File the decision on caching public pages

AGENTS.md

@@ -221,6 +221,9 @@ Revisit only if the stated reason stops holding.
   adds you to any group, including one holding every permission; `users:write` mints a recovery code
   for any account. The containment the split buys is real on the **read** half only (`users:read` is
   a safe helpdesk grant). Don't let the per-resource naming imply otherwise in docs.
+- **The CSRF token is deliberately unbound to the session** (`src/auth/csrf.ts`; README → Security
+  model): the usual cost of a signed double-submit token, covered by `SameSite=Lax` and HSTS.
+  **Valid while the session cookie stays `SameSite=Lax` or stricter.**
 - **Plainpages says "user" everywhere; Ory's word is "identity".** House style, not a renamed
   concept. The single exception is the `Identity` DTO in `src/auth/kratos-admin.ts`, which mirrors
   Kratos' wire shape — don't rename it.
@@ -367,7 +370,7 @@ docker compose -f compose.yml up --build -d              # production
    stack up and a *minimal* plugin live. Nothing comes before Quick start. Keep its commands
    copy-pasteable; deeper detail lives in its own section, linked.
 2. **Returning developer (rest).** A **Contents** ToC right after Quick start, then sections ordered
-   by **what an adopter reaches for first**, not by architectural layering: Overview → Users, groups
+   by **what an adopter reaches for first**, not by architectural layering: Goals → Audience → Overview → Users, groups
    & permissions → Building plugins → menu/blocks/interactivity → Configuration → Auth → Email →
    Architecture → Testing → Production → Observability → JWT-rotation runbook → Project-layout file
    map → Extending. Place a new section by how early an adopter needs it. **Users, groups &

CHANGELOG.md

@@ -4,6 +4,21 @@ The release version **is** the plugin contract version (`HOST_API_VERSION`), so
 contract break: a plugin's `apiVersion` must match the host's `major.minor` or discovery refuses it
 at boot. Entries start at 0.3.0.
 
+## Unreleased
+
+### One horizontal inset across the content pane
+
+The topbar, filter bar, pager, form page and empty-table message inset their content by `--pad-x`
+(12px), as table cells do, instead of 16px. A plugin that overrides `--pad-x` now moves all of them
+together.
+
+### Pages are not kept by a shared cache
+
+Every page and JSON response now carries `Cache-Control: private, no-store`, so a cache or reverse
+proxy in front of Plainpages can't store a signed-in page and serve it to the next visitor. A route
+whose page is the same for everyone can still set its own `cache-control` through
+`RouteResult.headers`; assets are unchanged.
+
 ## 0.4.0
 
 **Breaking.** Set `apiVersion: "0.4.0"`. The app shell no longer bounds the content column, so a page

README.md

@@ -70,6 +70,8 @@ From here, render real pages against the app shell and fetch upstream data — s
 
 ## Contents
 
+- [Goals](#goals)
+- [Audience](#audience)
 - [Overview](#overview)
 - [Users, groups & permissions](#users-groups--permissions)
   - [naming a permission](#naming-a-permission)
@@ -121,6 +123,64 @@ From here, render real pages against the app shell and fetch upstream data — s
 - [Project layout](#project-layout)
 - [Extending the core](#extending-the-core)
 
+## Goals
+
+1. **Achieve:** Less boilerplate for a server-rendered web GUI: a new project starts at
+   its own domain, not at sign-in, a menu or a table.
+   - Signal: a new plugin reaches a working, gated page with a menu, tables, filters and
+     forms built only from the core building blocks.
+   - Counter-signal: the core grows features only one project needs.
+   - Why: boilerplate removed by absorbing a project's domain only moves it into the core.
+2. **Achieve:** Every building block works for the person using it: a non-technical user
+   gets what they meant on the first try, and can't hurt themselves by accident.
+   - Signal: bugs users hit in a shipped flow get fixed and stay fixed, such as a pasted
+     code being refused, a double click acting twice, or an admin locking themselves out.
+   - Counter-signal: a fix adds client-side JavaScript, or a setting every plugin has to
+     know about.
+   - Why: a fix that breaks zero-JS or simplicity moves the irritation from users to every
+     plugin author.
+3. **Maintain:** Plainpages solves only the parts every project shares: auth, the GUI
+   building blocks (menu, tables, filters, forms) and the plugin structure. Everything
+   else is a plugin.
+4. **Maintain:** Every page is a server-rendered document that works with zero JavaScript,
+   on old and embedded browsers too. A newer browser only makes it nicer.
+5. **Maintain:** State lives in the URL. Only state the URL can't carry lives in the
+   database, never in local storage or JavaScript. A cookie carries only what must be bound
+   to the browser: the session and its CSRF token.
+6. **Maintain:** The frontend is semantic: the HTML says what each thing is, and the CSS
+   styles meaning, not looks. That keeps every page accessible.
+7. **Achieve:** Hackable: a plugin can change anything, and Plainpages stays out of its way.
+   - Signal: a plugin replaces any page, partial, menu entry or string without forking the
+     host, and fewer plugins need a host change to get unblocked.
+   - Counter-signal: plugins break on upgrade with no version signal, or a plugin mistake
+     shows up at runtime instead of at boot.
+   - Why: freedom that breaks silently costs more than the boilerplate it saved.
+8. **Achieve:** Speed: pages render fast and stay small.
+   - Signal: server time per page and bytes sent per page fall.
+   - Counter-signal: a building block loses a feature or accessibility to save bytes.
+   - Why: a fast page that does less only moves the work back into each plugin.
+9. **Achieve:** LLM-friendly: an LLM builds a correct plugin from the README and the
+   examples alone.
+   - Signal: an agent given only those writes a plugin that passes discovery and its E2E
+     test on the first try.
+   - Counter-signal: the README grows to restate what the code already shows.
+   - Why: docs that pass the test by growing cost every later reader, human or LLM.
+10. **Maintain:** Every behaviour is an explicit configuration choice, the same in every
+   environment.
+
+## Audience
+
+- **Developers starting a web GUI project, and the LLMs that build plugins with them.**
+  They may rely on `@plainpages/plugin-api`, the core partials and the packaging promises,
+  all versioned by `apiVersion`. A change that breaks a plugin takes a major bump.
+- **Operators** who self-host Plainpages. They may rely on semver-tagged images, on
+  configuration through environment variables and the drop-in mounts (`plugins/`,
+  `config/`, `locales/`), and on upgrades within a minor version that need no change on
+  their side.
+- **Visitors** to a site built on Plainpages, on any browser, including the older engines in
+  TVs, kiosks, thin clients and embedded panels. They may rely on pages that load fast and
+  work without JavaScript, with a keyboard and a screen reader.
+
 ## Overview
 
 Plainpages gives you the boring-but-hard parts of a web app — a design system, a menu,
@@ -136,11 +196,9 @@ public front with an authenticated area behind it.
 - **You add:** everything else domain-specific, as **plugins** — built from the same building
   blocks the admin plugin uses.
 
