Add NerdFont for symbols and version bump script

Add version bump script to synchronize GUI and daemon versions

Cargo.lock

@@ -869,7 +869,7 @@ dependencies = [
 
 [[package]]
 name = "spacesh-cli"
-version = "0.1.0"
+version = "0.1.3"
 dependencies = [
  "anyhow",
  "clap",
@@ -881,16 +881,17 @@ dependencies = [
 
 [[package]]
 name = "spacesh-core"
-version = "0.1.0"
+version = "0.1.3"
 dependencies = [
  "alacritty_terminal",
  "serde",
+ "serde_json",
  "spacesh-proto",
 ]
 
 [[package]]
 name = "spacesh-proto"
-version = "0.1.0"
+version = "0.1.3"
 dependencies = [
  "bytes",
  "serde",
@@ -902,7 +903,7 @@ dependencies = [
 
 [[package]]
 name = "spacesh-pty"
-version = "0.1.0"
+version = "0.1.3"
 dependencies = [
  "anyhow",
  "bytes",
@@ -912,7 +913,7 @@ dependencies = [
 
 [[package]]
 name = "spaceshd"
-version = "0.1.0"
+version = "0.1.3"
 dependencies = [
  "anyhow",
  "base64",

Cargo.toml

@@ -10,7 +10,7 @@ members = [
 
 [workspace.package]
 edition = "2021"
-version = "0.1.0"
+version = "0.1.3"
 
 [workspace.dependencies]
 tokio = { version = "1", features = ["full"] }

DOCS/superpowers/specs/2026-06-15-session-persistence-design.md

@@ -0,0 +1,216 @@
+# Session Persistence (resurrect + resume) — Design
+
+**Date:** 2026-06-15
+**Status:** Approved, ready for implementation plan
+
+## Goal
+
+Let a workspace survive both GUI loss and full power loss. Closing a tab or
+the whole GUI already keeps agents running (daemon owns PTYs, reattach via live
+grid snapshot — M1). This design adds the missing half: after the daemon itself
+dies (reboot, battery death, `kill -9`), the user can bring panels back —
+panels show their last on-screen state and offer a one-click **Resume** that
+restarts the agent with its session-continue flag (e.g. `claude --continue`).
+
+## Scope decisions (locked)
+
+- **Reboot behavior:** resurrect + resume. A live process cannot survive a
+  power-off — not even tmux does that. After a daemon restart we respawn the
+  panel from its persisted spec (cwd intact) and, for agents that support it,
+  relaunch with a resume flag so the conversation continues in a *new* process.
+- **Resurrect trigger:** manual, per-panel. After a daemon restart panels are
+  shown stopped with their last screen; nothing spawns until the user clicks.
+  This avoids surprise token burn from auto-launching many agents.
+- **Persisted scrollback:** visible screen only. We reuse the existing
+  `snapshot_ansi()` serializer (the same one that powers live reattach) and
+  write its output to disk. No scrollback history beyond the visible grid.
+- **Resume command source:** a `[resume]` table in `~/.spacesh/config.toml`
+  mapping a command basename to resume args, merged over built-in defaults.
+- **Snapshot cadence:** periodic + shutdown. A background task dumps changed
+  grids every N seconds (default 5), plus a full pass on graceful shutdown and
+  a final dump when an actor exits. This survives `kill -9` / battery (you lose
+  at most N seconds of the last screen).
+
+## What already exists (do not rebuild)
+
+- `state.json` (via `JsonStateStore` + debounced `Persister`) already persists
+  structure: groups, workspaces, layout, zoom, pinned, and per-surface
+  `SurfaceSpec` (`command`, `args`, `cwd`, `cols`, `rows`, `agent_label`,
+  `autostart`). On cold start `Registry::restore()` loads this; the `live`
+  actor map is empty, so every surface is "stopped" (spec present, no process).
+- `SurfaceView.running: bool` already tells the client a surface is stopped.
+- `spacesh_core::snapshot::snapshot_ansi(&GridSurface) -> Snapshot` serializes
+  the visible grid to an ANSI dump (`ansi`, `cols`, `rows`, `cursor_row`,
+  `cursor_col`). `Snapshot` currently derives `Serialize` only.
+- The surface actor already answers `SurfaceMsg::AttachSnapshot` by calling
+  `snapshot_ansi(&grid)`; the grid is the authoritative screen model.
+
+## Components
+
+### 1. Snapshot store — `crates/spaceshd/src/snapshot_store.rs` (new)
+
+Per-surface JSON file `~/.spacesh/snapshots/<surface_id>.json` holding the
+serialized visible-screen snapshot. Atomic write (temp file → `sync_all` →
+rename), mirroring `state_store::JsonStateStore`.
+
+```rust
+pub trait SnapshotStore: Send + Sync {
+    fn save(&self, sid: &SurfaceId, snap: &Snapshot) -> anyhow::Result<()>;
+    fn load(&self, sid: &SurfaceId) -> Option<Snapshot>;
+    fn remove(&self, sid: &SurfaceId);
+}
+```
+
+The store persists the core `spacesh_core::snapshot::Snapshot` directly
+(`ansi`, `cols`, `rows`, `cursor_row`, `cursor_col`) — `spaceshd` already
+depends on `spacesh-core`, so no separate daemon record type is introduced. A
+corrupt/missing file yields `None` (never an error that blocks resurrect).
+`remove` deletes the file and is called when a surface is closed or removed
+from the tree.
+
+### 2. On-demand snapshot from the actor — `crates/spaceshd/src/surface.rs`
+
+Add a message that returns the current snapshot without subscribing:
+
+```rust
+SurfaceMsg::Snapshot { reply: oneshot::Sender<(Snapshot, bool)> } // (snapshot, dirty)
+```
+
+The actor tracks a `dirty` flag: set inside `flush` whenever bytes are fed into
+the grid (`grid.feed`), cleared when a `Snapshot` reply is produced. The bool
+lets the periodic dumper skip unchanged grids.
+
+On actor exit (after `pty.wait()`), the actor takes a final `snapshot_ansi`
+and forwards `(id, snapshot)` to the writer channel (a cloned
+`mpsc::UnboundedSender<(SurfaceId, Snapshot)>` passed into the actor), so the
+last screen of a finished process is persisted even between ticker ticks.
+
+### 3. Writer task + periodic ticker — `crates/spaceshd/src/server.rs` / `main.rs`
+
+- **Writer task:** the sole owner of `Arc<dyn SnapshotStore>`. Receives
+  `(SurfaceId, Snapshot)` on an unbounded channel and writes to disk. Keeps all
+  snapshot disk I/O off the actor/PTY hot path and serializes writes.
+- **Periodic ticker:** every `snapshot_interval_secs` (config, default 5) the
+  router iterates live surface handles, sends `SurfaceMsg::Snapshot`, awaits the
+  reply, and forwards to the writer channel only when `dirty` is true.
+- **Graceful shutdown:** before the daemon exits it does one final synchronous
+  pass over all live surfaces into the writer, then flushes the writer.
+
+### 4. Resume config — `crates/spaceshd/src/config.rs`
+
+```toml
+[resume]
+commands = { claude = ["--continue"], codex = ["resume"] }
+```
+
+```rust
+#[derive(Debug, Clone, Default, Deserialize, Serialize)]
+pub struct ResumeConfig {
+    #[serde(default)]
+    pub commands: std::collections::HashMap<String, Vec<String>>,
+}
+```
+
+Added to `Config` as `#[serde(default)] pub resume: ResumeConfig`. A method
+`resume_args(command: &str) -> Option<Vec<String>>` resolves by command
+basename: user map first, then a built-in default table
+(`claude → ["--continue"]`, `codex → ["resume"]`), then `None`. The default
+table is a `const`/static, not inline literals in branching logic.
+
+### 5. Protocol — `crates/spacesh-proto/src/message.rs`
+
+The codebase already has `Cmd::RestartSurface { surface_id }` (starts a stopped
+surface from its spec, guarded by `is_running`) and an `Attach` response that
+already carries `{ snapshot, cols, rows, cursor_row, cursor_col, stopped }`.
+So no new command or wire type is needed beyond one field:
+
+- Extend `Cmd::RestartSurface` with `#[serde(default)] resume: bool`. `resume =
+  true` builds `command + resume_args(command)` (falling back to the original
+  args when no mapping exists); `resume = false` keeps the original
+  `command + args` (today's behavior). The `#[serde(default)]` keeps old frames
+  decoding to `resume = false`.
+- No `GetSnapshot`, no `StartSurface`, no `SnapshotView`: a stopped-panel
+  `Attach` returns the **disk** snapshot (see §6) using the existing response
+  shape.
+
+`spacesh-core::snapshot::Snapshot` gains `Deserialize` (alongside `Serialize`)
+so the store can load it back from disk.
+
+### 6. Server handlers — `crates/spaceshd/src/server.rs`
+
+- `RestartSurface { surface_id, resume }`: unchanged flow (spec lookup,
+  `spawn_from_spec`, `set_live`, `SurfaceRestarted` broadcast). When `resume`,
+  spawn with a spec whose `args` are replaced by `config.resume_args(command)`
+  (when present); otherwise spawn the original spec.
+- `Attach` for a **stopped** surface: instead of returning the empty
+  `{ snapshot: "", stopped: true }`, load the disk snapshot via the snapshot
+  store and return `{ snapshot: <ansi>, cols, rows, cursor_row, cursor_col,
+  stopped: true }`. Missing file → empty snapshot, still `stopped: true`.
+- Surface close/remove (`Close`, `CloseWorkspace`, `remove_surface` paths):
+  send a remove to the snapshot writer so stale `<sid>.json` files do not
+  accumulate.
+
+### 7. App — `app/src` and `app/src-tauri`
+
+- `socketBridge.ts`: `restartSurface(id, resume = false)` gains the `resume`
+  arg; `AttachResult` gains optional `cursor_row`/`cursor_col`/`stopped`.
+- `app/src-tauri/src/bridge.rs`: `restart_surface` forwards a `resume: bool`
+  arg into `Cmd::RestartSurface`.
+- `LayoutEngine.tsx` stopped branch (`running[id] === false`): paint the disk
+  snapshot into a dimmed, read-only `xterm` behind the controls, and offer two
+  buttons — **Resume** → `restartSurface(id, true)` and **Restart fresh** →
+  `restartSurface(id, false)`. On success the daemon's `workspace_changed`
+  flips `running` to true, the overlay unmounts, and the live `TerminalView`
+  mounts.
+
+## Data flow
+
+```
+running surface ──(every 5s, if dirty)──▶ ticker ──▶ writer task ──▶ <sid>.json
+running surface ──(on exit)─────────────────────────▶ writer task ──▶ <sid>.json
+daemon shutdown ──(final pass over live)────────────▶ writer task ──▶ <sid>.json
+
+reboot ▶ daemon cold start ▶ Registry::restore(state.json) ▶ all surfaces stopped
+client ▶ Attach(sid) [stopped] ▶ disk snapshot ▶ paint dimmed read-only screen
+user clicks Resume ▶ RestartSurface{resume:true} ▶ spawn(command + resume_args, cwd)
+                  ▶ SurfaceRestarted + running=true ▶ live TerminalView mounts
+```
+
+## Error handling
+
+- Missing/corrupt snapshot file → stopped `Attach` returns an empty snapshot;
+  the overlay shows an empty dimmed panel with the Resume/Restart controls.
+- `RestartSurface` on an already-running surface → no-op ok (existing
+  `is_running` guard); unknown surface → `NOT_FOUND`.
+- Resume command for an agent without a mapping → falls back to the original
+  spec args (plain restart), never fails the spawn.
+- Writer task failure to write one file is logged and dropped; it must not stall
+  the daemon or other surfaces.
+
+## Performance
+
+- A visible-screen snapshot is ≈ rows × cols bytes of ANSI; at a 5s cadence with
+  the `dirty` debounce, idle panels write nothing. All disk writes happen in the
+  single writer task, off the PTY/actor hot path, so the keypress→echo (<16 ms)
+  and output-batching budgets are untouched.
+
+## Testing
+
+- **snapshot_store:** save→load round-trip; atomic write; missing file → `None`;
+  corrupt file → `None`; `remove` deletes the file.
+- **config:** parse `[resume]` table; `resume_args` returns user override, then
+  built-in default, then `None`; missing section defaults cleanly.
+- **surface actor:** `SurfaceMsg::Snapshot` returns the current grid contents;
+  `dirty` is true after output and false immediately after a snapshot.
+- **server:** `RestartSurface{resume:true}` spawns with `command + resume_args`;
+  `{resume:false}` spawns with `command + args`; stopped `Attach` returns the
+  saved disk snapshot; `is_running` guard prevents a second actor.
+- **registry:** starting a stopped surface re-populates the live map and the
+  view flips `running` to true.
+
+## Out of scope
+
+- Resuming the literal in-flight process across power loss (impossible).
+- Scrollback history beyond the visible screen.
+- Auto-resume on daemon start (manual trigger chosen).
+- Per-surface resume command stored in the spec/wizard (config map chosen).

