Skip to content

armgit / mso

Public Imported from GitHub
0

Browser-based graphical shell and control plane for a Linux server you own — real terminal, file manager, live metrics and BYOK AI in one mobile-first workspace. Next.js 16 · React 19 · node-pty · Tailwind v4 · shadcn/ui. Public alpha, self-hosted.

fix(pwa,chat): notch clearance, a doubled iOS inset, and an assistant that was lying

Rahman7c998cd12c30
diff --git a/app/api/config/route.ts b/app/api/config/route.ts
index 648ec8de..1cb7905c 100644
--- a/app/api/config/route.ts
+++ b/app/api/config/route.ts
@@ -53,6 +53,13 @@ export async function GET() {
return NextResponse.json({
provider,
+ // Whether the ACTIVE provider path can carry function calls at all. The
+ // ChatGPT Codex OAuth backend cannot — app/api/assistant/route.ts takes a
+ // chat-only branch for it and drops `tools` entirely. Until this existed the
+ // UI advertised "22 tools" and "Alfa acts on your real VPS" on that provider
+ // while the model correctly answered that it had none, which reads as the
+ // assistant being broken rather than the provider being chat-only.
+ toolsSupported: provider !== "openai-codex",
model: cfg.model || defaultModelFor(provider),
hasApiKey: key.length > 0,
apiKeyMasked: stored ? mask(stored) : "",
diff --git a/app/globals.css b/app/globals.css
index 2b80e89b..bc3a5d21 100644
--- a/app/globals.css
+++ b/app/globals.css
@@ -459,8 +459,21 @@ body {
installed PWA (display-mode: standalone), so the browser tab / desktop
PhoneFrame preview stays visually unchanged while the real app clears the bar. */
[data-shell="ios"] { --sai-top: max(env(safe-area-inset-top, 0px), 2.75rem); }
-@media (display-mode: standalone) {
- [data-shell="android"] { --sai-top: max(env(safe-area-inset-top, 0px), 1.5rem); }
+/* Android floors BOTH insets, in every display mode.
+ It used to floor only --sai-top, only at 1.5rem (24px), and only inside
+ `display-mode: standalone`. Three problems on a real phone, all reported from
+ one: 24px is a 2019 status bar — a punch-hole or curved-corner device needs
+ ~36px, so the home grid and the top of every app sat UNDER the cutout; the
+ standalone gate meant an installed-to-home-screen PWA that Chrome does not
+ report as standalone got no floor at all; and --sai-bottom had no floor, so on
+ gesture navigation the 48px nav row sat beneath the system gesture bar and its
+ Back/Home/Recents could not be tapped.
+ max() throughout, so a device that reports a LARGER inset still wins and nothing
+ is capped. The desktop PhoneFrame preview gains 36px of top chrome, which is the
+ honest thing for a preview of a device that has a status bar. */
+[data-shell="android"] {
+ --sai-top: max(env(safe-area-inset-top, 0px), 2.25rem);
+ --sai-bottom: max(env(safe-area-inset-bottom, 0px), 1rem);
}
/* OS reduced-motion: collapse durations to ~instant. 1ms (not 0) so animationend
still fires — the window close/minimize finalize on that event. Scoped narrowly
diff --git a/frontend/slices/appshell/components/message-bubble.tsx b/frontend/slices/appshell/components/message-bubble.tsx
index 32bb7a45..b13d7eab 100644
--- a/frontend/slices/appshell/components/message-bubble.tsx
+++ b/frontend/slices/appshell/components/message-bubble.tsx
@@ -2,6 +2,7 @@
import { Sparkles, User } from "lucide-react";
import { cn } from "@/lib/utils";
+import { mdToHtml } from "../features/widgets/components/md";
export type ChatRole = "user" | "assistant" | "tool";
@@ -67,7 +68,25 @@ export function MessageBubble({ message, ios }: { message: ChatMessage; ios?: bo
),
)}
>
- {message.text || <span className="text-muted-foreground">…</span>}
+ {!message.text ? (
+ <span className="text-muted-foreground">…</span>
+ ) : isUser ? (
+ message.text
+ ) : (
+ // Assistant text is MARKDOWN. The model emits **bold**, `code`, bullets
+ // and fenced blocks on nearly every turn, and rendering it literally is
+ // why the thread read as unfinished — a wall of asterisks and backticks.
+ // mdToHtml escapes first and emits a fixed tag set (see md.test.ts),
+ // which is what makes dangerouslySetInnerHTML defensible on text that is
+ // model output, i.e. untrusted.
+ // USER text stays literal on purpose: someone typing ** means **.
+ // whitespace-normal because mdToHtml already turned newlines into <br/>;
+ // leaving the parent's pre-wrap on would double every line break.