-**Priorities:** simplicity, few dependencies, strict TypeScript, no build step, Docker-only,
+Identity, sessions, SSO, OAuth2 and permission checks are delegated to **Ory** sidecar
-environment-agnostic (no `NODE_ENV` — every behaviour is an explicit config toggle). Identity,
+services. The shape doesn't change as it grows: every plugin is the same self-contained folder,
-sessions, SSO, OAuth2 and permission checks are delegated to **Ory** sidecar services. The shape
+the hot path is the same I/O-free JWT check, and the host stores nothing of its own.
-doesn't change as it grows: every plugin is the same self-contained folder, the hot path is the
-same I/O-free JWT check, and there is no app database.
 
 **Plugins are the extension model.** The plugin API is the product's main surface: powerful,
 predictable, and overloadable — a plugin can take over as much of a page as it wants. The host
@@ -961,6 +1019,13 @@ underneath it — cramped, but nothing is unreachable. List state
 (`?q=…&status=…&sort=…&page=…`) lives **in the URL**, so a view is bookmarkable and shareable; the
 URL is the only state the UI keeps.
 
+**Two browser floors** serve goal 4. Every page's content and actions stay reachable on
+Chromium 87, Safari 15 and Firefox 115, the engines in years-old TVs, kiosks and Qt panels. Only
+features [Baseline Widely available](https://web-platform-dx.github.io/baseline/) (in every
+major engine for 30 months) may shape how a page looks; anything newer, like popovers or anchor
+positioning, only enhances. A feature the floor engines lack, such as `:has()` or `oklch()`, may
+decorate a page but never decide whether something shows or can be read.
+
 **A field whose options depend on another field is asked for in steps, one GET form each**, so every
 choice is in the URL, and nothing typed is lost to one because nothing is typed until they are made.
 Once the query names a value, that field renders read-only (`field` with `readonly`, and a `link` back
@@ -1319,6 +1384,11 @@ authorization graph, so treat it the same. `compose.yml` publishes none of the s
 so a stolen cookie jar is worth 30 days of re-mintable access, not ten minutes. Only our two cookies
 obey `SECURE_COOKIES`; the Kratos one takes its flags from Kratos' own config.
 
+**The CSRF token isn't bound to the session.** It is a signed double-submit token, so an attacker who
+can write cookies on your origin can plant one they know. `SameSite=Lax` keeps a cross-site page
+out, and HSTS (`SECURE_COOKIES=true`) a network attacker. A hostile page on a sibling subdomain can
+still set cookies, so host nothing untrusted on one.
+
 **Offboarding is not instant by default** — a revoked permission or deactivated identity lands
 within one token TTL, unless the [denylist](#instant-revoke-the-optional-denylist) is on.
 
@@ -1580,6 +1650,11 @@ Every response carries security headers (`src/http/security-headers.ts`): a stri
 so a branding logo must live under `/public/` or be a `data:` URI; a plugin route can override any
 header per-response via `RouteResult.headers`.
 
+Every page and JSON response is `Cache-Control: private, no-store`, so a cache in front of `web`
+never keeps one user's page for the next. A route may override it only for a page that is the same for
+every visitor: no user, no CSRF token, no `Set-Cookie`, since the host adds a fresh CSRF cookie to
+every plugin route.
+
 A deep link reached while signed out — or after the session JWT lapses mid-task — bounces to the
 themed sign-in and returns to the **page that was requested** (`return_to`, validated host-relative
 by `localPath` in `src/http/safe-url.ts`, so a crafted value can't turn login completion into an open

e2e-tests/playwright.config.ts

@@ -10,6 +10,7 @@ export default defineConfig({
   testDir: ".",
   outputDir: "artifacts/test-output",
   fullyParallel: true,
+  workers: 4, // Playwright's default is half the cores, so a run's parallelism would depend on the machine
   forbidOnly: true,
   reporter: [["list"], ["html", { open: "never", outputFolder: "artifacts/report" }]],
   use: {

examples/plugins/scheduling/views/shifts.ejs

@@ -2,7 +2,7 @@
   Scheduling · Shifts list (reference plugin). The handler fetched the rows from the upstream
   service; this view renders them with the core building blocks inside the native app shell
   (ctx.chrome). `include()` reaches the core partials (shell, nav-tree, filter-bar, data-table,
-  alert) — see docs/plugin-contract.md. Zero-JS: search round-trips the URL.
+  alert) — see the README. Zero-JS: search round-trips the URL.
   Data: chrome, title, breadcrumbs, count, filterBar, table, canWrite, newHref, error?
 %><%
   const navHtml = include("partials/nav-tree", { nodes: chrome.nav });