Makefile

@@ -43,8 +43,8 @@ targets: ## add rust targets for the universal build
 	rustup target add aarch64-apple-darwin x86_64-apple-darwin
 
 .PHONY: bump
-bump: ## increment the patch version in tauri.conf.json (single source of truth)
-	@node -e "const f='$(APP_DIR)/src-tauri/tauri.conf.json';const fs=require('fs');const j=JSON.parse(fs.readFileSync(f));const p=j.version.split('.').map(Number);p[2]=(p[2]||0)+1;j.version=p.join('.');fs.writeFileSync(f, JSON.stringify(j,null,2)+'\n');console.log('version → '+j.version)"
+bump: ## increment the patch version for BOTH the GUI (tauri.conf.json) and the daemon (workspace Cargo.toml)
+	@node scripts/bump_version.mjs
 
 .PHONY: dmg
 dmg: bump targets ## bump version + build the universal (Intel + Apple Silicon) .dmg — UNSIGNED

app/src-tauri/Cargo.lock

@@ -540,6 +540,17 @@ dependencies = [
  "libc",
 ]
 
+[[package]]
+name = "core-text"
+version = "22.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "333dab512ce710ca2d08574c373d246dbeac8b22769e47da4c0e72730ce442b7"
+dependencies = [
+ "core-foundation",
+ "core-graphics",
+ "foreign-types",
+]
+
 [[package]]
 name = "cpufeatures"
 version = "0.2.17"
@@ -3413,6 +3424,8 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "base64 0.22.1",
+ "core-foundation",
+ "core-text",
  "dirs 5.0.1",
  "reqwest 0.12.28",
  "serde",
@@ -3427,7 +3440,7 @@ dependencies = [
 
 [[package]]
 name = "spacesh-proto"
-version = "0.1.0"
+version = "0.1.3"
 dependencies = [
  "bytes",
  "serde",

app/src-tauri/Cargo.toml

@@ -25,3 +25,5 @@ anyhow = "1"
 dirs = "5"
 # rustls (no openssl) so the universal-apple-darwin cross-build stays self-contained.
 reqwest = { version = "0.12", default-features = false, features = ["rustls-tls", "json"] }
+core-text = "22"
+core-foundation = "0.10"

app/src-tauri/src/bridge.rs

@@ -78,6 +78,13 @@ fn find_daemon() -> PathBuf {
     PathBuf::from("spaceshd") // last resort: rely on PATH
 }
 
+/// The installed `spaceshd` binary's mtime as ms since the epoch (for staleness check).
+fn daemon_bin_mtime_ms() -> Option<u64> {
+    let meta = std::fs::metadata(find_daemon()).ok()?;
+    let mtime = meta.modified().ok()?;
+    Some(mtime.duration_since(std::time::UNIX_EPOCH).ok()?.as_millis() as u64)
+}
+
 async fn ensure_daemon(sock: &PathBuf) -> Result<UnixStream> {
     if let Ok(s) = UnixStream::connect(sock).await {
         return Ok(s);
@@ -120,7 +127,7 @@ impl Bridge {
         let pending: Arc<Mutex<HashMap<u64, oneshot::Sender<Envelope>>>> = Arc::default();
         let out_channels: Arc<Mutex<HashMap<String, Channel<Vec<u8>>>>> = Arc::default();
         let (tx, reader) = spawn_connection(&sock, &app, pending.clone(), out_channels.clone()).await?;
-        Ok(Self {
+        let bridge = Self {
             next_id: AtomicU64::new(1),
             app,
             sock,
@@ -130,7 +137,48 @@ impl Bridge {
             reader: Mutex::new(reader),
             pending,
             out_channels,
-        })
+        };
+        // The daemon outlives the GUI by design, so after an update the GUI may
+        // attach to a stale daemon — new features that need new daemon code then
+        // silently don't work. Restart it if it's out of date.
+        bridge.ensure_matching_daemon().await;
+        Ok(bridge)
+    }
+
+    /// Restart the running daemon if it predates the installed `spaceshd` binary
+    /// (or was built from a different commit). The bundled binary's mtime vs the
+    /// daemon's `started_at_ms` is the reliable signal: it catches every reinstall
+    /// even while developing dirty, where the git build id doesn't change.
+    async fn ensure_matching_daemon(&self) {
+        let Ok(reply) = self.request(Cmd::Health).await else { return };
+        let (daemon_build, started_at_ms) = match &reply {
+            Envelope::Res { data, .. } => (
+                data.get("build").and_then(|v| v.as_str()).map(str::to_string),
+                data.get("started_at_ms").and_then(|v| v.as_u64()),
+            ),
+            _ => (None, None),
+        };
+        let gui_build = option_env!("SPACESH_BUILD").unwrap_or("dev");
+        let build_mismatch = gui_build != "dev"
+            && daemon_build.as_deref().map(|b| b != gui_build).unwrap_or(false);
+        let binary_newer = match (daemon_bin_mtime_ms(), started_at_ms) {
+            (Some(bin_ms), Some(start_ms)) => bin_ms > start_ms,
+            _ => false,
+        };
+        if !build_mismatch && !binary_newer {
+            return;
+        }
+        // Ask the stale daemon to exit, wait for its socket to clear, then reconnect
+        // — which lazily spawns the fresh bundled daemon.
+        self.fire(Cmd::Shutdown).await;
+        for _ in 0..100 {
+            if UnixStream::connect(&self.sock).await.is_err() {
+                break;
+            }
+            tokio::time::sleep(Duration::from_millis(30)).await;
+        }
+        let seen = self.gen.load(Ordering::Acquire);
+        let _ = self.reconnect(seen).await;
     }
 
     /// Send a command without awaiting a reply or retrying. Used for Shutdown:
@@ -348,8 +396,8 @@ pub async fn apply_preset(state: BridgeState<'_>, workspace_id: String, preset_i
 }
 
 #[tauri::command]
-pub async fn restart_surface(state: BridgeState<'_>, surface_id: String) -> Result<Value, String> {
-    data_of(state.request(Cmd::RestartSurface { surface_id: SurfaceId(surface_id) }).await.map_err(|e| e.to_string())?)
+pub async fn restart_surface(state: BridgeState<'_>, surface_id: String, resume: bool) -> Result<Value, String> {
+    data_of(state.request(Cmd::RestartSurface { surface_id: SurfaceId(surface_id), resume }).await.map_err(|e| e.to_string())?)
 }
 
 #[tauri::command]
@@ -422,6 +470,11 @@ pub async fn health(state: BridgeState<'_>) -> Result<Value, String> {
     data_of(state.request(Cmd::Health).await.map_err(|e| e.to_string())?)
 }
 
+#[tauri::command]
+pub async fn which_agents(state: BridgeState<'_>, candidates: Vec<String>) -> Result<Value, String> {
+    data_of(state.request(Cmd::WhichAgents { candidates }).await.map_err(|e| e.to_string())?)
+}
+
 // ---- Update check ----
 
 /// Where the GUI looks for the published app version. Overridable via
@@ -481,6 +534,25 @@ pub fn open_external(url: String) -> Result<(), String> {
     Ok(())
 }
 
+/// List the user's installed font families (CoreText) so Settings can offer any of
+/// them for the terminal. Hidden system families (".SF NS" etc.) are dropped; the
+/// result is de-duplicated and sorted case-insensitively.
+#[tauri::command]
+pub fn list_fonts() -> Vec<String> {
+    use std::collections::BTreeSet;
+    let names = core_text::font_collection::get_family_names();
+    let mut set: BTreeSet<String> = BTreeSet::new();
+    for name in names.iter() {
+        let s = name.to_string();
+        if !s.is_empty() && !s.starts_with('.') {
+            set.insert(s);
+        }
+    }
+    let mut v: Vec<String> = set.into_iter().collect();
+    v.sort_by_key(|s| s.to_lowercase());
+    v
+}
+
 // ---- Settings commands ----
 
 #[tauri::command]

app/src-tauri/src/lib.rs

@@ -54,8 +54,10 @@ pub fn run() {
             bridge::mark_read,
             bridge::clear_events,
             bridge::health,
+            bridge::which_agents,
             bridge::check_update,
             bridge::open_external,
+            bridge::list_fonts,
             bridge::get_config,
             bridge::set_config,
             bridge::shutdown_daemon,

app/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "spacesh",
-  "version": "0.1.0",
+  "version": "0.1.3",
   "identifier": "xyz.spacesh.app",
   "build": {
     "frontendDist": "../dist",
@@ -10,8 +10,16 @@
     "beforeBuildCommand": "npm run build"
   },
   "app": {
-    "windows": [{ "title": "spacesh", "width": 1100, "height": 720 }],
-    "security": { "csp": null }
+    "windows": [
+      {
+        "title": "spacesh",
+        "width": 1100,
+        "height": 720
+      }
+    ],
+    "security": {
+      "csp": null
+    }
   },
   "bundle": {
     "active": true,

app/src/App.tsx

@@ -183,7 +183,7 @@ export function App() {
         <Sidebar railMode={!sidebarOpen} groups={groups} workspaces={workspaces} activeId={activeId} onSelect={selectWorkspace} onNew={() => setWizard(true)} onDelete={setDeleteTarget} health={health} connected={connected} />
         <div style={{ flex: 1, display: "flex", flexDirection: "column", minWidth: 0 }}>
           {active && (
-            <CenterToolbar selected="" onSelect={(p) => { if (active) void applyPreset(active.id, p, []); }} onOpenSearch={() => { if (effectiveFocus) { setSearchSurfaceId(effectiveFocus); setSearchNonce((n) => n + 1); } }} />
+            <CenterToolbar selected="" paneCount={leaves.length} onSelect={(p) => { if (active) void applyPreset(active.id, p, []); }} onOpenSearch={() => { if (effectiveFocus) { setSearchSurfaceId(effectiveFocus); setSearchNonce((n) => n + 1); } }} />
           )}
           <div style={{ flex: 1, minHeight: 0, position: "relative" }}>
             {active

app/src/CenterToolbar.tsx

@@ -3,10 +3,10 @@ import { COLORS, FONT } from "./theme";
 import { PresetPicker } from "./PresetPicker";
 
