks, **mock the boundary**, then dynamic-`import` the component so the mock is in place first:
+**Tier 2: React components.** `@testing-library/react` and happy-dom are already wired through `tests/setup.ts`. Render and assert. The template uses the classic JSX transform, so add `import React from "react"` in `.tsx` tests. For a component that reads Pylon data hooks, mock the boundary, then dynamically import the component so the mock is in place first:
 
 ```tsx
 import { test, expect, mock } from "bun:test";
@@ -70,7 +70,7 @@ test("renders orgs from the query", () => {
 });
 ```
 
-**Tier 3 — functions over HTTP (only when Tier 1 can't cover it).** A handler's full behavior (policies, `ctx.db`, auth) lives in the running app. Start `pylon dev` in another terminal and call the API; `resetDb()` from `@pylonsync/functions` clears the in-memory DB between cases (no-ops if the server's down, refuses production).
+**Tier 3: functions over HTTP.** Use this tier when pure logic tests cannot cover the behavior. A handler's policies, `ctx.db` calls, and auth run in the app. Start `pylon dev` in another terminal and call the API. `resetDb()` from `@pylonsync/functions` clears the in-memory database between cases; it does nothing when the server is down and refuses to run in production.
 
 ```ts
 import { afterEach, expect, test } from "bun:test";
@@ -90,9 +90,9 @@ test("createThing then read it back", async () => {
 });
 ```
 
-`pylon test:security` is a separate adversarial probe — it hits a running app and reports auth/policy holes (run `pylon dev`, then `pylon test:security`).
+`pylon test:security` is a separate adversarial probe; it hits a running app and reports auth/policy holes (run `pylon dev`, then `pylon test:security`).
 
-## Use the CLI — don't guess
+## Use the CLI
 
 | Need | Command |
 |---|---|

package/templates/todo/CLAUDE.md

@@ -1,7 +1,7 @@
 # CLAUDE.md
 
-Agent guidance for this Pylon project lives in **AGENTS.md** — the cross-editor
-standard that Claude Code, Cursor, Codex, and others all read. It's imported
-below so Claude Code picks it up automatically:
+Agent guidance for this Pylon project lives in **AGENTS.md**, the cross-editor
+standard read by Claude Code, Cursor, Codex, and other agents. The import below
+makes it available to Claude Code:
 
 @AGENTS.md

package/templates/todo/README.md

@@ -1,8 +1,7 @@
 # __APP_NAME__
 
-A full-stack [Pylon](https://pylonsync.com) todo app — a server-rendered page
-and a live, optimistic, per-user todo list over a synced database, all served
-from one binary on one port. No Next.js, no separate API server.
+A full-stack [Pylon](https://pylonsync.com) todo app with a server-rendered page
+and a live, optimistic, per-user list over a synced database.
 