public/css/styles.css

@@ -58,7 +58,7 @@
   --radius: 5px;
   --nav-w: 264px;
   --row-h: 34px;
-  --pad-x: 12px;
+  --pad-x: 12px;   /* side padding of the content pane's bars and form page, and of every table cell, so their text lines up */
   --fz: 13px;
   --fz-sm: 12px;
   --fz-xs: 11px;
@@ -336,7 +336,7 @@ span.nav-self { cursor: default; }                 /* static / non-clickable */
 .topbar {
   flex: 0 0 auto; height: 48px;
   display: flex; align-items: center; gap: 12px;
-  padding: 0 16px; border-bottom: 1px solid var(--border);
+  padding: 0 var(--pad-x); border-bottom: 1px solid var(--border);
   background: var(--surface);
 }
 .hamburger { display: none; }   /* shown only on narrow */
@@ -352,7 +352,7 @@ span.nav-self { cursor: default; }                 /* static / non-clickable */
 .filters {
   flex: 0 0 auto; background: var(--surface);
   border-bottom: 1px solid var(--border);
-  padding: 10px 16px;
+  padding: 10px var(--pad-x);
   display: flex; flex-direction: column; gap: 9px;
 }
 .filter-row {
@@ -584,7 +584,7 @@ table.table {
 .col-num { text-align: right; }
 .cell-strong { font-weight: 550; }
 .cell-muted { color: var(--text-muted); }
-.table-empty { text-align: center; color: var(--text-muted); padding: 28px 12px; }
+.table-empty { text-align: center; color: var(--text-muted); padding: 28px var(--pad-x); }
 .cell-mono { font-variant-numeric: tabular-nums;
   font-family: ui-monospace, "SF Mono", Menlo, monospace; font-size: var(--fz-sm); }
 
@@ -648,7 +648,7 @@ th[aria-sort="descending"] .sort-ico { transform: rotate(180deg); }
 .pager {
   flex: 0 0 auto;
   display: flex; align-items: center; gap: 12px;
-  padding: 8px 16px; border-top: 1px solid var(--border);
+  padding: 8px var(--pad-x); border-top: 1px solid var(--border);
   background: var(--surface); font-size: var(--fz-sm); color: var(--text-muted);
 }
 .pager .spacer { flex: 1 1 auto; }
@@ -696,7 +696,7 @@ th[aria-sort="descending"] .sort-ico { transform: rotate(180deg); }
 #nav-toggle { position: fixed; top: 0; left: 0; opacity: 0; pointer-events: none; }
 
 /* admin forms: create/edit user, account actions */
-.form-page { padding: 16px; display: flex; flex-direction: column; gap: 14px; max-width: 560px; }
+.form-page { padding: 16px var(--pad-x); display: flex; flex-direction: column; gap: 14px; max-width: 560px; }
 .form-card {
   display: flex; flex-direction: column; gap: 14px;
   padding: 18px; background: var(--surface);

src/http/app.test.ts

@@ -284,6 +284,39 @@ test("every response carries the security headers; HSTS follows SECURE_COOKIES",
   assert.match(res.headers.get("strict-transport-security") ?? "", /max-age=\d+/);
 });
 
+test("Cache-Control: pages and JSON are private, no-store unless the route sets its own; assets are left alone", async (t) => {
+  const NO_STORE = "private, no-store";
+  const csrf = (await fetch(base + "/")).headers.get("set-cookie")?.split(";")[0] ?? "";
+  assert.match(csrf, new RegExp(`^${CSRF_COOKIE}=`));
+
+  // The signed-out landing is no-store as well: it and the signed-in page share one URL.
+  assert.equal((await fetch(base + "/")).headers.get("cache-control"), NO_STORE);
+  assert.equal((await fetch(base + "/", { headers: { cookie: csrf } })).headers.get("cache-control"), NO_STORE);
+  assert.equal((await fetch(base + "/", { headers: { cookie: csrf }, method: "HEAD" })).headers.get("cache-control"), NO_STORE);
+  assert.equal((await fetch(base + "/", { headers: { cookie: `${csrf}; ${session()}` } })).headers.get("cache-control"), NO_STORE);
+
+  assert.equal((await fetch(base + "/dashboard", { headers: { cookie: session() } })).headers.get("cache-control"), NO_STORE);
+  assert.equal((await fetch(base + "/dashboard", { redirect: "manual" })).headers.get("cache-control"), NO_STORE); // sign-in bounce
+  assert.equal((await fetch(base + "/nope", { headers: { cookie: csrf } })).headers.get("cache-control"), NO_STORE); // 404
+  assert.equal((await fetch(base + "/public/css/styles.css")).headers.get("cache-control"), null);
+
+  const demo: Plugin = {
+    apiVersion: "1.0.0",
+    home: () => ({ html: "<p>plugin home</p>" }),
+    id: "demo",
+    routes: [
+      { handler: () => ({ html: "open" }), method: "GET", path: "/open", public: true },
+      { handler: () => ({ headers: { "Cache-Control": "public, max-age=600" }, html: "same" }), method: "GET", path: "/same", public: true },
+      { handler: () => ({ json: { ok: true } }), method: "GET", path: "/data", public: true },
+    ],
+  };
+  const url = await startApp(t, [demo]);
+  for (const path of ["/demo/open", "/demo/data", "/"]) {
+    assert.equal((await fetch(url + path)).headers.get("cache-control"), NO_STORE, path);
+  }
+  assert.equal((await fetch(url + "/demo/same")).headers.get("cache-control"), "public, max-age=600");
+});
+
 // Production caches compiled templates; rendering must stay correct across repeated requests.
 test("renders correctly with template caching enabled", async () => {
   const app = createApp({ cache: true });

src/http/app.ts

@@ -36,6 +36,8 @@ import { localPath } from "./safe-url.ts";
 import { routePublic, serveStatic } from "./static.ts";
 import { renderPluginView } from "../plugin-host/view-resolver.ts";
 
+const NO_STORE = "private, no-store";
+
 const rootDir = join(dirname(fileURLToPath(import.meta.url)), "..", "..");
 
 export interface AppOptions {
@@ -203,6 +205,11 @@ export function createApp(options: AppOptions = {}): Server {
       // A cache in front of us must key on the language. Set after the static branch: an asset is
       // the same bytes in every language, and a Vary there fragments its entry per raw header.
       res.setHeader("vary", "accept-language");
+      // Fail closed: a cache in front of us must not keep a page that may carry a user's data. Not
+      // even the signed-out `/` is shared: it and the signed-in page have one URL, and a cache keyed
+      // on URL and Accept-Language would hand the signed-out one to a signed-in visitor. A route that
+      // is the same for everyone sets its own value in RouteResult.headers; the host doesn't police it.
+      res.setHeader("cache-control", NO_STORE);
 