 /** Top-of-grid toolbar: layout presets on the left, scrollback search on the right (search is a mock). */
-export function CenterToolbar({ selected, onSelect, onOpenSearch }: { selected: string; onSelect: (id: string) => void; onOpenSearch: () => void }) {
+export function CenterToolbar({ selected, onSelect, onOpenSearch, paneCount }: { selected: string; onSelect: (id: string) => void; onOpenSearch: () => void; paneCount: number }) {
   return (
     <div style={{ display: "flex", alignItems: "center", gap: 8, padding: "0 12px", height: 46, borderBottom: `1px solid ${COLORS.borderSubtle}` }}>
-      <PresetPicker selected={selected} onSelect={onSelect} />
+      <PresetPicker selected={selected} onSelect={onSelect} minSlots={paneCount} />
       <div style={{ flex: 1 }} />
       <div
         title="Search scrollback"

app/src/LayoutEngine.tsx

@@ -1,11 +1,12 @@
 import { useEffect, useRef, useState } from "react";
-import { Maximize2, Minimize2, RotateCw, GripVertical } from "lucide-react";
+import { Maximize2, Minimize2, RotateCw, GripVertical, Play } from "lucide-react";
+import { Terminal } from "@xterm/xterm";
 import { TerminalView } from "./TerminalView";
 import { SearchBar } from "./SearchBar";
 import { StatusRing } from "./StatusRing";
 import { COLORS, FONT, STATE_COLOR } from "./theme";
 import type { LayoutNode, SurfaceState, SurfaceView } from "./layoutTypes";
-import { setRatios, restartSurface, setZoom, moveSurface } from "./socketBridge";
+import { setRatios, restartSurface, setZoom, moveSurface, attachSurface, detachSurface } from "./socketBridge";
 
 interface Props {
   workspaceId: string;
@@ -116,6 +117,43 @@ function Node({ node, path, ...rest }: NodeProps) {
   return <SplitView split={node.split} path={path} {...rest} />;
 }
 
+const NERD_FALLBACK_LE = "'Symbols Nerd Font Mono'";
+const fontStackLE = (family: string | null) =>
+  family ? `'${family}', ${NERD_FALLBACK_LE}, monospace`
+         : `'JetBrains Mono Variable', 'JetBrains Mono', ${NERD_FALLBACK_LE}, monospace`;
+
+function xtermThemeLE(p: Record<string, string>) {
+  return {
+    background: p["bg-panel"],
+    foreground: p["text-primary"],
+    cursor: p["text-primary"],
+    selectionBackground: p["search-match"],
+  };
+}
+
+function StoppedSnapshot({ surfaceId, font, palette }: { surfaceId: string; font: { family: string; size: number } | null; palette: Record<string, string> | null }) {
+  const hostRef = useRef<HTMLDivElement | null>(null);
+  useEffect(() => {
+    const host = hostRef.current;
+    if (!host) return;
+    const term = new Terminal({
+      fontFamily: fontStackLE(font?.family ?? null),
+      fontSize: font?.size ?? 13,
+      theme: palette ? xtermThemeLE(palette) : undefined,
+      cursorBlink: false,
+      disableStdin: true,
+      scrollback: 0,
+    });
+    term.open(host);
+    let disposed = false;
+    void attachSurface(surfaceId, () => {}).then((res) => {
+      if (!disposed && res.snapshot) term.write(res.snapshot);
+    });
+    return () => { disposed = true; term.dispose(); void detachSurface(surfaceId); };
+  }, [surfaceId, font, palette]); // eslint-disable-line react-hooks/exhaustive-deps
+  return <div ref={hostRef} style={{ position: "absolute", inset: 0, opacity: 0.45, pointerEvents: "none" }} />;
+}
+
 function Leaf({ id, workspaceId, running, states, surfaces, focusedId, onFocus, zoomed, drop, onStartPanelDrag, searchSurfaceId, searchNonce, onCloseSearch, font, palette }: Omit<NodeProps, "node" | "path"> & { id: string }) {
   const focused = focusedId === id;
   const dropEdge = drop && drop.id === id ? drop.edge : null;
@@ -142,19 +180,26 @@ function Leaf({ id, workspaceId, running, states, surfaces, focusedId, onFocus,
 
   if (running[id] === false) {
     return card(
-      <div style={{ display: "flex", alignItems: "center", justifyContent: "center", height: "100%", width: "100%", color: COLORS.textSecondary, flexDirection: "column", gap: 10 }}>
-        <div style={{ fontFamily: FONT.mono, fontSize: 13 }}>Process exited</div>
-        <div style={{ display: "flex", gap: 8 }}>
-          <button onClick={() => void restartSurface(id)}
-            style={{ display: "flex", alignItems: "center", gap: 6, padding: "6px 14px", background: COLORS.bgElevated, color: COLORS.textPrimary, border: `1px solid ${COLORS.borderStrong}`, borderRadius: 7, fontSize: 12 }}>
-            <RotateCw size={13} /> Restart
-          </button>
-          {zoomed === id && (
-            <button onClick={() => void setZoom(workspaceId, null)}
-              style={{ display: "flex", alignItems: "center", gap: 6, padding: "6px 14px", background: "transparent", color: COLORS.textSecondary, border: `1px solid ${COLORS.borderStrong}`, borderRadius: 7, fontSize: 12 }}>
-              <Minimize2 size={13} /> Exit zoom
+      <div style={{ position: "relative", height: "100%", width: "100%" }}>
+        <StoppedSnapshot surfaceId={id} font={font} palette={palette} />
+        <div style={{ position: "absolute", inset: 0, display: "flex", alignItems: "center", justifyContent: "center", flexDirection: "column", gap: 10, color: COLORS.textSecondary, background: "rgba(0,0,0,0.35)" }}>
+          <div style={{ fontFamily: FONT.mono, fontSize: 13 }}>Stopped</div>
+          <div style={{ display: "flex", gap: 8 }}>
+            <button onClick={() => void restartSurface(id, true)}
+              style={{ display: "flex", alignItems: "center", gap: 6, padding: "6px 14px", background: COLORS.accent, color: COLORS.bgApp, border: "none", borderRadius: 7, fontSize: 12, fontWeight: 600 }}>
+              <Play size={13} /> Resume
             </button>
-          )}
+            <button onClick={() => void restartSurface(id, false)}
+              style={{ display: "flex", alignItems: "center", gap: 6, padding: "6px 14px", background: COLORS.bgElevated, color: COLORS.textPrimary, border: `1px solid ${COLORS.borderStrong}`, borderRadius: 7, fontSize: 12 }}>
+              <RotateCw size={13} /> Restart fresh
+            </button>
+            {zoomed === id && (
+              <button onClick={() => void setZoom(workspaceId, null)}
+                style={{ display: "flex", alignItems: "center", gap: 6, padding: "6px 14px", background: "transparent", color: COLORS.textSecondary, border: `1px solid ${COLORS.borderStrong}`, borderRadius: 7, fontSize: 12 }}>
+                <Minimize2 size={13} /> Exit zoom
+              </button>
+            )}
+          </div>
         </div>
       </div>
     );

app/src/PresetPicker.tsx

@@ -13,10 +13,12 @@ export const PRESETS: { id: string; label: string; slots: number }[] = [
 
 import { COLORS, FONT } from "./theme";
 
-export function PresetPicker({ selected, onSelect }: { selected: string; onSelect: (id: string) => void }) {
+// `minSlots` hides presets smaller than the current pane count — applying a preset
+// only ever ADDS panes (never destroys running ones); shrink by closing panels.
+export function PresetPicker({ selected, onSelect, minSlots = 0 }: { selected: string; onSelect: (id: string) => void; minSlots?: number }) {
   return (
     <div style={{ display: "flex", gap: 4, flexWrap: "wrap" }}>
-      {PRESETS.map((p) => {
+      {PRESETS.filter((p) => p.slots >= minSlots).map((p) => {
         const on = p.id === selected;
         return (
           <button key={p.id} onClick={() => onSelect(p.id)}

app/src/Settings.tsx

@@ -1,10 +1,74 @@
-import { useEffect, useRef, useState } from "react";
-import { X } from "lucide-react";
+import { useEffect, useMemo, useRef, useState } from "react";
+import { X, Search, Check } from "lucide-react";
 import { COLORS, FONT, ACCENTS } from "./theme";
-import { setConfig, restartDaemon } from "./socketBridge";
+import { setConfig, restartDaemon, listFonts } from "./socketBridge";
 import type { ConfigView, DaemonHealth } from "./socketBridge";
 
-const FONTS = ["JetBrains Mono", "Menlo", "Monaco", "SF Mono", "Fira Code", "Cascadia Code"];
+// Pinned defaults shown first; the rest are the user's installed families (list_fonts).
+const DEFAULT_FONTS = ["JetBrains Mono", "Menlo", "Monaco", "SF Mono", "Fira Code", "Cascadia Code"];
+
+/** Searchable font picker: type to filter, click to apply. Defaults pinned on top. */
+function FontPicker({ value, onPick }: { value: string; onPick: (family: string) => void }) {
+  const [installed, setInstalled] = useState<string[]>([]);
+  const [query, setQuery] = useState("");
+  const [open, setOpen] = useState(false);
+  const boxRef = useRef<HTMLDivElement>(null);
+
+  useEffect(() => { void listFonts().then(setInstalled).catch(() => {}); }, []);
+
+  // Close on outside click.
+  useEffect(() => {
+    if (!open) return;
+    const onDown = (e: MouseEvent) => { if (boxRef.current && !boxRef.current.contains(e.target as Node)) setOpen(false); };
+    document.addEventListener("mousedown", onDown);
+    return () => document.removeEventListener("mousedown", onDown);
+  }, [open]);
+
+  const options = useMemo(() => {
+    const seen = new Set<string>();
+    const merged: string[] = [];
+    for (const f of [...DEFAULT_FONTS, ...installed]) {
+      const k = f.toLowerCase();
+      if (!seen.has(k)) { seen.add(k); merged.push(f); }
+    }
+    const q = query.trim().toLowerCase();
+    return q ? merged.filter((f) => f.toLowerCase().includes(q)) : merged;
+  }, [installed, query]);
+
+  return (
+    <div ref={boxRef} style={{ position: "relative", marginBottom: 10 }}>
+      <div style={{ position: "relative" }}>
+        <Search size={14} style={{ position: "absolute", left: 9, top: "50%", transform: "translateY(-50%)", color: COLORS.textMuted }} />
+        <input
+          value={open ? query : value}
+          placeholder={value}
+          onFocus={() => { setOpen(true); setQuery(""); }}
+          onChange={(e) => { setQuery(e.target.value); setOpen(true); }}
+          style={{ width: "100%", padding: "8px 8px 8px 30px", background: COLORS.bgPanel, color: COLORS.textPrimary, border: `1px solid ${open ? COLORS.accent : COLORS.borderStrong}`, borderRadius: 8, fontFamily: FONT.ui }}
+        />
+      </div>
+      {open && (
+        <div style={{ position: "absolute", top: "calc(100% + 4px)", left: 0, right: 0, zIndex: 10, maxHeight: 240, overflowY: "auto",
+          background: COLORS.bgPanel, border: `1px solid ${COLORS.borderStrong}`, borderRadius: 8, boxShadow: "0 8px 24px rgba(0,0,0,0.4)" }}>
+          {options.length === 0 && <div style={{ padding: 10, fontSize: 12, color: COLORS.textMuted }}>Ничего не найдено</div>}
+          {options.map((f) => {
+            const isDefault = DEFAULT_FONTS.some((d) => d.toLowerCase() === f.toLowerCase());
+            return (
+              <button key={f} onClick={() => { onPick(f); setOpen(false); }}
+                style={{ display: "flex", alignItems: "center", gap: 8, width: "100%", textAlign: "left", padding: "7px 10px",
+                  background: f === value ? COLORS.bgElevated : "transparent", border: "none", color: COLORS.textPrimary,
+                  fontFamily: `'${f}', ${FONT.mono}`, fontSize: 13 }}>
+                <Check size={13} style={{ opacity: f === value ? 1 : 0, color: COLORS.accent, flex: "0 0 auto" }} />
+                <span style={{ flex: 1, whiteSpace: "nowrap", overflow: "hidden", textOverflow: "ellipsis" }}>{f}</span>
+                {isDefault && <span style={{ fontFamily: FONT.ui, fontSize: 10, color: COLORS.textMuted }}>default</span>}
+              </button>
+            );
+          })}
+        </div>
+      )}
+    </div>
+  );
+}
 