 ## Develop
 
@@ -10,9 +9,8 @@ from one binary on one port. No Next.js, no separate API server.
 __RUN_DEV__
 ```
 
-Open http://localhost:4321. Add a todo — it appears instantly (optimistic) and
-syncs; open a second tab to watch writes arrive live. Edit any file under
-`app/` and save — the page reloads instantly.
+Open http://localhost:4321 and add a todo. It appears optimistically and syncs
+to a second tab. Editing a file under `app/` reloads the page.
 
 ## Layout
 

package/templates/waitlist/AGENTS.md

@@ -1,47 +1,47 @@
-# AGENTS.md — working in a Pylon project
+# AGENTS.md: working in a Pylon project
 
-Operating rules for a coding agent in this Pylon app. You — the agent — are a first-class user of Pylon: one Rust binary (`pylon`) serves the API, auth, sync, WebSocket, SSE, and native React 19 SSR from one process on one port, so you build, run, and ship a whole app without wiring services together or leaving the codebase. This is production infrastructure, not a sandbox — real auth, SQLite or Postgres, row-level policies, jobs, search, and one-command deploy — so build like it ships. You declare entities, policies, and server functions in TypeScript; the binary does the rest. The full API reference is the **llms-full.txt** at https://docs.pylonsync.com/llms-full.txt — read it before guessing an API name.
+Pylon serves the API, auth, sync, WebSocket, SSE, and native React 19 SSR from one Rust process on one port. Treat this app as production infrastructure: it supports real auth, SQLite or Postgres, row-level policies, jobs, search, and one-command deploys. Declare entities, policies, and server functions in TypeScript; the binary handles the runtime. Read the full API reference at https://docs.pylonsync.com/llms-full.txt before guessing an API name.
 
 ## Directory conventions
 
 **Unified SSR app:**
-- `app.ts` — data model + manifest (`entity()` + `field.*`, queries/actions/policies, `routes: await discoverAppRoutes()`). Ends with `console.log(JSON.stringify(manifest))`.
-- `app/` — file-based SSR routes. `app/page.tsx` → `/`, `app/about/page.tsx` → `/about`, `app/blog/[slug]/page.tsx` → `/blog/:slug`. `app/layout.tsx` is the shell; `app/error.tsx` / `app/not-found.tsx` are boundaries.
-- `app/globals.css` — Tailwind v4 entrypoint (auto-compiled and injected).
-- `functions/` — server functions, one per file, `default`-exported.
-- `.pylon/` — local dev state (sqlite, jobs, sessions, uploads). Created by `pylon dev`. Do not commit.
+- `app.ts`: data model + manifest (`entity()` + `field.*`, queries/actions/policies, `routes: await discoverAppRoutes()`). Ends with `console.log(JSON.stringify(manifest))`.
+- `app/`: file-based SSR routes. `app/page.tsx` → `/`, `app/about/page.tsx` → `/about`, `app/blog/[slug]/page.tsx` → `/blog/:slug`. `app/layout.tsx` is the shell; `app/error.tsx` / `app/not-found.tsx` are boundaries.
+- `app/globals.css`: Tailwind v4 entrypoint (auto-compiled and injected).
+- `functions/`: server functions, one per file, `default`-exported.
+- `.pylon/`: local dev state (SQLite, jobs, sessions, uploads). Created by `pylon dev`. Do not commit.
 
-**Monorepo app:** backend is `apps/api/` (entry `apps/api/schema.ts`, handlers in `apps/api/functions/`); frontend in `apps/web/`. `pylon.manifest.json` / `pylon.client.ts` are generated — do not hand-edit.
+**Monorepo app:** backend is `apps/api/` (entry `apps/api/schema.ts`, handlers in `apps/api/functions/`); frontend in `apps/web/`. `pylon.manifest.json` / `pylon.client.ts` are generated; do not hand-edit.
 
 ## The core authoring loop
 
-1. **Define an entity** — `entity("Thing", { name: field.string(), done: field.boolean().default(false) })`. Modifiers: `.optional()`, `.unique()`, `.readonly()` (settable on insert, rejected on client update — use for `authorId`/`orgId`), `.serverOnly()` (never in HTTP responses), `.encrypted()` (AEAD at rest, needs `PYLON_ENCRYPTION_KEY`), `.crdt("text")` (collaborative).
-2. **Write a policy** — `policy({ entity: "Thing", allowRead, allowInsert, allowUpdate, allowDelete })` with CEL-like expressions over `auth.*` / `data.*` (e.g. `"auth.userId == data.authorId"`). **Omitted actions DENY by default.** Wide-open dev policies (`allow*: "true"`) are flagged by `pylon lint` — tighten before shipping.
-3. **Author a function** in `functions/<name>.ts` — `query` (read-only), `mutation` (transactional read+write), or `action` (external I/O, no direct `ctx.db`). Import `{ query, mutation, action, v }` from `@pylonsync/functions`. `auth` defaults to `"user"` (secure-by-default); set `"public"` explicitly for unauthenticated access. Use `ctx.db.*`, `ctx.auth.userId`, `ctx.error(code, msg)`.
-4. **Read it on the client** — `db.useQuery("Thing")` (live, re-renders on any write) or `db.useQueryOne("Thing", id)`. Call functions with `db.fn(name, args)` / `callFn`. On SSR pages, read via `use(serverData.list("Thing"))` inside `<Suspense>`.
+1. **Define an entity:** `entity("Thing", { name: field.string(), done: field.boolean().default(false) })`. Modifiers: `.optional()`, `.unique()`, `.readonly()` (settable on insert, rejected on client update; use for `authorId`/`orgId`), `.serverOnly()` (never in HTTP responses), `.encrypted()` (AEAD at rest, needs `PYLON_ENCRYPTION_KEY`), `.crdt("text")` (collaborative).