       // Canonical host (APP_URL): send an off-host visitor to the configured origin so the browser,
       // the themed forms and the cross-origin Kratos POST share one cookie host — otherwise the

todo.md

@@ -1,75 +1,200 @@
-# Primary todo
+# todo
 
-## Unfinnished work
+## Scoring
 
-- [ ] **Publish every release to Docker Hub, and give the Hub page its short description.** Today Hub's newest tag is 0.4.0 (2026-09-10), while git has v0.4.1–v0.4.12, and the overview still says 0.4.0. No `release.yml` run exists for the tags auto-release pushes as `renovate-bot`, so neither the images nor the overview reach Hub; find out why Gitea doesn't fire `push: tags` for them, fix it, and backfill the missing tags. The short description is empty. fejkdata already did this (`../fejkdata`, commits `51d924a` and `34c8234`), and two parts port directly: a `dockerhub/description.txt`-style one-liner, which here can ride the existing PATCH as its `description` field; and a unit test that the description opens a README sentence and every command the overview shows appears in the README. fejkdata also syncs the page on any push to `main` that changes it, with its own Read/Write/Delete token in an environment only `main` may use, so the image-push token needs no Delete scope. Decide whether that fits here, where `{{VERSION}}` must render a released tag.
+`Score = 100 * V / (2**S + 8)`
-- [ ] Rename the plugin "admin" to something less generic, like "auth-admin" or "users-groups-admin".
-- [ ] Guard the group paths against self-lockout, or accept them explicitly. The self-revoke guard covers only your own *direct* grants on the Users screen; unticking a permission on a group you belong to, removing yourself from that group, or deleting it can all still strip your own effective access with no warning. Recorded in AGENTS.md as a known gap — the robust fix is a "last effective holder" check, which needs a reverse Keto query.
-- [ ] The permission picker has no concurrency baseline, so two operators editing the same user/group silently discard each other's change. Sketch: post the rendered set as a hidden baseline; if it no longer matches Keto, re-render with "this changed while you had the page open" rather than applying.
-- [ ] A grant whose plugin was uninstalled is invisible and unremovable in the GUI. `grantDiff` deliberately never revokes an undeclared name, but nothing *shows* it either — so it can't be audited or cleaned, and reinstalling that plugin silently reactivates access nobody remembers granting. Sketch: a read-only "held, but no installed plugin offers this" list with a remove action.
-- [ ] A plugin may gate a route on a permission it never declares — declaring stays optional on purpose. The cost is a dead end: the picker is built from declarations only, so that route is ungrantable from the GUI with no boot error and a permanent 403 as the operator's only clue. Sketch: a discovery *warning* (not an error) naming the gated-but-undeclared permission.
-- [ ] Saving permissions gives no confirmation, and a partial failure is silent. `applyGrants` loops writes then deletes with no transaction, so a Keto error midway leaves a half-applied set behind the generic error page; and a successful save is indistinguishable from "nothing changed". The `alert alert-pos` pattern the recovery-code banner uses is already available.
-- [ ] In Playwright tests, try different resolutions and sizes, from BIG desktop down to tiny phone.
-- [ ] Decide whether `e2e-tests/` should be typechecked. It is outside `tsconfig.include`, so the gate never checks its most logic-bearing file (`console-guard.ts`). Including it needs the DOM lib and `@playwright/test` present wherever `npm run typecheck` runs, which today is the `web` image that installs neither.
-- [ ] Decide whether Playwright's `workers` should be pinned. Unset, it sizes the pool from `os.cpus()`, which reports the host's cores regardless of a container CPU quota — and with `retries: 0` a starved runner turns a slow test straight into a red gate. Fine on the current act_runner; revisit if CI ever runs constrained.
-- [ ] Record the browser floor Plainpages requires, and whether the fallback is the contract or a courtesy. The stylesheet needs `:has()` (Dec 2023); the menus need the popover API (Safari 17) and CSS anchor positioning for placement (newer still, and unguarded — the `@supports` test covers popover only). An iPadOS 16 tablet, capped at Safari 16, therefore gets panels flowing inline rather than working menus. Either state a supported floor or accept the fallback for those devices; nobody has rendered that path on real hardware.
-- [ ] Decide whether the profile dropdown still earns a dropdown. It holds one item, Sign out, behind a click, and its "Signed in as X" head repeats what the trigger already shows.
-- [ ] Trim whitespace around the verification code — a copy+pasted code from the email fails, today as a browser `pattern` refusal rather than a Kratos rejection, now carrying a digits-only hint so that refusal isn't bare. The paste still fails. **Own session.** No zero-JS fix exists: `pattern` is rejective and cannot transform, so loosening it only lets the untrimmed value through. The real fix is a host-side proxy of the flow POST, which is feasible and half-built — `submitFlow` (`src/auth/kratos-public.ts:132`) already relays cookies and normalises a 422 `redirect_browser_to`, and has no callers outside tests. CSRF improves rather than breaks: the host already writes Kratos' cookie onto its own origin (`src/auth/routes.ts:77`), so the POST becomes same-origin, and `form-action 'self'` could finally join the CSP. Two risks: one `flow.ui.action` serves all five flows, so this rewrites the sign-in path, not just code entry; and Go's nosurf checks Referer only over https, which no test here covers. **Check first:** `kratos-admin.ts:75` returns a `recovery_link` alongside the code — if the stock courier mail carries one (unverified), a template change fixes the UX with no host code. Code entry has no E2E coverage today either way.
-- [ ] Guard against the double-clicked submit, without client-side JavaScript. A non-technical user clicks a button twice when nothing happens fast enough, so a second identical POST is an expected event, not misuse — today it creates two users, mints two recovery codes, or registers two OAuth2 clients. Constraints: HTML/CSS only, and it must not break an action that is *legitimately* repeatable. Sketch: a CSS-only affordance so the second click has nothing to hit, paired with the host recognising a duplicate on the server — same session, route and payload within a short window — then logging and dropping it. Open questions: what identifies "the same submission" (a one-time token minted into each rendered form beats hashing the payload, and the CSRF plumbing already mints per-request tokens), the window length, where the record lives given the app is stateless, and how a plugin declares a route repeatable.
-- [ ] Decide the caching contract for rendered pages. Responses carry `Vary: Accept-Language` but nothing sets `Cache-Control`, so a shared cache has no instruction and a signed-in page is not marked `private`. Either set the headers deliberately (public cacheable, gated `private, no-store`) or record in AGENTS.md that the reverse proxy owns this.
-- [ ] **Cache every stylesheet and script in the browser until its content changes, so a page view downloads none of them again.** Nothing under `/public/` carries a `Cache-Control` header (`serveStatic`, `src/http/static.ts`), so every page view downloads `styles.css`, `auth.css` and each plugin's stylesheet again: about 50 kB on a plugin's issue page. On a GPRS connection that took seconds per page (measured in Firefox, 2026-10-30), and in those seconds a browser extension that forces layout early makes Firefox paint the page unstyled ("Layout was forced before the page was fully loaded"). Sketch: put a content version in each asset's URL, as a query (`?v=<hash of the file>`) or in the path, and answer a versioned URL with `Cache-Control: public, max-age=31536000, immutable`, so a release still reaches every browser at once. The shell's `<link>`s, the error pages' and the `styles` and script hrefs a plugin passes in (`views/partials/shell.ejs`) all need to come out versioned, so a plugin names its asset once and the host adds the version. Open: whether the host versions a plugin's hrefs itself or hands plugins a helper, and what an unversioned `/public/` URL answers (a short `max-age` with revalidation is the safe default). Requested by plainpages-plugin-fastjira, whose item 142 waits on the release that ships this.
-- [ ] Decide whether the single generic Keto `Resource` namespace should become per-domain namespaces (`Shift`, `Document`, …), as Ory's own examples model it. One global `Resource` bucket is the project's own "no catch-all names" rule applied to namespaces. A design question, not a naming one.
-- [ ] Decide what `ICON_NAMES` (`src/ui/icons.ts`) actually is. `i-chart`, `i-copy`, `i-download` and `i-sliders` have no caller anywhere — so either they go, or the comment should say the palette is curated and may carry an id ahead of its first use. Not cosmetic: the sprite is inlined into every page, and the rule decides whether a future removal is routine cleanup or a plugin-facing regression.
-- [ ] Decide (once) whether the CSRF token staying unbound to `sub`/session is accepted. `src/auth/csrf.ts` signs `<nonce>.<HMAC(secret, nonce)>` with no session binding, so any validly-signed token passes for any user — an attacker who can write cookies on the origin can fix a token they know. Standard for unbound signed double-submit and plausibly fine behind `SameSite=Lax` + HSTS. Accepted ⇒ record it in AGENTS.md and README → Security model; not accepted ⇒ bind the nonce to `sub`.
-- [ ] Verify the documented Docker commands on macOS and fix whatever misbehaves — **macOS is a supported dev host**, but nothing here has been run on one. Two suspects, both from the `--user "$(id -u):$(id -g)"` idiom: a macOS `id -g` is `20`, which is `dialout` inside the noble image rather than a user group, and Docker Desktop remaps bind-mount ownership in its own VM layer. The same question covers rootless Docker, where README already says to *drop* the flag.
-- [ ] Map Kratos' 401 on a self-service flow init, so an anonymous `GET /settings` with no `?flow` renders instead of 500ing. `flowPage` (`src/auth/routes.ts`) maps 403/404/410 → restart the flow, 400 `session_already_available` → `/auth/complete`, and ≥500 → the themed 503, then rethrows everything else — and Kratos answers the settings-flow init with 401 when there is no session. `/settings` is correctly `public` (the recovery flow lands there with a live Kratos session but no app JWT), so the gate is not the fix: a 401 should redirect to `/login` with the page as `return_to`. No E2E covers an anonymous hit on a flow page that needs a session.
 