+ <span
+ className="block whitespace-normal [&_code]:font-mono [&_pre]:whitespace-pre [&_strong]:font-semibold"
+ dangerouslySetInnerHTML={{ __html: mdToHtml(message.text) }}
+ />
+ )}
</div>
</div>
);
diff --git a/frontend/slices/appshell/components/mobile-shell.tsx b/frontend/slices/appshell/components/mobile-shell.tsx
index dc607ad7..93c3cf04 100644
--- a/frontend/slices/appshell/components/mobile-shell.tsx
+++ b/frontend/slices/appshell/components/mobile-shell.tsx
@@ -239,13 +239,18 @@ export function MobileShell() {
</div>
</header>
{/* The home-indicator overlays the content edge-to-edge (real-iOS), so
- raise --sai-bottom INSIDE the app pane to include its 34px zone —
- every app already pads with var(--sai-bottom), so bumping the var
- clears the pill centrally without double-padding anyone. */}
+ --sai-bottom INSIDE the app pane must clear its 34px band — every app
+ already pads with var(--sai-bottom), so setting the var here clears the
+ pill centrally without double-padding anyone.
+ max(), NOT env() + 34px. On a notched iPhone env(safe-area-inset-bottom)
+ is ALREADY the 34px home-indicator zone, and our pill is drawn INSIDE
+ that same physical band — adding them reserved 68px and pushed every
+ app's bottom row visibly up the screen. On a device that reports 0 the
+ floor still applies, so nothing regresses where the sum used to be right. */}
<main
onScrollCapture={(e) => setAppScrolled((e.target as HTMLElement).scrollTop > 4)}
className="relative min-h-0 flex-1 overflow-auto [container-type:inline-size]"
- style={{ "--sai-bottom": "calc(env(safe-area-inset-bottom, 0px) + 34px)" } as React.CSSProperties}
+ style={{ "--sai-bottom": "max(env(safe-area-inset-bottom, 0px), 34px)" } as React.CSSProperties}
>
<WindowContent app={top.app} payload={top.payload} />
</main>
diff --git a/frontend/slices/appshell/features/widgets/components/md.test.ts b/frontend/slices/appshell/features/widgets/components/md.test.ts
index 2d651c41..c4a0a2ae 100644
--- a/frontend/slices/appshell/features/widgets/components/md.test.ts
+++ b/frontend/slices/appshell/features/widgets/components/md.test.ts
@@ -19,3 +19,32 @@ describe("mdToHtml", () => {
expect(out).toContain("&quot;onmouseover"); // it stays escaped INSIDE the value
});
});
+
+describe("fenced code blocks", () => {
+ it("renders a fence as <pre><code>, not as literal backticks", () => {
+ const out = mdToHtml("run this:\n```sh\nls -la ~\n```");
+ expect(out).toContain("<pre");
+ expect(out).toContain("ls -la ~");
+ expect(out).not.toContain("```");
+ });
+
+ it("does NOT apply inline markdown inside a fence", () => {
+ // `rm -rf **` inside a code block used to render as bold nothing, which is
+ // both wrong and dangerous-looking in a VPS assistant.
+ const out = mdToHtml("```\nrm -rf ** and `x`\n```");
+ expect(out).not.toContain("<strong>");
+ expect(out).toContain("rm -rf ** and `x`");
+ });
+
+ it("escapes inside a fence — it is model output, i.e. untrusted", () => {
+ const out = mdToHtml("```\n<img src=x onerror=alert(1)>\n```");
+ expect(out).not.toContain("<img");
+ expect(out).toContain("&lt;img");
+ });
+
+ it("still renders inline code and bold OUTSIDE a fence", () => {
+ const out = mdToHtml("**bold** and `code`\n```\nraw\n```");
+ expect(out).toContain("<strong>bold</strong>");
+ expect(out).toContain("<code");
+ });
+});
diff --git a/frontend/slices/appshell/features/widgets/components/md.ts b/frontend/slices/appshell/features/widgets/components/md.ts
index 7573b19b..8cfbf483 100644
--- a/frontend/slices/appshell/features/widgets/components/md.ts
+++ b/frontend/slices/appshell/features/widgets/components/md.ts
@@ -1,15 +1,33 @@
-// Minimal, SAFE markdown → HTML for the Markdown widget. Escape FIRST (including
-// the double-quote, so a link URL can't break out of href="…" and inject an
-// event-handler attribute), THEN apply a fixed set of transforms. Links are
-// restricted to http(s) — no javascript:/data:. The output is rendered via
-// dangerouslySetInnerHTML, so the escaping here is load-bearing (see md.test.ts).
-// ponytail: headings/bold/italic/code/links/bullets — no tables/nested lists.