+2. **Write a policy:** `policy({ entity: "Thing", allowRead, allowInsert, allowUpdate, allowDelete })` with CEL-like expressions over `auth.*` / `data.*` (e.g. `"auth.userId == data.authorId"`). Omitted actions deny by default. `pylon lint` flags wide-open development policies such as `allow*: "true"`; tighten them before shipping.
+3. **Author a function** in `functions/<name>.ts`: `query` (read-only), `mutation` (transactional read+write), or `action` (external I/O, no direct `ctx.db`). Import `{ query, mutation, action, v }` from `@pylonsync/functions`. `auth` defaults to `"user"` (secure-by-default); set `"public"` explicitly for unauthenticated access. Use `ctx.db.*`, `ctx.auth.userId`, `ctx.error(code, msg)`.
+4. **Read it on the client:** `db.useQuery("Thing")` (live, re-renders on any write) or `db.useQueryOne("Thing", id)`. Call functions with `db.fn(name, args)` / `callFn`. On SSR pages, read via `use(serverData.list("Thing"))` inside `<Suspense>`.
 
 ## Key gotchas
 
-- **Policies deny by default; server functions BYPASS them.** Direct client CRUD (`/api/entities/*`) and sync are policy-checked. Functions run with full DB access — enforce trust with `ctx.auth` checks inside the handler, not policies.
-- **Type page props from the SDK, don't hand-roll them.** `import type { PageProps, Metadata } from "@pylonsync/react"`. Every page/layout gets `{ url, params, searchParams, auth, response, serverData }`; `PageProps<{ slug: string }>` types a `[slug]` route's params. Request headers/cookies are intentionally NOT on `PageProps` — they're server-only and stripped from hydration, so reading them in the render would mismatch.
-- **Anonymous output caching is opt-in + earned.** `export const revalidate = 60` (seconds) on a page makes it CDN-cacheable (`public, s-maxage=60`) — but ONLY if the render is auth-INDEPENDENT: it must NOT read `props.auth` (reading it at all opts out, even for anonymous), set no cookie, and the app must not run strict per-caller policies (`PYLON_STRICT_FN_POLICIES`). `export const dynamic = "force-static"` caches until the next deploy; `"force-dynamic"` never caches. Fail-closed: without the opt-in (or if any condition fails) the page is `no-cache`. A page that reads `auth` or sets a cookie is never shared. The SAME earned render is also kept in an **origin disk cache** (`.pylon/.cache/ssr`): a cookie-less GET with no query string is served straight off disk for the TTL — skipping the render entirely — then re-rendered live when stale. The disk cache is namespaced per deploy (wiped on each new build) and OFF in `pylon dev` (so an edit is never masked by a stale entry); invalidation is by the `revalidate` TTL or the next deploy.
-- **No-JS forms use `route.ts` + `<Form>`.** Drop `app/.../route.ts` exporting `export const POST: RouteHandler = async ({ form, db, response, auth }) => { await db.insert("X", {...}); response.redirect("/x?ok=1"); }` (303 POST-redirect-GET by default). Render `<Form action="/x">` (from @pylonsync/react) with plain `<input name=...>` — works with JS off (native POST→handler→redirect) and is enhanced to no-reload when JS is on. The handler's `db` is read+write (mutation trust model — gate on `auth`); CSRF is automatic (Origin gate + SameSite=Lax). Multipart/file uploads aren't supported yet — use urlencoded forms + `/api/files`.
-- **`loading.tsx` streams a skeleton while the page's data resolves.** Drop `app/.../loading.tsx` (default export, page props) and the nearest one becomes a route-level Suspense fallback: Pylon flushes the shell + skeleton immediately, then reveals the real page when its top-level `use(serverData…)` resolves (no blank page). It only shows when the PAGE suspends — a page that wraps its own `<Suspense>` around a child (like `/dashboard` in this template) handles that itself. The skeleton is SERVER-ONLY: don't read `serverData` in it. A page with no `loading.tsx` is buffered (unchanged).
-- **`export const streaming = true` streams a page's OWN inner `<Suspense>` boundaries.** Without it (and without a `loading.tsx`), a page is BUFFERED — the whole document, including suspended children, resolves before the first byte. Opt in and the shell + each inner `<Suspense>` fallback flush immediately, then each boundary's real content streams in as its data resolves (multi-boundary progressive streaming). It's opt-in because it changes the response timing contract: a streaming render commits its HTTP head BEFORE suspended subtrees finish, so (a) it's never CDN/disk cacheable — don't combine with `export const revalidate`; (b) `response.setStatus/setCookie/redirect/notFound` only take effect from the SYNCHRONOUS shell render — a call from inside a suspended subtree is dropped (the runtime logs a loud warning naming what was lost); (c) a `throw` from a deep `<Suspense>` child resolves via its nearest `error.tsx` at HTTP 200, not a 5xx. Hydration is clean for any number of boundaries (the data blob ships before hydration runs). Type the config with `import type { RouteSegmentConfig } from "@pylonsync/react"`.