 export function Settings({ config, health, onClose, onReload }: { config: ConfigView; health: DaemonHealth | null; onClose: () => void; onReload: () => void }) {
   const ref = useRef<HTMLDivElement>(null);
@@ -30,10 +94,7 @@ export function Settings({ config, health, onClose, onReload }: { config: Config
         </div>
 
         <div style={{ fontSize: 12, color: COLORS.textSecondary, marginBottom: 6 }}>Terminal font</div>
-        <select value={config.font_family} onChange={(e) => void setConfig({ font_family: e.target.value })}
-          style={{ width: "100%", padding: 8, marginBottom: 10, background: COLORS.bgPanel, color: COLORS.textPrimary, border: `1px solid ${COLORS.borderStrong}`, borderRadius: 8 }}>
-          {FONTS.map((f) => <option key={f} value={f}>{f}</option>)}
-        </select>
+        <FontPicker value={config.font_family} onPick={(f) => void setConfig({ font_family: f })} />
         <div style={{ display: "flex", alignItems: "center", gap: 10, marginBottom: 18 }}>
           <span style={{ fontSize: 12, color: COLORS.textSecondary }}>Size {sizeLocal}</span>
           <input type="range" min={10} max={20} value={sizeLocal}

app/src/TerminalView.tsx

@@ -9,6 +9,13 @@ import { registerSearch, unregisterSearch } from "./searchRegistry";
 const decoder = new TextDecoder();
 const encoder = new TextEncoder();
 
+// Appended after the user font so Nerd Font icon glyphs (Private Use Area) render
+// via fallback instead of blank boxes, without changing the base monospace font.
+const NERD_FALLBACK = "'Symbols Nerd Font Mono'";
+const fontStack = (family: string | null) =>
+  family ? `'${family}', ${NERD_FALLBACK}, monospace`
+         : `'JetBrains Mono Variable', 'JetBrains Mono', ${NERD_FALLBACK}, monospace`;
+
 function xtermTheme(p: Record<string, string>) {
   return {
     background: p["bg-panel"],
@@ -30,7 +37,7 @@ export function TerminalView({ surfaceId, font, palette }: { surfaceId: string;
     // call registerMarker/registerDecoration (proposed API). Without it findNext
     // throws and the scrollback search counter never updates.
     const term = new Terminal({
-      fontFamily: font ? `'${font.family}', monospace` : "'JetBrains Mono Variable', 'JetBrains Mono', monospace",
+      fontFamily: fontStack(font?.family ?? null),
       fontSize: font?.size ?? 13,
       convertEol: false,
       scrollback: 10000,
@@ -81,6 +88,12 @@ export function TerminalView({ surfaceId, font, palette }: { surfaceId: string;
 
     let disposed = false;
 
+    // The Nerd Font fallback may finish loading after the first paint; once it's
+    // ready, drop the WebGL glyph atlas so cached blank cells re-rasterize with icons.
+    void document.fonts.load("16px 'Symbols Nerd Font Mono'").then(() => {
+      if (!disposed) webglRef.current?.clearTextureAtlas();
+    }).catch(() => {});
+
     // Attach: fresh xterm instance, write snapshot, then stream live output.
     void attachSurface(surfaceId, (bytes) => {
       if (!disposed) term.write(decoder.decode(bytes));
@@ -112,7 +125,7 @@ export function TerminalView({ surfaceId, font, palette }: { surfaceId: string;
     const t = termRef.current;
     if (!t) return;
     if (font) {
-      t.options.fontFamily = `'${font.family}', monospace`;
+      t.options.fontFamily = fontStack(font.family);
       t.options.fontSize = font.size;
       // The WebGL renderer caches rasterized glyphs in a texture atlas keyed by
       // the old font/size; without clearing it the grid keeps rendering stale

app/src/Wizard.tsx

@@ -1,16 +1,22 @@
 import { useEffect, useRef, useState } from "react";
 import { PresetPicker, PRESETS } from "./PresetPicker";
-import { openWorkspace, applyPreset } from "./socketBridge";
+import { openWorkspace, applyPreset, whichAgents } from "./socketBridge";
+
+// Agents we know about; only the installed ones are offered (probed via whichAgents).
+const KNOWN_AGENTS = ["claude", "codex", "gemini"];
+const CUSTOM = "custom…";
 
 export function Wizard({ onDone, onCancel }: { onDone: (workspaceId: string) => void; onCancel: () => void }) {
   const [path, setPath] = useState(".");
   const [preset, setPreset] = useState("2x2");
   const [agents, setAgents] = useState<string[]>([]);
+  const [customCmds, setCustomCmds] = useState<string[]>([]);
+  const [installed, setInstalled] = useState<string[]>([]);
   const [busy, setBusy] = useState(false);
   const [error, setError] = useState<string | null>(null);
   const pathRef = useRef<HTMLInputElement>(null);
   const slots = PRESETS.find((p) => p.id === preset)?.slots ?? 1;
-  const agentChoices = ["shell", "claude", "codex", "gemini"];
+  const agentChoices = ["shell", ...installed, CUSTOM];
 
   // Grab focus on open — otherwise keystrokes leak to the xterm panel behind us
   // (its helper textarea sits at z-index 1000 and keeps the live focus).
@@ -19,6 +25,9 @@ export function Wizard({ onDone, onCancel }: { onDone: (workspaceId: string) =>
     pathRef.current?.select();
   }, []);
 
+  // Only offer agents the user actually has installed.
+  useEffect(() => { void whichAgents(KNOWN_AGENTS).then(setInstalled).catch(() => {}); }, []);
+
   async function create() {
     if (busy) return;
     setBusy(true);
@@ -27,7 +36,12 @@ export function Wizard({ onDone, onCancel }: { onDone: (workspaceId: string) =>
       const ws = await openWorkspace(path);
       const slotSpecs = Array.from({ length: slots }, (_, i) => {
         const a = agents[i] ?? "shell";
-        return a === "shell" ? {} : { command: a };
+        if (a === "shell") return {};
+        if (a === CUSTOM) {
+          const parts = (customCmds[i] ?? "").trim().split(/\s+/).filter(Boolean);
+          return parts.length ? { command: parts[0], args: parts.slice(1) } : {};
+        }
+        return { command: a };
       });
       await applyPreset(ws, preset, slotSpecs);
       onDone(ws);
@@ -61,12 +75,22 @@ export function Wizard({ onDone, onCancel }: { onDone: (workspaceId: string) =>
         <div style={{ margin: "8px 0 16px" }}><PresetPicker selected={preset} onSelect={setPreset} /></div>
         <label style={{ fontSize: 12, color: "#8B97A6" }}>Agents</label>
         <div style={{ display: "grid", gridTemplateColumns: "1fr 1fr", gap: 8, margin: "8px 0 20px" }}>
-          {Array.from({ length: slots }, (_, i) => (
-            <select key={i} value={agents[i] ?? "shell"} onChange={(e) => setAgents((a) => { const n = [...a]; n[i] = e.target.value; return n; })}
-              style={{ padding: 8, background: "#1A2029", color: "#E6EDF3", border: "1px solid #323C49", borderRadius: 6 }}>
-              {agentChoices.map((c) => <option key={c} value={c}>{c}</option>)}
-            </select>
-          ))}
+          {Array.from({ length: slots }, (_, i) => {
+            const val = agents[i] ?? "shell";
+            return (
+              <div key={i} style={{ display: "flex", flexDirection: "column", gap: 6 }}>
+                <select value={val} onChange={(e) => setAgents((a) => { const n = [...a]; n[i] = e.target.value; return n; })}
+                  style={{ padding: 8, background: "#1A2029", color: "#E6EDF3", border: "1px solid #323C49", borderRadius: 6 }}>
+                  {agentChoices.map((c) => <option key={c} value={c}>{c}</option>)}
+                </select>
+                {val === CUSTOM && (
+                  <input value={customCmds[i] ?? ""} placeholder="e.g. npm run dev"
+                    onChange={(e) => setCustomCmds((c) => { const n = [...c]; n[i] = e.target.value; return n; })}
+                    style={{ padding: 8, background: "#0A0D12", color: "#E6EDF3", border: "1px solid #4C8DFF", borderRadius: 6, fontFamily: "monospace", fontSize: 12 }} />
+                )}
+              </div>
+            );
+          })}
         </div>
         {error && <div style={{ margin: "0 0 14px", padding: "8px 10px", background: "#3A1418", border: "1px solid #6B2230", borderRadius: 8, fontSize: 12, color: "#FF9AA6" }}>{error}</div>}
         <div style={{ display: "flex", justifyContent: "flex-end", gap: 10 }}>

app/src/socketBridge.ts

@@ -42,6 +42,9 @@ export interface AttachResult {
   snapshot: string;
   cols: number;
   rows: number;
+  cursor_row?: number;
+  cursor_col?: number;
+  stopped?: boolean;
 }
 
 export async function attachSurface(
@@ -147,8 +150,8 @@ export async function applyPreset(workspaceId: string, presetId: string, slots:
   return data.surface_ids;
 }
 
-export async function restartSurface(surfaceId: string): Promise<void> {
-  await invoke("restart_surface", { surfaceId });
+export async function restartSurface(surfaceId: string, resume = false): Promise<void> {
+  await invoke("restart_surface", { surfaceId, resume });
 }
 
 export async function closeWorkspaceCmd(workspaceId: string): Promise<void> {
@@ -199,6 +202,16 @@ export async function openExternal(url: string): Promise<void> {
   await invoke("open_external", { url });
 }
 
