Add worktrees, per-hunk review, effort per message, regenerate, voice input

- Isolated git worktree per session (create, status, merge, discard)
- Permission card: accept file-edit hunks individually; agent learns what was rejected
- Reasoning effort per message; regenerate the last answer with another model/effort
- Voice input via the Web Speech API
- Folder-up button in the file browser; ssh, rsync and mc in the image

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

web/src/components/Voice.tsx

@@ -0,0 +1,68 @@
+import { useEffect, useRef, useState } from "react";
+import { getLang, useT } from "../i18n";
+import { toast } from "../dialogs";
+import { Icon } from "./Icon";
+
+/* Web Speech API (Chrome/Edge/Safari; Firefox has none). */
+type Recognition = {
+  lang: string;
+  continuous: boolean;
+  interimResults: boolean;
+  start(): void;
+  stop(): void;
+  onresult: ((e: any) => void) | null;
+  onend: (() => void) | null;
+  onerror: ((e: any) => void) | null;
+};
+const Impl: (new () => Recognition) | undefined = (window as any).SpeechRecognition ?? (window as any).webkitSpeechRecognition;
+
+export const voiceSupported = () => !!Impl;
+
+const LOCALES: Record<string, string> = { de: "de-DE", en: "en-US" };
+
+/** Microphone button: dictated text is appended to the message. */
+export function VoiceButton({ onText, onInterim }: { onText: (text: string) => void; onInterim: (text: string) => void }) {
+  const t = useT();
+  const [listening, setListening] = useState(false);
+  const rec = useRef<Recognition | null>(null);
+
+  useEffect(() => () => rec.current?.stop(), []);
+  if (!Impl) return null;
+
+  const toggle = () => {
+    if (listening) {
+      rec.current?.stop();
+      return;
+    }
+    const r = new Impl!();
+    r.lang = LOCALES[getLang()] ?? navigator.language;
+    r.continuous = true;
+    r.interimResults = true;
+    r.onresult = (e: any) => {
+      let interim = "";
+      for (let i = e.resultIndex; i < e.results.length; i++) {
+        const res = e.results[i];
+        if (res.isFinal) onText(res[0].transcript.trim());
+        else interim += res[0].transcript;
+      }
+      onInterim(interim);
+    };
+    r.onerror = (e: any) => {
+      if (e.error === "not-allowed" || e.error === "service-not-allowed") toast(t("voice.denied"));
+      else if (e.error !== "no-speech" && e.error !== "aborted") toast(`${t("voice.error")}: ${e.error}`);
+    };
+    r.onend = () => {
+      setListening(false);
+      onInterim("");
+    };
+    rec.current = r;
+    r.start();
+    setListening(true);
+  };
+
+  return (
+    <button type="button" className={`ghost small voice-btn ${listening ? "on" : ""}`} onClick={toggle} title={t(listening ? "voice.stop" : "voice.start")} aria-label={t(listening ? "voice.stop" : "voice.start")} aria-pressed={listening}>
+      <Icon name="mic" />
+    </button>
+  );
+}