-- **`error.tsx` / `not-found.tsx` boundaries are HYDRATED (interactive).** `app/.../error.tsx` catches a throw below it (HTTP 500) and receives `{ error: { message, digest }, reset }` (`import type { ErrorBoundaryProps }`) — `reset()` re-attempts the route; the stack NEVER reaches the client (dev overlay + logs only). `app/.../not-found.tsx` renders at 404 (also for `response.notFound()`) and gets the page props (`NotFoundProps`), no `reset`. Both run useState/onClick/hooks.
-- **Client navigation hooks live in @pylonsync/react.** `useRouter()` → `{ push, replace, back, forward, refresh, prefetch }`; `useSearchParams()` → reactive `URLSearchParams`; `usePathname()` → reactive pathname. The hooks are CLIENT-reactive — during SSR they return defaults (empty params / "/"); for server-side URL values read the `url` / `searchParams` page props.
-- **Dynamic + catch-all routes follow Next conventions.** `app/blog/[slug]/page.tsx` → `params.slug`. `app/docs/[...path]/page.tsx` is a catch-all (matches `/docs/a/b/c`; `params.path === "a/b/c"` — `.split("/")` for segments). `app/shop/[[...filters]]/page.tsx` is an optional catch-all (also matches the bare `/shop`, with `params.filters === ""`). A catch-all must be the last segment; static beats dynamic beats catch-all on overlap.
+- **Policies deny by default; server functions BYPASS them.** Direct client CRUD (`/api/entities/*`) and sync are policy-checked. Functions run with full DB access; enforce trust with `ctx.auth` checks inside the handler, not policies.
+- **Type page props from the SDK, don't hand-roll them.** `import type { PageProps, Metadata } from "@pylonsync/react"`. Every page/layout gets `{ url, params, searchParams, auth, response, serverData }`; `PageProps<{ slug: string }>` types a `[slug]` route's params. Request headers/cookies are intentionally NOT on `PageProps`; they're server-only and stripped from hydration, so reading them in the render would mismatch.
+- **Anonymous output caching is opt-in and conditional.** `export const revalidate = 60` makes a page CDN-cacheable (`public, s-maxage=60`) only when the render is auth-independent: it must not read `props.auth`, set a cookie, or run with strict per-caller policies (`PYLON_STRICT_FN_POLICIES`). `export const dynamic = "force-static"` caches until the next deploy; `"force-dynamic"` never caches. When any condition fails, the page is `no-cache`. Eligible renders also use the origin disk cache at `.pylon/.cache/ssr`: a cookie-less GET without a query string is served from disk for the TTL and rerendered when stale. The cache is namespaced per deploy, cleared by each build, disabled in `pylon dev`, and invalidated by the `revalidate` TTL or the next deploy.
+- **No-JS forms use `route.ts` and `<Form>`.** Add `app/.../route.ts` exporting `export const POST: RouteHandler = async ({ form, db, response, auth }) => { await db.insert("X", {...}); response.redirect("/x?ok=1"); }` (303 POST-redirect-GET by default). Render `<Form action="/x">` from `@pylonsync/react` with plain `<input name=...>`. It uses native POST → handler → redirect without JavaScript and no-reload enhancement with JavaScript. The handler's `db` is read+write under the mutation trust model, so gate it on `auth`. CSRF protection is automatic through the Origin gate and SameSite=Lax. Multipart uploads are not supported yet; use URL-encoded forms and `/api/files`.
+- **`loading.tsx` streams a skeleton while the page's data resolves.** Drop `app/.../loading.tsx` (default export, page props) and the nearest one becomes a route-level Suspense fallback: Pylon flushes the shell + skeleton immediately, then reveals the real page when its top-level `use(serverData…)` resolves (no blank page). It only shows when the PAGE suspends; a page that wraps its own `<Suspense>` around a child (like `/dashboard` in this template) handles that itself. The skeleton is SERVER-ONLY: don't read `serverData` in it. A page with no `loading.tsx` is buffered (unchanged).
+- **`export const streaming = true` streams a page's inner `<Suspense>` boundaries.** Without it or a `loading.tsx`, the page is buffered until all suspended children resolve. With it, the shell and fallbacks flush immediately, then each boundary streams its content. Streaming commits the HTTP head before suspended subtrees finish, so the page is never CDN- or disk-cacheable; do not combine it with `export const revalidate`. Calls to `response.setStatus`, `setCookie`, `redirect`, or `notFound` only take effect during the synchronous shell render. A call from a suspended subtree is dropped and logged. An error from a deep `<Suspense>` child resolves through the nearest `error.tsx` at HTTP 200 rather than 5xx. Type the config with `import type { RouteSegmentConfig } from "@pylonsync/react"`.
+- **`error.tsx` / `not-found.tsx` boundaries are HYDRATED (interactive).** `app/.../error.tsx` catches a throw below it (HTTP 500) and receives `{ error: { message, digest }, reset }` (`import type { ErrorBoundaryProps }`); `reset()` re-attempts the route; the stack NEVER reaches the client (dev overlay + logs only). `app/.../not-found.tsx` renders at 404 (also for `response.notFound()`) and gets the page props (`NotFoundProps`), no `reset`. Both run useState/onClick/hooks.