+export async function listFonts(): Promise<string[]> {
+  return await invoke<string[]>("list_fonts");
+}
+
+/** Which of the given CLI candidates are actually installed on the daemon's spawn PATH. */
+export async function whichAgents(candidates: string[]): Promise<string[]> {
+  const data = await invoke<{ available: string[] }>("which_agents", { candidates });
+  return data.available;
+}
+
 export async function setZoom(workspaceId: string, surfaceId: string | null): Promise<void> {
   await invoke("set_zoom", { workspaceId, surfaceId });
 }

app/src/styles.css

@@ -1,3 +1,12 @@
+/* Nerd Font symbols (icons, powerline, devicons) used as a fallback in the
+   terminal so glyphs in the Private Use Area render instead of blank boxes.
+   Base monospace font is untouched; this only fills missing glyphs. */
+@font-face {
+  font-family: "Symbols Nerd Font Mono";
+  src: url("./assets/SymbolsNerdFontMono-Regular.ttf") format("truetype");
+  font-display: swap;
+}
+
 :root {
   color-scheme: dark;
 }

crates/spacesh-cli/src/cli.rs

@@ -38,7 +38,11 @@ pub enum Sub {
     },
     Close { surface_id: String },
     Focus { surface_id: String },
-    Restart { surface_id: String },
+    Restart {
+        surface_id: String,
+        /// Relaunch the agent with its session-continue flag (e.g. claude --continue).
+        #[arg(long)] resume: bool,
+    },
     Notify {
         #[arg(long)] surface: String,
         #[arg(long, value_enum)] state: StateArg,

crates/spacesh-cli/src/mapping.rs

@@ -29,7 +29,7 @@ pub fn to_cmd(sub: Sub) -> Cmd {
         },
         Sub::Close { surface_id } => Cmd::Close { surface_id: SurfaceId(surface_id) },
         Sub::Focus { surface_id } => Cmd::Focus { surface_id: SurfaceId(surface_id) },
-        Sub::Restart { surface_id } => Cmd::RestartSurface { surface_id: SurfaceId(surface_id) },
+        Sub::Restart { surface_id, resume } => Cmd::RestartSurface { surface_id: SurfaceId(surface_id), resume },
         Sub::Notify { surface, state } => Cmd::SetState { surface_id: SurfaceId(surface), state: state_of(state) },
         Sub::ApplyPreset { workspace_id, preset, agents } => Cmd::ApplyPreset {
             workspace_id: WorkspaceId(workspace_id),

crates/spacesh-core/Cargo.toml

@@ -7,3 +7,6 @@ version.workspace = true
 alacritty_terminal.workspace = true
 serde.workspace = true
 spacesh-proto = { path = "../spacesh-proto" }
+
+[dev-dependencies]
+serde_json.workspace = true

crates/spacesh-core/src/grid.rs

@@ -1,4 +1,6 @@
-use alacritty_terminal::event::VoidListener;
+use std::sync::{Arc, Mutex};
+
+use alacritty_terminal::event::{Event, EventListener};
 use alacritty_terminal::grid::Dimensions;
 use alacritty_terminal::index::{Column, Line, Point};
 use alacritty_terminal::term::{Config, Term};
@@ -23,24 +25,55 @@ impl Dimensions for GridSize {
     }
 }
 
+/// Collects the escape sequences the terminal model wants written back to the PTY
+/// (Primary/Secondary Device Attributes, DSR cursor/status reports, etc.). Programs
+/// like fish block on these replies at startup; with a void listener they hang ~2s
+/// and then warn ("could not read response to Primary Device Attribute query").
+#[derive(Clone, Default)]
+pub struct ReplyCollector {
+    buf: Arc<Mutex<Vec<u8>>>,
+}
+
+impl EventListener for ReplyCollector {
+    fn send_event(&self, event: Event) {
+        if let Event::PtyWrite(text) = event {
+            if let Ok(mut b) = self.buf.lock() {
+                b.extend_from_slice(text.as_bytes());
+            }
+        }
+    }
+}
+
 /// Owns an alacritty terminal model and feeds raw PTY bytes into it.
 pub struct GridSurface {
-    term: Term<VoidListener>,
+    term: Term<ReplyCollector>,
     parser: Processor,
     size: GridSize,
+    replies: ReplyCollector,
 }
 
 impl GridSurface {
     pub fn new(cols: u16, rows: u16) -> Self {
         let size = GridSize { cols: cols as usize, lines: rows as usize };
-        let term = Term::new(Config::default(), &size, VoidListener);
-        Self { term, parser: Processor::new(), size }
+        let replies = ReplyCollector::default();
+        let term = Term::new(Config::default(), &size, replies.clone());
+        Self { term, parser: Processor::new(), size, replies }
     }
 
     pub fn feed(&mut self, bytes: &[u8]) {
         self.parser.advance(&mut self.term, bytes);
     }
 
+    /// Drain any escape sequences the model produced in response to queries fed so
+    /// far. The caller must write these back to the PTY for query-driven programs
+    /// (fish, vim, etc.) to proceed without timing out.
+    pub fn take_replies(&mut self) -> Vec<u8> {
+        match self.replies.buf.lock() {
+            Ok(mut b) => std::mem::take(&mut *b),
+            Err(_) => Vec::new(),
+        }
+    }
+
     pub fn resize(&mut self, cols: u16, rows: u16) {
         self.size = GridSize { cols: cols as usize, lines: rows as usize };
         self.term.resize(self.size);
@@ -56,7 +89,7 @@ impl GridSurface {
         self.term.grid()[point].c
     }
 
-    pub fn term(&self) -> &Term<VoidListener> {
+    pub fn term(&self) -> &Term<ReplyCollector> {
         &self.term
     }
 
@@ -97,4 +130,15 @@ mod tests {
         assert_eq!(g.char_at(0, 0), 'a');
         assert_eq!(g.char_at(1, 0), 'c');
     }
+
+    #[test]
+    fn primary_device_attribute_query_gets_a_reply() {
+        // fish (and friends) send ESC[c at startup and block on the response.
+        let mut g = GridSurface::new(20, 5);
+        g.feed(b"\x1b[c");
+        let reply = g.take_replies();
+        assert!(reply.starts_with(b"\x1b[?"), "expected a DA1 reply, got {reply:?}");
+        // Replies are drained, not duplicated.
+        assert!(g.take_replies().is_empty());
+    }
 }

crates/spacesh-core/src/snapshot.rs

@@ -1,11 +1,11 @@
-use serde::Serialize;
+use serde::{Deserialize, Serialize};
 use alacritty_terminal::index::Point;
 use alacritty_terminal::term::cell::Flags;
 use alacritty_terminal::vte::ansi::Color;
 use crate::grid::GridSurface;
 
 /// Serializable snapshot returned by `attach`.
-#[derive(Debug, Clone, Serialize)]
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
 pub struct Snapshot {
     /// ANSI byte dump suitable for `xterm.write()`.
     pub ansi: String,
@@ -120,6 +120,18 @@ mod tests {
         assert_eq!(a.rows, 3);
     }
 
+    #[test]
+    fn snapshot_round_trips_through_json() {
+        let mut g = GridSurface::new(20, 4);
+        g.feed(b"hello");
+        let snap = snapshot_ansi(&g);
+        let json = serde_json::to_string(&snap).unwrap();
+        let back: Snapshot = serde_json::from_str(&json).unwrap();
+        assert_eq!(back.ansi, snap.ansi);
+        assert_eq!((back.cols, back.rows), (snap.cols, snap.rows));
+        assert_eq!((back.cursor_row, back.cursor_col), (snap.cursor_row, snap.cursor_col));
+    }
+
     #[test]
     fn cursor_is_one_based_after_input() {
         let mut g = GridSurface::new(10, 3);

crates/spacesh-proto/src/message.rs

@@ -92,7 +92,11 @@ pub enum Cmd {
     SetRatios { workspace_id: WorkspaceId, node_path: Vec<u32>, ratios: Vec<f32> },
     MoveSurface { surface_id: SurfaceId, target_surface_id: SurfaceId, edge: Edge },
     ApplyPreset { workspace_id: WorkspaceId, preset_id: String, slots: Vec<PresetSlot> },
-    RestartSurface { surface_id: SurfaceId },
+    RestartSurface {
+        surface_id: SurfaceId,
+        #[serde(default)]
+        resume: bool,
+    },
     CloseWorkspace { workspace_id: WorkspaceId },
     SetWorkspaceMeta {
         workspace_id: WorkspaceId,
@@ -131,6 +135,8 @@ pub enum Cmd {
         surface_id: Option<SurfaceId>,
     },
     Health,
+    /// Which of the given CLI candidates are actually installed on the spawn PATH.
+    WhichAgents { candidates: Vec<String> },
     Status,
     Shutdown,
     GetConfig,
@@ -366,6 +372,21 @@ mod tests {
         assert_eq!(back, env);
     }
 
+    #[test]
+    fn restart_surface_resume_defaults_false_and_round_trips() {
+        // Legacy frame without `resume` decodes to false.
+        let legacy = r#"{"kind":"req","id":5,"cmd":{"cmd":"restart_surface","args":{"surface_id":"s_1"}}}"#;
+        let env: Envelope = serde_json::from_str(legacy).unwrap();
+        match env {
+            Envelope::Req { cmd: Cmd::RestartSurface { resume, .. }, .. } => assert!(!resume),
+            _ => panic!("wrong variant"),
+        }
+        // resume=true round-trips.
+        let e = Envelope::Req { id: 6, cmd: Cmd::RestartSurface { surface_id: SurfaceId("s_1".into()), resume: true } };
+        let back: Envelope = serde_json::from_str(&serde_json::to_string(&e).unwrap()).unwrap();
+        assert_eq!(back, e);
+    }
+
     #[test]
     fn event_log_cmd_no_limit_round_trips() {
         let env = Envelope::Req { id: 9, cmd: Cmd::EventLog { limit: None } };

crates/spaceshd/src/config.rs

@@ -22,6 +22,20 @@ pub struct AppearanceConfig {
     pub accent: Option<String>,
 }
 