-### Architectural review findings (2026-07-02)
+`Bar = 1`
 
-Prioritized. Overall verdict: architecture is sound; these are refinements.
+`Next ID = 34`
 
-- [ ] **MEDIUM — Add complexity/method-size static analysis to the CI gate.** Only `tsc --strict` today; a size/complexity rule would have caught the `app.ts` growth.
+## Items
-- [ ] **LOW — The users list offers a pencil "Edit" row action to a `users:read` holder.** The link is harmless (it opens the read-only detail page), but the label contradicts what the reader can do. Needs `canWrite` threaded into `listTable` plus a `common.view` core catalog key and an `i-eye` entry in `ICON_NAMES`.
-- [ ] **MEDIUM→LOW — Add a list-page view-model helper in `src/ui/`.** Every list screen hand-rewrites the same ~40 lines bridging `parseListQuery`/`paginate` to the EJS partials; at minimum a `buildPaginationModel(page, hrefFor)` block.
-- [ ] **LOW→MEDIUM — Retire `src/ui/shell-context.ts`.** `ShellModel`/`buildShellContext` has one consumer left (dashboard) and duplicates `PageChrome` on almost every field. Fold the dashboard onto `ctx.chrome` + title/breadcrumbs; keep `shellUser` as the shared primitive.
-- [ ] **LOW — Fix stale doc references to removed `docs/plugin-contract.md`** in `views/index.ejs` (user-visible dashboard text; also links /scheduling as if pre-installed) and `examples/plugins/scheduling/views/shifts.ejs`.
-- [ ] **LOW — Decide (once) on a `ctx.system` facade.** `@plainpages/plugin-api` exposes raw Ory client shapes, so an Ory client refactor is a major `apiVersion` bump. AGENTS.md accepts this; revisit only if external plugin authors appear.
-- [ ] **LOW — README/AGENTS.md gaps:** state the intended lifetime/horizon explicitly, add a short domain glossary (host, manifest, chrome, nav fragment, permission token, system plugin, denylist…), and note the expected plugin-author population and plugin count — per-plugin dependency isolation prices N dependency trees on disk and in RSS, and that number is what says whether the trade needs revisiting (build-time dedupe for baked images stays open).
 