+- **Client navigation hooks live in @pylonsync/react.** `useRouter()` → `{ push, replace, back, forward, refresh, prefetch }`; `useSearchParams()` → reactive `URLSearchParams`; `usePathname()` → reactive pathname. The hooks are CLIENT-reactive; during SSR they return defaults (empty params / "/"); for server-side URL values read the `url` / `searchParams` page props.
+- **Dynamic + catch-all routes follow Next conventions.** `app/blog/[slug]/page.tsx` → `params.slug`. `app/docs/[...path]/page.tsx` is a catch-all (matches `/docs/a/b/c`; `params.path === "a/b/c"`; `.split("/")` for segments). `app/shop/[[...filters]]/page.tsx` is an optional catch-all (also matches the bare `/shop`, with `params.filters === ""`). A catch-all must be the last segment; static beats dynamic beats catch-all on overlap.
 - **`serverData` (SSR) is READ-ONLY.** No write methods; the runtime rejects write frames (`SSR_WRITE_FORBIDDEN`). Mutations belong in actions/functions, never in a page render.
-- **`response.*` / `response.redirect()` / `response.notFound()` must fire in the synchronous shell render**, before any `await` / `<Suspense>`. The HTTP head commits when the shell is ready — status/headers/cookies set from a suspended subtree are lost, and `redirect`/`notFound` thrown below a Suspense boundary are swallowed.
-- **`ctx.llm` and `ctx.connections` are on mutation + action only, NOT query** (reactive purity). `action` has no direct `ctx.db` — use `ctx.runQuery` / `ctx.runMutation`.
+- **`response.*` / `response.redirect()` / `response.notFound()` must fire in the synchronous shell render**, before any `await` / `<Suspense>`. The HTTP head commits when the shell is ready; status/headers/cookies set from a suspended subtree are lost, and `redirect`/`notFound` thrown below a Suspense boundary are swallowed.
+- **`ctx.llm` and `ctx.connections` are on mutation + action only, NOT query** (reactive purity). `action` has no direct `ctx.db`; use `ctx.runQuery` / `ctx.runMutation`.
 - **It's `db.useQueryOne`, not `useOne`.** Validators and field types have aliases: `v.bool`/`v.boolean`, `v.float`/`v.number`.
-- **There is no `ctx.files` or `defineWorkflow`/`defineJob`.** Files go through `<FileUpload>` + `/api/files/*`. Deferred (one-shot) execution is `ctx.scheduler.runAfter/runAt/cancel`. Recurring work is a **cron**: `cron("0 * * * *", "fnName")` in `buildManifest({ crons: [...] })` (import `cron` from `@pylonsync/sdk`) — it fires the named function (make it `internal: true`) on the schedule; the function runs with anonymous auth — its own `ctx.db.*` is server-side (not policy-gated), so write directly; only `ctx.auth.elevate({ admin: true, reason: "..." })` (reason mandatory) to chain an `internal: true` function via `ctx.scheduler`.
+- **Use the supported file and scheduling APIs.** Files go through `<FileUpload>` and `/api/files/*`; there is no `ctx.files`. One-shot work uses `ctx.scheduler.runAfter`, `runAt`, or `cancel`; there is no `defineWorkflow` or `defineJob`. Recurring work uses `cron("0 * * * *", "fnName")` in `buildManifest({ crons: [...] })`, imported from `@pylonsync/sdk`. Make the target function `internal: true`. It runs with anonymous auth, but its own `ctx.db.*` calls are server-side and bypass policies. Use `ctx.auth.elevate({ admin: true, reason: "..." })`, with a mandatory reason, only when chaining another internal function through `ctx.scheduler`.
 
 ## Testing
 
-`pylon test` discovers every `*.test.ts` / `*.test.tsx` file under `tests/` (or `functions/`) and runs it with **Bun's test runner** (`import { test, expect } from "bun:test"`). Run the suite with `pylon test` (or `npm test`); filter with `pylon test <substring>`. This template ships `bunfig.toml` + `tests/setup.ts` (registers happy-dom) so component tests render out of the box, plus starter tests under `tests/` — replace them with your own.
+`pylon test` discovers every `*.test.ts` / `*.test.tsx` file under `tests/` (or `functions/`) and runs it with **Bun's test runner** (`import { test, expect } from "bun:test"`). Run the suite with `pylon test` (or `npm test`); filter with `pylon test <substring>`. This template ships `bunfig.toml` + `tests/setup.ts` (registers happy-dom) so component tests render out of the box, plus starter tests under `tests/`; replace them with your own.
 
-**Tier 1 — pure logic (reach for this first).** Keep the decisions that matter — access/plan gating, pricing, credit math, validation, formatting — in pure functions in `lib/`, and test those exhaustively. No server, instant, and it's where the real bugs live. Keep your `query`/`mutation`/`action` handlers as thin wrappers around them, so the logic is testable without a running app.
+**Tier 1: pure logic (start here).** Keep access and plan gating, pricing, credit math, validation, and formatting in pure functions under `lib/`, and test them exhaustively. These tests need no server and run instantly. Keep `query`, `mutation`, and `action` handlers as thin wrappers so their decision logic remains testable without a running app.
 