+/// Built-in resume args for known agents, used when config has no override.
+/// (command basename, resume args)
+const DEFAULT_RESUME: &[(&str, &[&str])] = &[
+    ("claude", &["--continue"]),
+    ("codex", &["resume"]),
+];
+
+#[derive(Debug, Clone, Default, Deserialize, Serialize)]
+pub struct ResumeConfig {
+    /// command basename -> args that continue its previous session.
+    #[serde(default)]
+    pub commands: std::collections::HashMap<String, Vec<String>>,
+}
+
 #[derive(Debug, Clone, Default, Deserialize, Serialize)]
 pub struct Config {
     /// Shell launched for plain (no-command) panels. When unset, the daemon
@@ -32,6 +46,11 @@ pub struct Config {
     pub terminal: TerminalConfig,
     #[serde(default)]
     pub appearance: AppearanceConfig,
+    #[serde(default)]
+    pub resume: ResumeConfig,
+    /// How often (seconds) the daemon dumps changed grids to disk.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub snapshot_interval_secs: Option<u64>,
 }
 
 impl Config {
@@ -85,6 +104,25 @@ impl Config {
             .map_err(|e| std::io::Error::new(std::io::ErrorKind::Other, e.to_string()))?;
         std::fs::write(path, s)
     }
+
+    /// Resume args for a command, by basename: user map → built-in default → None.
+    pub fn resume_args(&self, command: &str) -> Option<Vec<String>> {
+        let base = std::path::Path::new(command)
+            .file_name()
+            .map(|s| s.to_string_lossy().to_string())
+            .unwrap_or_else(|| command.to_string());
+        if let Some(args) = self.resume.commands.get(&base) {
+            return Some(args.clone());
+        }
+        DEFAULT_RESUME.iter()
+            .find(|(name, _)| *name == base)
+            .map(|(_, args)| args.iter().map(|s| s.to_string()).collect())
+    }
+
+    /// Snapshot dump cadence in seconds (config → default 5, clamped to [1, 3600]).
+    pub fn snapshot_interval_secs(&self) -> u64 {
+        self.snapshot_interval_secs.unwrap_or(5).clamp(1, 3600)
+    }
 }
 
 /// Resolve the shell to spawn for a plain panel.
@@ -110,6 +148,98 @@ pub fn default_shell() -> String {
     "/bin/sh".into()
 }
 
+/// The user's full PATH for spawning panels: their login-shell PATH (which sources
+/// `.zprofile`/`.zshrc`), merged with the daemon's current PATH and common install
+/// dirs. Cached for the daemon's lifetime.
+///
+/// Why: when the GUI launches the daemon (Finder/launchd), the inherited PATH is
+/// minimal (`/usr/bin:/bin:…`), so agents like `claude`, `codex`, `gemini` — installed
+/// in `~/.local/bin`, npm-global, or Homebrew — aren't found and the panel exits
+/// immediately with "Process exited". A bare `/bin/zsh` still works, which is why
+/// shells launched fine but agents didn't.
+pub fn enriched_path() -> String {
+    use std::collections::HashSet;
+    use std::sync::OnceLock;
+    static CACHE: OnceLock<String> = OnceLock::new();
+    CACHE
+        .get_or_init(|| {
+            let mut dirs: Vec<String> = Vec::new();
+            let mut seen: HashSet<String> = HashSet::new();
+            let mut merge = |src: &str| {
+                for d in src.split(':') {
+                    if !d.is_empty() && seen.insert(d.to_string()) {
+                        dirs.push(d.to_string());
+                    }
+                }
+            };
+            if let Some(p) = login_shell_path() {
+                merge(&p);
+            }
+            if let Ok(p) = std::env::var("PATH") {
+                merge(&p);
+            }
+            merge(&fallback_path_dirs());
+            dirs.join(":")
+        })
+        .clone()
+}
+
+/// Whether `cmd` is an executable resolvable on the spawn PATH (or an existing path
+/// if it contains a slash). Used to only offer agents the user actually has installed.
+pub fn is_installed(cmd: &str) -> bool {
+    use std::os::unix::fs::PermissionsExt;
+    if cmd.is_empty() {
+        return false;
+    }
+    let is_exec = |p: &std::path::Path| {
+        p.metadata().map(|m| m.is_file() && m.permissions().mode() & 0o111 != 0).unwrap_or(false)
+    };
+    if cmd.contains('/') {
+        return is_exec(std::path::Path::new(cmd));
+    }
+    enriched_path().split(':').any(|dir| !dir.is_empty() && is_exec(&std::path::Path::new(dir).join(cmd)))
+}
+
+/// Common locations user-installed CLIs land in, as a colon-joined fallback.
+fn fallback_path_dirs() -> String {
+    let mut v = vec![
+        "/opt/homebrew/bin".to_string(),
+        "/usr/local/bin".to_string(),
+        "/usr/bin".to_string(),
+        "/bin".to_string(),
+        "/usr/sbin".to_string(),
+        "/sbin".to_string(),
+    ];
+    if let Ok(home) = std::env::var("HOME") {
+        if !home.is_empty() {
+            for d in [".local/bin", ".npm-global/bin", ".cargo/bin", ".bun/bin", ".deno/bin", ".volta/bin", "go/bin"] {
+                v.push(format!("{home}/{d}"));
+            }
+        }
+    }
+    v.join(":")
+}
+
+/// Capture PATH from the user's login+interactive shell so rc-file PATH edits apply.
+/// Parses `env` output (the exported PATH is colon-joined regardless of shell, so this
+/// works for fish too, where `$PATH` would otherwise print space-separated).
+#[cfg(unix)]
+fn login_shell_path() -> Option<String> {
+    let shell = login_shell().or_else(|| std::env::var("SHELL").ok())?;
+    let out = std::process::Command::new(&shell)
+        .args(["-lic", "env"])
+        .output()
+        .ok()?;
+    String::from_utf8_lossy(&out.stdout)
+        .lines()
+        .find_map(|l| l.strip_prefix("PATH="))
+        .map(|p| p.trim().to_string())
+        .filter(|p| !p.is_empty())
+}
+
+#[cfg(not(unix))]
+fn login_shell_path() -> Option<String> { None }
+
 /// The current user's login shell from the passwd database (`getpwuid`).
 #[cfg(unix)]
 fn login_shell() -> Option<String> {
@@ -206,4 +336,38 @@ mod tests {
         assert_eq!(back.appearance.accent.as_deref(), Some("purple"));
         let _ = std::fs::remove_file(&path);
     }
+
+    #[test]
+    fn resume_args_user_then_default_then_none() {
+        let mut c = Config::default();
+        // built-in defaults present without any config
+        assert_eq!(c.resume_args("claude").as_deref(), Some(&["--continue".to_string()][..]));
+        assert_eq!(c.resume_args("codex").as_deref(), Some(&["resume".to_string()][..]));
+        // a path is reduced to its basename before lookup
+        assert_eq!(c.resume_args("/usr/local/bin/claude").as_deref(), Some(&["--continue".to_string()][..]));
+        // unknown command → None
+        assert_eq!(c.resume_args("bash"), None);
+        // user override wins over the default
+        c.resume.commands.insert("claude".into(), vec!["--resume".into(), "last".into()]);
+        assert_eq!(c.resume_args("claude"), Some(vec!["--resume".into(), "last".into()]));
+    }
+
+    #[test]
+    fn snapshot_interval_defaults_to_5s() {
+        let c = Config::default();
+        assert_eq!(c.snapshot_interval_secs(), 5);
+    }
+
+    #[test]
+    fn parses_resume_table_and_interval() {
+        let dir = std::env::temp_dir().join(format!("spacesh-cfg-resume-{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let path = dir.join("config.toml");
+        std::fs::write(&path,
+            "snapshot_interval_secs = 10\n[resume.commands]\ngemini = [\"--resume\"]\n").unwrap();
+        let c = Config::from_path(&path);
+        assert_eq!(c.snapshot_interval_secs(), 10);
+        assert_eq!(c.resume_args("gemini"), Some(vec!["--resume".into()]));
+        let _ = std::fs::remove_file(&path);
+    }
 }

crates/spaceshd/src/main.rs

@@ -7,6 +7,7 @@ mod lifecycle;
 mod persist;
 mod registry;
 mod server;
+mod snapshot_store;
 mod state_store;
 mod surface;
 
@@ -60,6 +61,9 @@ async fn run_daemon() -> Result<()> {
     let events_path = lifecycle::spacesh_dir()?.join("events.json");
     let event_store: std::sync::Arc<dyn event_store::EventStore> =
         std::sync::Arc::new(event_store::JsonEventStore::new(events_path));
+    let snapshots_dir = lifecycle::spacesh_dir()?.join("snapshots");
+    let snapshot_store: std::sync::Arc<dyn snapshot_store::SnapshotStore> =
+        std::sync::Arc::new(snapshot_store::JsonSnapshotStore::new(snapshots_dir));
     eprintln!("spaceshd listening on {}", sock.display());
-    server::serve(&sock, store, event_store).await
+    server::serve(&sock, store, event_store, snapshot_store).await
 }

crates/spaceshd/src/registry.rs

@@ -114,6 +114,10 @@ impl Registry {
     pub fn is_running(&self, sid: &SurfaceId) -> bool {
         self.live.contains_key(sid)
     }
+    /// Ids of all currently-live surfaces.
+    pub fn live_ids(&self) -> Vec<SurfaceId> {
+        self.live.keys().cloned().collect()
+    }
 
     // ---- surface state ----
 

crates/spaceshd/src/server.rs

@@ -13,6 +13,7 @@ use crate::event_log::EventLog;
 use crate::event_store::{self, EventPersister, EventStore};
 use crate::persist::{self, Persister};
 use crate::registry::Registry;
+use crate::snapshot_store::{SnapshotStore, SnapshotMsg, spawn_writer};
 use crate::state_store::StateStore;
 use crate::surface::{SurfaceMsg};
 
@@ -33,11 +34,13 @@ enum ServerMsg {
     ClientDisconnected { client: ClientId },
     /// A status change detected internally (OSC 133 / fallback) by a surface actor.
     StateDetected { surface_id: SurfaceId, state: SurfaceState },
+    /// Periodic snapshot tick: ask all live surfaces for a snapshot and persist dirty ones.
+    SnapshotTick,
 }
 
 type ClientId = u64;
 
-pub async fn serve(socket: &Path, store: Arc<dyn StateStore>, event_store: Arc<dyn EventStore>) -> Result<()> {
+pub async fn serve(socket: &Path, store: Arc<dyn StateStore>, event_store: Arc<dyn EventStore>, snapshot_store: Arc<dyn SnapshotStore>) -> Result<()> {
     let listener = UnixListener::bind(socket)?;
     let (router_tx, router_rx) = mpsc::channel::<ServerMsg>(256);
 
@@ -58,6 +61,20 @@ pub async fn serve(socket: &Path, store: Arc<dyn StateStore>, event_store: Arc<d
         }
     });
 