-## Finnished work
+| ID | Release | Blocked by | Exempt | Item | L | E | I | R | S | Moves | Score |
+|---|---|---|---|---|---|---|---|---|---|---|---|
+| 30 | 0.4.21 |  | defect | **Give every `oklch()` colour a fallback, so primary buttons stay readable on an engine without it.** | 1 | 2 | 2 | 3 | 2 | 4 by 5 | 29.9 |
+| 29 | 0.4.21 |  | defect | **Show the whole nav tree on an engine without `:has()`, and fold it only where `:has()` works.** | 2 | 2 | 2 | 4 | 3 | 4 by 6 | 26.9 |
+| 31 | 0.4.21 |  |  | **Delete the design-preview rules in `public/css/auth.css` that no view uses.** | 1 | 1 | 1 | 1 | 2 | code health by 2, 8 by 1 | 13.3 |
+| 26 | 0.4.21 |  |  | **Add a domain glossary and the project's intended horizon to the README.** | 1 | 2 | 1 | 2 | 3 | 9 by 3, code health by 2 | 11.9 |
+| 8 | 0.4.21 |  |  | **Run the visual E2E specs at viewports from a large desktop down to a small phone.** | 2 | 1 | 1 | 2 | 4 | code health by 3 | 9.0 |
+| 9 | 0.4.21 |  |  | **Typecheck `e2e-tests/` in the gate.** | 1 | 1 | 1 | 1 | 3 | code health by 2 | 9.0 |
+| 24 | 0.4.21 |  |  | **Retire `src/ui/shell-context.ts` and render the dashboard from `ctx.chrome`.** | 2 | 1 | 2 | 3 | 4 | code health by 3 | 9.0 |
+| 17 | 0.4.21 |  |  | **Remove the icons no caller uses from `ICON_NAMES` and the sprite.** | 1 | 1 | 1 | 1 | 1 | 8 by 1, code health by 1 | 8.8 |
+| 21 | 0.4.21 |  |  | **Fail the gate on a function that exceeds a size or complexity limit.** | 3 | 1 | 1 | 3 | 5 | code health by 4 | 7.2 |
+| 32 | 0.4.21 | 29, 30 |  | **Fail the gate on a stylesheet feature the browser floor lacks, unless it is guarded.** | 3 | 1 | 1 | 3 | 4 | code health by 2 | 6.0 |
+| 27 | 0.4.21 |  |  | **Inset the scheduling example's lead text, shift count, alerts and call-to-action by `--pad-x`, like the bars around them.** | 1 | 2 | 1 | 2 | 1 | 9 by 2, 2 by 1 | 4.8 |
+| 20 | 0.4.21 |  |  | **Redirect an anonymous hit on a session-needing flow page to sign-in, instead of a 500.** | 2 | 2 | 2 | 4 | 3 | 2 by 4 | 4.2 |
+| 4 | 0.4.21 |  |  | **Refuse a permission save whose baseline no longer matches Keto, so concurrent edits are not silently lost.** | 2 | 2 | 2 | 4 | 4 | 2 by 4 | 2.8 |
+| 7 | 0.4.21 |  |  | **Confirm a successful permission save, and report a partial failure instead of hiding it.** | 2 | 2 | 2 | 4 | 4 | 2 by 4 | 2.8 |
+| 22 | 0.4.21 |  |  | **Offer a `users:read` holder a View row action on the users list, not Edit.** | 2 | 2 | 1 | 3 | 3 | 2 by 2 | 2.1 |
+| 3 | 0.4.21 |  |  | **Guard the group paths against self-lockout, or accept the gap explicitly.** | 3 | 2 | 2 | 5 | 6 | 2 by 6 | 1.4 |
+| 1 | 0.4.21 |  | owner | **Publish every release to Docker Hub, and give the Hub page its short description.** | 3 | 3 | 2 | 6 | 5 | 1 by 3 | 1.3 |
+| 13 | 0.4.21 |  |  | **Accept a verification code pasted with surrounding whitespace.** | 4 | 2 | 3 | 7 | 7 | 2 by 8 | 1.0 |
+| 19 | 0.5.0 |  | question | **Run the documented Docker commands on macOS and fix whatever misbehaves.** | 2 | 2 | 1 | 3 | 3 | 1 by 3 | 3.3 |
+| 33 | 0.5.0 |  | question | **Decide whether any public page gets a shared-cache `Cache-Control`, now that the host sends `private, no-store` on every page.** | 2 | 3 | 4 | 7 | 3 | 8 by 2 | 2.0 |
+| 16 | 0.5.0 |  | question | **Decide whether the single Keto `Resource` namespace splits into per-domain namespaces.** | 3 | 2 | 3 | 6 | 6 | 7 by 2 | 0.5 |
+| 6 | 0.5.0 |  |  | **Warn at discovery when a route or nav node gates on a permission no plugin declares.** | 2 | 1 | 1 | 2 | 2 | 7 by 3, 9 by 3 | 8.0 |
+| 23 | 0.5.0 |  |  | **Give list pages one view-model helper from query to pagination, instead of ~40 hand-written lines each.** | 3 | 1 | 2 | 4 | 5 | 1 by 4, 9 by 2, code health by 3 | 7.9 |
+| 28 | 0.5.0 |  |  | **Let a plugin add entries to the profile menu through the plugin API.** | 2 | 2 | 2 | 4 | 4 | 7 by 4, 1 by 2 | 4.2 |
+| 2 | 0.5.0 |  |  | **Rename the admin example plugin to a name that says what it administers.** | 3 | 1 | 2 | 4 | 6 | 9 by 3, code health by 2 | 2.6 |
+| 5 | 0.5.0 |  |  | **List grants that no installed plugin declares, with a remove action.** | 2 | 2 | 2 | 4 | 4 | 2 by 3 | 2.1 |
+| 14 | 0.5.0 |  |  | **Drop a double-clicked duplicate submit, with no client-side JavaScript.** | 4 | 2 | 3 | 7 | 7 | 2 by 7, 1 by 4 | 1.4 |
 