 ```ts
 import { expect, test } from "bun:test";
@@ -52,7 +52,7 @@ test("unknown slug → undefined", () => {
 });
 ```
 
-**Tier 2 — React components.** `@testing-library/react` + happy-dom are already wired (`tests/setup.ts`). Render and assert. The template uses the classic JSX transform, so add `import React from "react"` in `.tsx` tests. For a component that reads Pylon data hooks, **mock the boundary**, then dynamic-`import` the component so the mock is in place first:
+**Tier 2: React components.** `@testing-library/react` and happy-dom are already wired through `tests/setup.ts`. Render and assert. The template uses the classic JSX transform, so add `import React from "react"` in `.tsx` tests. For a component that reads Pylon data hooks, mock the boundary, then dynamically import the component so the mock is in place first:
 
 ```tsx
 import { test, expect, mock } from "bun:test";
@@ -70,7 +70,7 @@ test("renders orgs from the query", () => {
 });
 ```
 
-**Tier 3 — functions over HTTP (only when Tier 1 can't cover it).** A handler's full behavior (policies, `ctx.db`, auth) lives in the running app. Start `pylon dev` in another terminal and call the API; `resetDb()` from `@pylonsync/functions` clears the in-memory DB between cases (no-ops if the server's down, refuses production).
+**Tier 3: functions over HTTP.** Use this tier when pure logic tests cannot cover the behavior. A handler's policies, `ctx.db` calls, and auth run in the app. Start `pylon dev` in another terminal and call the API. `resetDb()` from `@pylonsync/functions` clears the in-memory database between cases; it does nothing when the server is down and refuses to run in production.
 
 ```ts
 import { afterEach, expect, test } from "bun:test";
@@ -90,9 +90,9 @@ test("createThing then read it back", async () => {
 });
 ```
 
-`pylon test:security` is a separate adversarial probe — it hits a running app and reports auth/policy holes (run `pylon dev`, then `pylon test:security`).
+`pylon test:security` is a separate adversarial probe; it hits a running app and reports auth/policy holes (run `pylon dev`, then `pylon test:security`).
 
-## Use the CLI — don't guess
+## Use the CLI
 
 | Need | Command |
 |---|---|

package/templates/waitlist/CLAUDE.md

@@ -1,7 +1,7 @@
 # CLAUDE.md
 
-Agent guidance for this Pylon project lives in **AGENTS.md** — the cross-editor
-standard that Claude Code, Cursor, Codex, and others all read. It's imported
-below so Claude Code picks it up automatically:
+Agent guidance for this Pylon project lives in **AGENTS.md**, the cross-editor
+standard read by Claude Code, Cursor, Codex, and other agents. The import below
+makes it available to Claude Code:
 
 @AGENTS.md