+// Minimal, SAFE markdown → HTML. Escape FIRST (including the double-quote, so a
+// link URL can't break out of href="…" and inject an event-handler attribute),
+// THEN apply a fixed set of transforms. Links are restricted to http(s) — no
+// javascript:/data:. The output is rendered via dangerouslySetInnerHTML, so the
+// escaping here is load-bearing (see md.test.ts).
+//
+// Used by the Markdown widget AND by the Alfa chat bubble, which is why fenced
+// code blocks exist: an assistant reporting command output emits them on almost
+// every turn, and rendering them literally is what made the thread look unfinished.
+// ponytail: headings/bold/italic/code/fences/links/bullets — no tables, no nested
+// lists. Add one only when something actually needs it.
+
+const esc = (s: string) =>
+ s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;");
+
+// A placeholder that cannot appear in real text and survives the escape pass
+// untouched (it contains none of & < > ").
+const MARK = "\u0000F";
+
export function mdToHtml(src: string): string {
- return src
- .replace(/&/g, "&amp;")
- .replace(/</g, "&lt;")
- .replace(/>/g, "&gt;")
- .replace(/"/g, "&quot;")
+ // Fenced blocks are lifted out BEFORE anything else and restored last. The
+ // inline passes below would otherwise eat the ** and ` inside a shell snippet —
+ // `rm -rf **` in a code block would have rendered as bold nothing.
+ const fences: string[] = [];
+ const body = src.replace(/```[a-zA-Z0-9_-]*\r?\n?([\s\S]*?)```/g, (_m, code: string) => {
+ fences.push(esc(code.replace(/\n$/, "")));
+ return `${MARK}${fences.length - 1}${MARK}`;
+ });
+
+ return esc(body)
.replace(/^### (.+)$/gm, '<div class="text-sm font-semibold">$1</div>')
.replace(/^## (.+)$/gm, '<div class="text-base font-bold">$1</div>')
.replace(/^# (.+)$/gm, '<div class="text-lg font-bold">$1</div>')
@@ -18,5 +36,8 @@ export function mdToHtml(src: string): string {
.replace(/\*([^*\n]+)\*/g, "<em>$1</em>")
.replace(/`([^`\n]+)`/g, '<code class="rounded bg-black/20 px-1">$1</code>')
.replace(/\[([^\]]+)\]\((https?:\/\/[^)\s]+)\)/g, '<a href="$2" target="_blank" rel="noopener noreferrer" class="underline">$1</a>')
- .replace(/\n/g, "<br/>");
+ .replace(/\n/g, "<br/>")
+ .replace(new RegExp(`${MARK}(\\d+)${MARK}`, "g"), (_m, i: string) =>
+ `<pre class="my-1.5 overflow-x-auto rounded-md bg-black/25 p-2 text-[0.92em] leading-snug"><code>${fences[Number(i)]}</code></pre>`,
+ );
}
diff --git a/frontend/slices/assistant/components/agent-switcher.tsx b/frontend/slices/assistant/components/agent-switcher.tsx
index 43497aba..109389ca 100644
--- a/frontend/slices/assistant/components/agent-switcher.tsx
+++ b/frontend/slices/assistant/components/agent-switcher.tsx
@@ -9,6 +9,7 @@ import {
DropdownMenuTrigger,
} from "@/components/ui/dropdown-menu";
import type { AIStore } from "../lib/store";
+import { useToolsSupported } from "../lib/use-tools-supported";
import { OS_TOOLS } from "../lib/tools";
import { GlyphTile } from "./agent-avatar";
@@ -16,14 +17,19 @@ import { GlyphTile } from "./agent-avatar";
// persona the system context for the next reply; "New agent" opens the editor.