-- [x] Configure a plugin at install time: the manifest declares `settings`, the operator sets one `PLUGIN_SETTING_<ID>_<KEY>` variable per key, and the resolved values arrive on `onBoot` typed from the declaration. Read-only admin screen at `/admin/plugin-settings`.
+## Details
-- [x] Give a plugin persistent storage: `storage: true` provisions a Postgres database + login role named `plugin_<id>`, credentials arrive on `onBoot`, passwords are derived from `PLUGIN_DB_SECRET` rather than stored.
+
-- [x] Refuse a stray `package.json`/`node_modules` in `config/` by name, as plugin folders already are.
+### 30. Give every `oklch()` colour a fallback, so primary buttons stay readable on an engine without it.
-- [x] Let Renovate reach the example plugins' manifests (`ignorePaths` overrides `config:recommended`).
+
-- [x] The seeded admin is granted each permission once — `seedPermissions` dedupes and the grant PUT is idempotent.
+The accent, focus and status tokens in `public/css/styles.css` are `oklch()`, which needs Chromium
-- [x] Run the E2E runner as the invoking user so its artifacts aren't root-owned.
+111. Below it (Tizen 2024, webOS TV 24) `--accent` is invalid, so `.btn-primary` is white text on
-- [x] Install node_modules above `WORKDIR /app` so no mount leaves a root-owned dir in the checkout.
+no background: the sign-in button reads as blank.
-- [x] Enforce `<resource>:<action>` permission names at discovery; split `admin` per screen.
+
-- [x] Make the declared-permission catalog the fixed list; delete the Permissions screen, move granting onto Users and Groups.
+### 29. Show the whole nav tree on an engine without `:has()`, and fold it only where `:has()` works.
-- [x] Fail a Playwright test on any browser console warning, error or uncaught exception, in every engine.
+
-- [x] Skip the CI gate when only markdown changed.
+README → "Two browser floors": Chromium 87 must reach every page. `public/css/styles.css` hides
-- [x] Replace the `<details>` popup menus with the Popover API so an outside click dismisses them.
+`.nav-children` and only a `:has()` rule reveals them, so below Chromium 105 (webOS TV 22 and 23)
-- [x] Organize the files in src into folders.
+no nested entry can be reached. Hide the children only inside `@supports selector(:has(a))`. The
-- [x] Move docs/plugin-contract.md into README.md and remove the docs folder.
+theme switch is inert there too, since every theme rule is a `:has()`: hide it under the same guard.
-- [x] Move the scheduling example out of `plugins/` into `examples/`.
+
-- [x] Make `config/` an empty drop-in mount with the defaults as fallback.
+### 31. Delete the design-preview rules in `public/css/auth.css` that no view uses.
-- [x] Turn the built-in admin pages into a drop-in example plugin.
+
-- [x] CI/CD — test on push to any branch except main.
+`.auth-view`, the `:target` screen switch, `.tpl-*`, the `fstate`/`rstate` radios and the
-- [x] CI/CD — require a PR to main, gated on a green build, fast-forward-only.
+forgot/register alerts match nothing in `views/` or `examples/`.
-- [x] CI/CD — force-push mirror to GitHub after every merge to main.
+
-- [x] CI/CD — build and push the app image, tagged with the commit hash, as part of the gate.
+### 26. Add a domain glossary and the project's intended horizon to the README.
-- [x] CI/CD — re-tag the hash image to semver on a `vX.Y.Z` tag.
+
-- [x] CI/CD — sync released tags to Docker Hub.
+Glossary: host, manifest, chrome, nav fragment, permission token, system plugin, denylist. Also the
-- [x] Write README-dockerhub.md for the Docker Hub overview.
+expected plugin count, since per-plugin dependency isolation prices N dependency trees on disk and
-- [x] CI/CD — set up the Renovate bot.
+in RSS.
-- [x] CI/CD — give Renovate a read-only `GITHUB_COM_TOKEN` so github.com lookups aren't rate-limited.
+
-- [x] CI/CD — auto-release on Renovate updates, versioned from the `Release-Bump:` trailer.
+### 24. Retire `src/ui/shell-context.ts` and render the dashboard from `ctx.chrome`.
-- [x] Add an e2e test for the admin plugin's OAuth2-clients (Hydra) screen.
+
-- [x] Document the auth security model in the README.
+`ShellModel`/`buildShellContext` has one consumer, the dashboard, and duplicates `PageChrome` on
-- [x] Add i18n support.
+almost every field. Keep `shellUser` as the shared primitive.
-- [x] Settle the identity-vs-user vocabulary.
+
-- [x] Use one uniform verb per action in the English UI (sign in / sign out / create account).
+### 17. Remove the icons no caller uses from `ICON_NAMES` and the sprite.
-- [x] Remove the dead "Profile" link from the sidebar profile menu.
+
-- [x] Remove the unspecified "Settings"/"Preferences" cog from the sidebar footer.
+`i-chart`, `i-copy` and `i-download` have no caller, and the sprite is inlined into every page.
-- [x] **HIGH — Split `handleRequest` in `src/http/app.ts`** — extract the built-in endpoints into named handlers on an internal route table.
+AGENTS.md → "`ICON_NAMES` is a host-owned registry" already makes this routine cleanup.
+
+### 21. Fail the gate on a function that exceeds a size or complexity limit.
+
+Only `tsc --strict` runs today. A new tool is a new dependency, so justify it.
+
+### 32. Fail the gate on a stylesheet feature the browser floor lacks, unless it is guarded.
+
+Nothing checks README → "Two browser floors", and no E2E engine is old enough to show a break.
+Lint every stylesheet, the examples' included, against Chromium 87, Safari 15 and Firefox 115, and
+allow a feature only inside an `@supports` that guards it.
+
+### 27. Inset the scheduling example's lead text, shift count, alerts and call-to-action by `--pad-x`, like the bars around them.
+
+They sit flush against the content pane's edge while the filter bar, table and pager sit 12px in.
+The fix belongs in `examples/plugins/scheduling/public/scheduling.css`; check each page in a browser.
+
+### 20. Redirect an anonymous hit on a session-needing flow page to sign-in, instead of a 500.
+
+`flowPage` (`src/auth/routes.ts`) rethrows a 401, which Kratos answers to a settings-flow init with no
+session. `/settings` stays `public`, since recovery lands there with a Kratos session but no app JWT;
+a 401 should redirect to `/login` with the page as `return_to`. Add the missing E2E.
+
+### 4. Refuse a permission save whose baseline no longer matches Keto, so concurrent edits are not silently lost.