+    let snapshot_tx = spawn_writer(snapshot_store.clone());
+
+    // Periodic snapshot tick → router.
+    let tick_router = router_tx.clone();
+    let interval_secs = crate::config::Config::load().snapshot_interval_secs();
+    tokio::spawn(async move {
+        let mut tick = tokio::time::interval(Duration::from_secs(interval_secs));
+        tick.tick().await; // consume the immediate first tick
+        loop {
+            tick.tick().await;
+            if tick_router.send(ServerMsg::SnapshotTick).await.is_err() { break; }
+        }
+    });
+
     let persister = persist::spawn(store.clone(), Duration::from_millis(500));
     let initial = store.load().unwrap_or_default();
     let event_persister = event_store::spawn(event_store.clone(), Duration::from_millis(500));
@@ -66,7 +83,7 @@ pub async fn serve(socket: &Path, store: Arc<dyn StateStore>, event_store: Arc<d
     let shutdown = tokio::spawn(router(
         router_rx, router_tx.clone(), exit_tx, state_tx,
         persister, initial, event_persister, event_initial,
-        started_at_ms,
+        started_at_ms, snapshot_store, snapshot_tx,
     ));
 
     let mut next_client: ClientId = 0;
@@ -128,6 +145,8 @@ async fn router(
     event_persister: EventPersister,
     event_initial: crate::event_log::EventLogState,
     started_at_ms: u64,
+    snapshot_store: Arc<dyn SnapshotStore>,
+    snapshot_tx: mpsc::UnboundedSender<SnapshotMsg>,
 ) {
     let mut reg = Registry::new();
     reg.restore(initial);
@@ -175,10 +194,24 @@ async fn router(
                     }
                 }
             }
+            ServerMsg::SnapshotTick => {
+                let ids = reg.live_ids();
+                for sid in ids {
+                    let Some(handle) = reg.live(&sid) else { continue };
+                    let (reply_tx, reply_rx) = oneshot::channel();
+                    if handle.tx.send(SurfaceMsg::Snapshot { reply: reply_tx }).await.is_err() { continue; }
+                    if let Ok((snap, dirty)) = reply_rx.await {
+                        if dirty {
+                            let _ = snapshot_tx.send(SnapshotMsg::Save(sid.clone(), snap));
+                        }
+                    }
+                }
+            }
             ServerMsg::Request { id, cmd, client, out } => {
                 handle_request(id, cmd, client, out, &mut reg, &mut subs, &clients,
                     &router_tx, &exit_tx, &state_tx, &persister,
-                    &mut event_log, &event_persister, started_at_ms, &mut config).await;
+                    &mut event_log, &event_persister, started_at_ms, &mut config,
+                    &snapshot_store, &snapshot_tx).await;
             }
         }
     }
@@ -241,7 +274,7 @@ fn err(id: u64, code: &str, msg: &str) -> Envelope {
 /// Compute spawn env (hooks for claude agents, zsh integration for zsh shells)
 /// and whether a deterministic hook source is active.
 fn spawn_env(sid: &SurfaceId, spec: &spacesh_proto::workspace::SurfaceSpec) -> (Vec<(String, String)>, bool) {
-    if crate::hooks::is_agent(&spec.command, spec.agent_label.as_deref()) {
+    let (mut env, active) = if crate::hooks::is_agent(&spec.command, spec.agent_label.as_deref()) {
         let env = crate::hooks::prepare(sid, &crate::hooks::spacesh_bin());
         let active = !env.is_empty();
         (env, active)
@@ -249,7 +282,30 @@ fn spawn_env(sid: &SurfaceId, spec: &spacesh_proto::workspace::SurfaceSpec) -> (
         (crate::hooks::shell_env(sid), false)
     } else {
         (vec![], false)
+    };
+    // Ensure the child sees the user's full PATH; the GUI/launchd-launched daemon
+    // otherwise can't find agents (claude/codex/gemini) and the panel exits at once.
+    if !env.iter().any(|(k, _)| k == "PATH") {
+        env.push(("PATH".to_string(), crate::config::enriched_path()));
     }
+    (env, active)
+}
+
+/// Build the spawn spec for a (re)start. When `resume` and the command has a
+/// resume mapping, its args are replaced with the resume args; otherwise the
+/// original spec args are kept.
+fn resume_spec(
+    spec: &spacesh_proto::workspace::SurfaceSpec,
+    resume: bool,
+    cfg: &crate::config::Config,
+) -> spacesh_proto::workspace::SurfaceSpec {
+    let mut out = spec.clone();
+    if resume {
+        if let Some(args) = cfg.resume_args(&spec.command) {
+            out.args = args;
+        }
+    }
+    out
 }
 
 /// Emit a `layout_changed` event for a workspace's current tree.
@@ -278,6 +334,8 @@ async fn handle_request(
     event_persister: &EventPersister,
     started_at_ms: u64,
     config: &mut crate::config::Config,
+    snapshot_store: &Arc<dyn SnapshotStore>,
+    snapshot_tx: &mpsc::UnboundedSender<SnapshotMsg>,
 ) {
     use spacesh_proto::message::SplitDir;
     use spacesh_proto::layout::{LayoutNode, Orient};
@@ -306,7 +364,7 @@ async fn handle_request(
                 agent_label: command, cols, rows, autostart: false,
             };
             let (env, hooks_active) = spawn_env(&sid, &spec);
-            match crate::surface::spawn_from_spec(sid.clone(), workspace_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone()) {
+            match crate::surface::spawn_from_spec(sid.clone(), workspace_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone(), snapshot_tx.clone()) {
                 Ok(handle) => {
                     spawn_output_bridge(sid.clone(), &handle, router_tx.clone());
                     reg.set_live(handle);
@@ -338,7 +396,7 @@ async fn handle_request(
             let shell = command.clone().unwrap_or_else(|| config.resolved_shell());
             let spec = SurfaceSpec { command: shell, args, cwd: ws.path.clone(), agent_label: command, cols: 80, rows: 24, autostart: false };
             let (env, hooks_active) = spawn_env(&new_sid, &spec);
-            match crate::surface::spawn_from_spec(new_sid.clone(), ws_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone()) {
+            match crate::surface::spawn_from_spec(new_sid.clone(), ws_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone(), snapshot_tx.clone()) {
                 Ok(handle) => {
                     spawn_output_bridge(new_sid.clone(), &handle, router_tx.clone());
                     reg.set_live(handle);
@@ -395,55 +453,56 @@ async fn handle_request(
             let Some(ws) = reg.workspace(&workspace_id).cloned() else {
                 let _ = out.send(err(id, "NOT_FOUND", "workspace")).await; return;
             };
-            // Kill current panels of this workspace.
-            let existing: Vec<SurfaceId> = ws.surfaces.keys().cloned().collect();
-            for sid in &existing {
-                if let Some(h) = reg.live(sid) { let _ = h.tx.send(crate::surface::SurfaceMsg::Close).await; }
-                reg.remove_surface(sid);
-                subs.remove(sid);
-            }
-            // Spawn `count` panels (slots padded/truncated to count).
-            let mut new_ids = Vec::new();
-            for i in 0..count {
-                let slot = slots.get(i);
+            // Additive: keep existing panels (and their live processes) in their
+            // current visual order, spawn only the delta needed to reach `count`,
+            // then rebuild the tree to the preset shape. Presets never destroy
+            // running panels — shrinking is done by closing panels via the X. The
+            // GUI only offers presets whose count >= the current pane count, so
+            // `count >= existing.len()` and `ids.len() == count` after the loop.
+            let existing: Vec<SurfaceId> = ws.layout.as_ref()
+                .map(spacesh_core::ops::leaves)
+                .unwrap_or_else(|| ws.surfaces.keys().cloned().collect());
+            let mut ids = existing.clone();
+            let to_spawn = count.saturating_sub(existing.len());
+            for j in 0..to_spawn {
+                let slot = slots.get(existing.len() + j);
                 let new_sid = reg.new_surface_id();
                 let command = slot.and_then(|s| s.command.clone());
                 let shell = command.clone().unwrap_or_else(|| config.resolved_shell());
                 let args = slot.map(|s| s.args.clone()).unwrap_or_default();
                 let spec = SurfaceSpec { command: shell, args, cwd: ws.path.clone(), agent_label: command, cols: 80, rows: 24, autostart: false };
                 let (env, hooks_active) = spawn_env(&new_sid, &spec);
-                match crate::surface::spawn_from_spec(new_sid.clone(), workspace_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone()) {
+                match crate::surface::spawn_from_spec(new_sid.clone(), workspace_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone(), snapshot_tx.clone()) {
                     Ok(handle) => {
                         spawn_output_bridge(new_sid.clone(), &handle, router_tx.clone());
                         reg.set_live(handle);
                         reg.set_state(&new_sid, spacesh_proto::SurfaceState::Idle);
                         reg.add_surface_spec(&workspace_id, new_sid.clone(), spec);
-                        new_ids.push(new_sid);
+                        broadcast_evt(clients, &Envelope::Evt(Evt::SurfaceCreated { surface_id: new_sid.clone(), workspace_id: workspace_id.clone() }));
+                        ids.push(new_sid);
                     }
                     Err(e) => { let _ = out.send(err(id, "SPAWN_FAILED", &e.to_string())).await; return; }
                 }
             }
-            if let Some(tree) = spacesh_core::presets::build(&preset_id, &new_ids) {
+            if let Some(tree) = spacesh_core::presets::build(&preset_id, &ids) {
                 if let Some(w) = reg.workspace_mut(&workspace_id) { w.layout = Some(tree); }
             }
-            for sid in &new_ids {
-                broadcast_evt(clients, &Envelope::Evt(Evt::SurfaceCreated { surface_id: sid.clone(), workspace_id: workspace_id.clone() }));
-            }
             emit_layout(reg, &workspace_id, clients);
             persister.mark_dirty(reg.persist_state());
-            let _ = out.send(ok(id, serde_json::json!({ "surface_ids": new_ids.iter().map(|s| s.0.clone()).collect::<Vec<_>>() }))).await;
+            let _ = out.send(ok(id, serde_json::json!({ "surface_ids": ids.iter().map(|s| s.0.clone()).collect::<Vec<_>>() }))).await;
         }
 
-        Cmd::RestartSurface { surface_id } => {
+        Cmd::RestartSurface { surface_id, resume } => {
             if reg.is_running(&surface_id) {
                 let _ = out.send(ok(id, serde_json::Value::Null)).await; return; // already running
             }
             let Some(spec) = reg.surface_spec(&surface_id) else {
                 let _ = out.send(err(id, "NOT_FOUND", "surface")).await; return;
             };
+            let spec = resume_spec(&spec, resume, config);
             let ws_id = reg.workspace_of(&surface_id).unwrap();
             let (env, hooks_active) = spawn_env(&surface_id, &spec);
-            match crate::surface::spawn_from_spec(surface_id.clone(), ws_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone()) {
+            match crate::surface::spawn_from_spec(surface_id.clone(), ws_id.clone(), &spec, env, hooks_active, state_tx.clone(), exit_tx.clone(), snapshot_tx.clone()) {
                 Ok(handle) => {
                     spawn_output_bridge(surface_id.clone(), &handle, router_tx.clone());
                     reg.set_live(handle);
@@ -458,6 +517,7 @@ async fn handle_request(
         Cmd::CloseWorkspace { workspace_id } => {
             let ids = reg.close_workspace(&workspace_id);
             for sid in &ids { crate::hooks::cleanup(sid); crate::hooks::cleanup_shell(sid); subs.remove(sid); }
+            for sid in &ids { let _ = snapshot_tx.send(SnapshotMsg::Remove(sid.clone())); }
             broadcast_evt(clients, &Envelope::Evt(Evt::WorkspaceClosed { workspace_id: workspace_id.clone() }));
             persister.mark_dirty(reg.persist_state());
             let _ = out.send(ok(id, serde_json::Value::Null)).await;
@@ -547,8 +607,18 @@ async fn handle_request(
                 }
                 let _ = out.send(err(id, "INTERNAL", "attach failed")).await;
             } else {
-                // stopped panel: no live stream, return an empty snapshot so the GUI shows the restart overlay.
-                let _ = out.send(ok(id, serde_json::json!({ "snapshot": "", "cols": 0, "rows": 0, "stopped": true }))).await;
+                // stopped panel: no live stream. Paint the last on-disk screen if we have one.
+                match snapshot_store.load(&surface_id) {
+                    Some(snap) => {
+                        let _ = out.send(ok(id, serde_json::json!({
+                            "snapshot": snap.ansi, "cols": snap.cols, "rows": snap.rows,
+                            "cursor_row": snap.cursor_row, "cursor_col": snap.cursor_col, "stopped": true,
+                        }))).await;
+                    }
+                    None => {
+                        let _ = out.send(ok(id, serde_json::json!({ "snapshot": "", "cols": 0, "rows": 0, "stopped": true }))).await;
+                    }
+                }
             }
         }
 