export function AgentSwitcher({ store, onNew }: { store: AIStore; onNew: () => void }) {
const a = store.activeAgent;
+ const toolsSupported = useToolsSupported();
return (
<DropdownMenu>
<DropdownMenuTrigger className="flex items-center gap-2 rounded-lg p-1 pr-2 outline-none hover:bg-accent [@media(pointer:coarse)]:min-h-[44px] [@media(pointer:coarse)]:min-w-[44px]">
<GlyphTile glyph={a.glyph} color={a.color} size={30} />
<div className="text-left leading-tight">
<div className="text-[13px] font-semibold">{a.name}</div>
+ {/* OS_TOOLS.length is what the CATALOG holds, which is not what the model
+ receives: a chat-only provider gets none of them. Printing the catalog
+ size unconditionally is how the header came to say "22 tools" next to an
+ assistant that correctly reported having zero. */}
<div className="text-[10px] text-muted-foreground">
- {OS_TOOLS.length} tools
+ {toolsSupported ? `${OS_TOOLS.length} tools` : "chat only"}
</div>
</div>
<ChevronDown className="size-3.5 text-muted-foreground" />
diff --git a/frontend/slices/assistant/components/chat-panel.tsx b/frontend/slices/assistant/components/chat-panel.tsx
index a45fddce..8100a3c5 100644
--- a/frontend/slices/assistant/components/chat-panel.tsx
+++ b/frontend/slices/assistant/components/chat-panel.tsx
@@ -12,6 +12,7 @@ import {
import { ScrollArea } from "@/components/ui/scroll-area";
import { cn } from "@/lib/utils";
import { errText } from "./error-text";
+import { useToolsSupported } from "../lib/use-tools-supported";
import {
registerAlfaRunner,
updateAlfa,
@@ -68,6 +69,7 @@ export function ChatPanel({
}) {
const api = useOsApi();
const ios = useActiveShell().id === "ios";
+ const toolsSupported = useToolsSupported();
// The display list lives in the SHARED Alfa store, not in this component. That
// is what makes the per-app Alfa sheet show this very conversation instead of a
// second, disconnected one — and it is less memory, not more: one array for
@@ -226,12 +228,22 @@ export function ChatPanel({
<div
className={cn(
"px-3 py-1 text-center text-[11px] font-medium",
- api.mode === "live" ? "bg-destructive/10 text-destructive" : "bg-muted text-muted-foreground",
+ !toolsSupported
+ ? "bg-amber-500/10 text-amber-600 dark:text-amber-400"
+ : api.mode === "live"
+ ? "bg-destructive/10 text-destructive"
+ : "bg-muted text-muted-foreground",
)}
>
- {api.mode === "live"
- ? "● LIVE — Alfa acts on your real VPS; every change needs your approval"
- : "MOCK — actions are simulated (switch to Live in Settings → Server)"}
+ {/* Three states, not two. `api.mode` only says which DATA source the tools
+ would hit — it says nothing about whether the model can call one, and on
+ a chat-only provider the old two-state banner promised a VPS the
+ assistant could not touch. */}
+ {!toolsSupported
+ ? "CHAT ONLY — this provider cannot call tools. Switch provider in Settings → AI to let Alfa act."
+ : api.mode === "live"
+ ? "● LIVE — Alfa acts on your real VPS; every change needs your approval"
+ : "MOCK — actions are simulated (switch to Live in Settings → Server)"}
</div>
{messages.length === 0 ? (
<div className="flex-1 overflow-y-auto">
diff --git a/frontend/slices/assistant/lib/use-tools-supported.ts b/frontend/slices/assistant/lib/use-tools-supported.ts
new file mode 100644
index 00000000..6cd5fb6f
--- /dev/null
+++ b/frontend/slices/assistant/lib/use-tools-supported.ts
@@ -0,0 +1,47 @@
+"use client";
+
+import { useEffect, useState } from "react";
+import { IS_DEMO } from "@/lib/demo";
+
+// Can the ACTIVE provider actually call tools?
+//
+// Not a cosmetic question. app/api/assistant/route.ts takes a chat-only branch for
+// the ChatGPT Codex OAuth backend and drops the `tools` array entirely, so on that
+// provider every tool the UI advertises is unreachable. Before this, the header
+// claimed "22 tools" and the banner claimed "Alfa acts on your real VPS" while the
+// model answered, correctly, that it had none — which reads as a broken assistant
+// rather than a chat-only provider.
+//
+// Optimistic default: true. A provider that CAN use tools is the common case, and
+// flashing a "chat only" warning for one frame on every open would be its own lie.
+// The GET is cached at module scope because several surfaces ask (chat panel
+// banner, agent switcher badge) and the answer changes only when the owner picks a
+// different provider in Settings.
+
+let cached: Promise<boolean> | null = null;
+
+function fetchSupported(): Promise<boolean> {
+ cached ??= fetch("/api/config", { cache: "no-store" })
+ .then((r) => (r.ok ? r.json() : null))
+ .then((c: { toolsSupported?: boolean } | null) => c?.toolsSupported !== false)
+ .catch(() => true); // unreachable config is not evidence of a chat-only provider
+ return cached;
+}
+
+/** Call after changing the provider so the next read re-asks. */
+export function invalidateToolsSupported(): void {
+ cached = null;
+}
+
+export function useToolsSupported(): boolean {
+ const [supported, setSupported] = useState(true);
+ useEffect(() => {
+ if (IS_DEMO) return; // demo makes no /api calls; its tools are simulated anyway
+ let alive = true;
+ void fetchSupported().then((v) => alive && setSupported(v));
+ return () => {
+ alive = false;
+ };
+ }, []);
+ return supported;
+}