+
+Sketch: post the rendered set as a hidden baseline; when it no longer matches Keto, re-render with
+"this changed while you had the page open" rather than applying.
+
+### 7. Confirm a successful permission save, and report a partial failure instead of hiding it.
+
+`applyGrants` loops writes then deletes with no transaction, so a Keto error midway leaves a
+half-applied set behind the generic error page. The `alert alert-pos` pattern of the recovery-code
+banner is available.
+
+### 22. Offer a `users:read` holder a View row action on the users list, not Edit.
+
+The pencil opens the read-only detail page, so it is harmless, but the label contradicts what the
+reader can do. Needs `canWrite` threaded into `listTable`, a `common.view` core catalog key and an
+`i-eye` icon.
+
+### 3. Guard the group paths against self-lockout, or accept the gap explicitly.
+
+The self-revoke guard covers only your own direct grants on the Users screen. Unticking a permission
+on a group you belong to, leaving that group, or deleting it can still strip your own access with no
+warning. AGENTS.md records this as a known gap. The robust fix is a "last effective holder" check,
+which needs a reverse Keto query.
+
+### 1. Publish every release to Docker Hub, and give the Hub page its short description.
+
+Hub's newest tag is 0.4.0, while git has v0.4.1 onwards, and the overview still says 0.4.0. No
+`release.yml` run exists for the tags auto-release pushes as `renovate-bot`; find out why Gitea
+doesn't fire `push: tags` for them, fix it, and backfill the missing tags. fejkdata did this
+(`../fejkdata`, commits `51d924a` and `34c8234`), and two parts port directly: a one-line
+description riding the existing PATCH as its `description` field, and a unit test that the
+description opens a README sentence and every command the overview shows appears in the README.
+fejkdata also syncs the page on any push to `main`, with its own token in an environment only `main`
+may use; decide whether that fits here, where `{{VERSION}}` must render a released tag.
+
+### 13. Accept a verification code pasted with surrounding whitespace.
+
+A pasted code fails the browser's `pattern` check. No zero-JS fix exists: `pattern`
+cannot transform, so loosening it only lets the untrimmed value through to Kratos. **Check first:**
+`kratos-admin.ts` returns a `recovery_link` alongside the code; if the stock courier mail can carry
+one, a template change fixes this with no host code. Otherwise proxy the flow POST through the host:
+`submitFlow` (`src/auth/kratos-public.ts`) already relays cookies and normalises a 422
+`redirect_browser_to`, and has no callers outside tests. CSRF improves, since the POST becomes
+same-origin and `form-action 'self'` can join the CSP. Risks: one `flow.ui.action` serves all five
+flows, so this rewrites the sign-in path; and Go's nosurf checks Referer only over https, which no
+test covers. Code entry has no E2E coverage today.
+
+### 19. Run the documented Docker commands on macOS and fix whatever misbehaves.
+
+Needs a Mac. Two suspects, both from `--user "$(id -u):$(id -g)"`: a macOS `id -g` is `20`, which is
+`dialout` inside the noble image, and Docker Desktop remaps bind-mount ownership in its VM.
+
+### 33. Decide whether any public page gets a shared-cache `Cache-Control`, now that the host sends `private, no-store` on every page.
+
+Every non-asset response is `private, no-store`. The built-in signed-out `/` and its signed-in
+variant share one URL, so a shared cache keyed on URL and `Accept-Language` would serve the
+signed-out page to a signed-in visitor for the cache lifetime, and a cache that bypasses on cookies
+would never store it. Options: leave it, so operators cache only assets; add `Vary: Cookie`; or give
+the signed-out page its own URL. A plugin route can already opt in through `RouteResult.headers`.
+I is 4 because a wrong rule caches a signed-in page for everyone, exposing PII.
+
+### 16. Decide whether the single Keto `Resource` namespace splits into per-domain namespaces.
+
+Ory's own examples model `Shift`, `Document` and so on. One global `Resource` bucket is the
+project's "no catch-all names" rule applied to namespaces.
+
+### 6. Warn at discovery when a route or nav node gates on a permission no plugin declares.
+
+Declaring stays optional on purpose, but the picker is built from declarations only, so such a route
+is ungrantable from the GUI and answers a permanent 403. A warning, not an error.
+
+### 23. Give list pages one view-model helper from query to pagination, instead of ~40 hand-written lines each.
+
+Every list screen bridges `parseListQuery`/`paginate` to the partials by hand. At minimum a
+`buildPaginationModel(page, hrefFor)`. Exported through the barrel, so it is contract surface.
+
+### 28. Let a plugin add entries to the profile menu through the plugin API.
+
+The profile dropdown stays a dropdown: Sign out is its only entry today, and it will gain more.
+`views/partials/shell.ejs` hand-rolls it (AGENTS.md → UI, the dropdown rule), so the entries need a
+manifest field that discovery checks, gated like a nav node.
+
+### 2. Rename the admin example plugin to a name that says what it administers.
+
+Candidates: `auth-admin`, `users-groups-admin`. The id is also the mount path, so the compose files,
+E2E stacks and README move with it.
+
+### 5. List grants that no installed plugin declares, with a remove action.
+
+`grantDiff` never revokes an undeclared name, and nothing shows one either, so it can't be audited or
+cleaned, and reinstalling its plugin silently reactivates access nobody remembers granting.
+
+### 14. Drop a double-clicked duplicate submit, with no client-side JavaScript.
+
+A second identical POST is expected from a non-technical user, and today creates two
+users, mints two recovery codes, or registers two OAuth2 clients. It must not break a legitimately
+repeatable action. Sketch: a CSS-only affordance so the second click has nothing to hit, plus the
+host recognising a duplicate (same session, route and payload within a short window), logging and
+dropping it. Open: what identifies "the same submission" (a one-time token minted into each form
+beats hashing the payload), the window, where the record lives given the host is stateless, and how
+a plugin declares a route repeatable.