@@ -574,6 +644,7 @@ async fn handle_request(
                 subs.remove(&surface_id);
                 crate::hooks::cleanup(&surface_id);
                 crate::hooks::cleanup_shell(&surface_id);
+                let _ = snapshot_tx.send(SnapshotMsg::Remove(surface_id.clone()));
                 broadcast_evt(clients, &Envelope::Evt(Evt::SurfaceClosed { surface_id: surface_id.clone() }));
                 if let Some(ws_id) = ws_id {
                     emit_layout(reg, &ws_id, clients);
@@ -628,6 +699,11 @@ async fn handle_request(
             }))).await;
         }
 
+        Cmd::WhichAgents { candidates } => {
+            let available: Vec<String> = candidates.into_iter().filter(|c| crate::config::is_installed(c)).collect();
+            let _ = out.send(ok(id, serde_json::json!({ "available": available }))).await;
+        }
+
         Cmd::Status => {
             let (groups, workspaces) = reg.status();
             let _ = out.send(ok(id, serde_json::json!({ "groups": groups, "workspaces": workspaces }))).await;
@@ -656,6 +732,19 @@ async fn handle_request(
         }
 
         Cmd::Shutdown => {
+            // Final snapshot pass: capture each live surface's visible screen so a
+            // clean restart (e.g. Settings → Restart daemon) repaints last screens.
+            // Written synchronously through the store (the async writer task would
+            // not drain before process::exit).
+            for sid in reg.live_ids() {
+                let Some(handle) = reg.live(&sid) else { continue };
+                let (reply_tx, reply_rx) = oneshot::channel();
+                if handle.tx.send(SurfaceMsg::Snapshot { reply: reply_tx }).await.is_ok() {
+                    if let Ok((snap, _dirty)) = reply_rx.await {
+                        snapshot_store.save(&sid, &snap);
+                    }
+                }
+            }
             let _ = out.send(ok(id, serde_json::Value::Null)).await;
             std::process::exit(0);
         }
@@ -728,6 +817,7 @@ fn spawn_output_bridge(
 mod tests {
     use super::*;
     use base64::Engine;
+    use crate::snapshot_store::NullSnapshotStore;
 
     async fn req(stream: &mut UnixStream, id: u64, cmd: Cmd) -> Envelope {
         write_frame(stream, &Envelope::Req { id, cmd }).await.unwrap();
@@ -776,7 +866,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         let mut s = UnixStream::connect(&sock).await.unwrap();
@@ -818,7 +908,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
         let mut s = UnixStream::connect(&sock).await.unwrap();
         let r = req(&mut s, 1, Cmd::Input {
@@ -844,7 +934,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // First client: open, new surface that prints a marker, attach, then disconnect.
@@ -883,7 +973,7 @@ mod tests {
             std::sync::Arc::new(crate::state_store::JsonStateStore::new(dir.join("state.json")));
         let event_store = make_event_store(&dir);
         let sock2 = sock.clone();
-        tokio::spawn(async move { let _ = serve(&sock2, store, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock2, store, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
         let mut s = UnixStream::connect(&sock).await.unwrap();
 
@@ -913,7 +1003,7 @@ mod tests {
             std::sync::Arc::new(crate::state_store::JsonStateStore::new(dir.join("state.json")));
         let event_store = make_event_store(&dir);
         let sock2 = sock.clone();
-        tokio::spawn(async move { let _ = serve(&sock2, store, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock2, store, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
         let mut s = UnixStream::connect(&sock).await.unwrap();
 
@@ -957,7 +1047,7 @@ mod tests {
             // per-test dir so instance B reads from disk what instance A persisted.
             let event_store = make_event_store(&dir);
             let sock2 = sock.clone();
-            tokio::spawn(async move { let _ = serve(&sock2, store, event_store).await; });
+            tokio::spawn(async move { let _ = serve(&sock2, store, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
             wait_for_socket(&sock).await;
             let mut s = UnixStream::connect(&sock).await.unwrap();
             let r = req(&mut s, 1, Cmd::Open { path: std::env::temp_dir().to_string_lossy().into() }).await;
@@ -974,7 +1064,7 @@ mod tests {
             std::sync::Arc::new(crate::state_store::JsonStateStore::new(state_path.clone()));
         let event_store_b = make_event_store(&dir);
         let sb2 = sock_b.clone();
-        tokio::spawn(async move { let _ = serve(&sock_b, store_b, event_store_b).await; });
+        tokio::spawn(async move { let _ = serve(&sock_b, store_b, event_store_b, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sb2).await;
         let mut s2 = UnixStream::connect(&sb2).await.unwrap();
         let r = req(&mut s2, 1, Cmd::Status).await;
@@ -996,7 +1086,7 @@ mod tests {
             std::sync::Arc::new(crate::state_store::JsonStateStore::new(dir.join("state.json")));
         let event_store = make_event_store(&dir);
         let sock2 = sock.clone();
-        tokio::spawn(async move { let _ = serve(&sock2, store, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock2, store, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
         let mut s = UnixStream::connect(&sock).await.unwrap();
 
@@ -1034,7 +1124,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // Control connection: open workspace and spawn surface.
@@ -1087,7 +1177,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // Control connection: open workspace and spawn surface.
@@ -1139,7 +1229,7 @@ mod tests {
             std::sync::Arc::new(crate::state_store::JsonStateStore::new(dir.join("state.json")));
         let event_store = make_event_store(&dir);
         let sock2 = sock.clone();
-        tokio::spawn(async move { let _ = serve(&sock2, store, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock2, store, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // Control connection: open workspace and spawn a surface that emits OSC 133.
@@ -1190,7 +1280,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // Observer connection to catch the EventsRead broadcast.
@@ -1258,7 +1348,7 @@ mod tests {
                 std::sync::Arc::new(crate::state_store::JsonStateStore::new(state_path.clone()));
             let event_store = make_event_store(&dir);
             let sock2 = sock.clone();
-            tokio::spawn(async move { let _ = serve(&sock2, store, event_store).await; });
+            tokio::spawn(async move { let _ = serve(&sock2, store, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
             wait_for_socket(&sock).await;
 
             let mut s = UnixStream::connect(&sock).await.unwrap();
@@ -1300,7 +1390,7 @@ mod tests {
             std::sync::Arc::new(crate::state_store::JsonStateStore::new(state_path.clone()));
         let event_store_b = make_event_store(&dir);
         let sb2 = sock_b.clone();
-        tokio::spawn(async move { let _ = serve(&sock_b, store_b, event_store_b).await; });
+        tokio::spawn(async move { let _ = serve(&sock_b, store_b, event_store_b, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sb2).await;
 
         let mut s2 = UnixStream::connect(&sb2).await.unwrap();
@@ -1328,7 +1418,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // Observer connection.
@@ -1390,7 +1480,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
         let mut s = UnixStream::connect(&sock).await.unwrap();
 
@@ -1413,7 +1503,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
         let mut s = UnixStream::connect(&sock).await.unwrap();
 
@@ -1460,7 +1550,7 @@ mod tests {
         let event_store = make_event_store(&dir);
         let sock_for_task = sock.clone();
         let store2 = store.clone();
-        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store).await; });
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, std::sync::Arc::new(NullSnapshotStore)).await; });
         wait_for_socket(&sock).await;
 
         // Control connection: open, spawn, zoom.
@@ -1500,4 +1590,73 @@ mod tests {
         }
         assert!(saw_cleared, "expected a WorkspaceChanged broadcast with cleared zoom after closing the zoomed surface");
     }
+
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn stopped_attach_returns_disk_snapshot() {
+        let _serial = crate::test_support::serial();
+        let dir = tempdir_path();
+        let sock = dir.join("sock");
+        let store: std::sync::Arc<dyn crate::state_store::StateStore> =
+            std::sync::Arc::new(crate::state_store::JsonStateStore::new(dir.join("state.json")));
+        let event_store = make_event_store(&dir);
+        // Use a real JsonSnapshotStore so the on-exit dump actually lands on disk.
+        let snap_dir = dir.join("snapshots");
+        let snapshot_store: std::sync::Arc<dyn crate::snapshot_store::SnapshotStore> =
+            std::sync::Arc::new(crate::snapshot_store::JsonSnapshotStore::new(snap_dir.clone()));
+        let sock_for_task = sock.clone();
+        let store2 = store.clone();
+        let snap_store2 = snapshot_store.clone();
+        tokio::spawn(async move { let _ = serve(&sock_for_task, store2, event_store, snap_store2).await; });
+        wait_for_socket(&sock).await;
+
+        // Open workspace, spawn a surface that prints a unique marker then exits quickly.
+        let surface_id;
+        {
+            let mut s = UnixStream::connect(&sock).await.unwrap();
+            let r = req(&mut s, 1, Cmd::Open { path: std::env::temp_dir().to_string_lossy().into() }).await;
+            let ws = res_data(&r)["workspace_id"].as_str().unwrap().to_string();
+            let r = req(&mut s, 2, Cmd::NewSurface {
+                workspace_id: spacesh_proto::WorkspaceId(ws),
+                command: Some("/bin/sh".into()),
+                args: vec!["-c".into(), "printf SNAPDISK; sleep 0.2".into()],
+                cols: 80, rows: 24,
+            }).await;
+            surface_id = spacesh_proto::SurfaceId(res_data(&r)["surface_id"].as_str().unwrap().to_string());
+            // Give the process time to run, exit, and the actor to dump its snapshot to the writer.
+            tokio::time::sleep(tokio::time::Duration::from_millis(1500)).await;
+            // s drops here
+        }
+
+        // Re-verify the socket is still alive.
+        wait_for_socket(&sock).await;
+
+        // Fresh client: attach to the now-stopped surface.
+        let mut s2 = UnixStream::connect(&sock).await.unwrap();
+        let r = req(&mut s2, 1, Cmd::Attach { surface_id: surface_id.clone() }).await;
+        let data = res_data(&r);
+        assert_eq!(data["stopped"].as_bool(), Some(true), "surface should be stopped");
+        let snap = data["snapshot"].as_str().unwrap_or("");
+        assert!(snap.contains("SNAPDISK"), "on-disk snapshot should contain SNAPDISK, got: {snap:?}");
+    }
+
+    #[test]
+    fn resume_spec_swaps_args_when_mapped() {
+        use spacesh_proto::workspace::SurfaceSpec;
+        let spec = SurfaceSpec {
+            command: "claude".into(), args: vec!["--foo".into()], cwd: "/tmp".into(),
+            agent_label: Some("claude".into()), cols: 80, rows: 24, autostart: false,
+        };
+        let cfg = crate::config::Config::default();
+        // resume=false → original args
+        let plain = resume_spec(&spec, false, &cfg);
+        assert_eq!(plain.args, vec!["--foo".to_string()]);
+        // resume=true with a default mapping → resume args
+        let resumed = resume_spec(&spec, true, &cfg);
+        assert_eq!(resumed.args, vec!["--continue".to_string()]);
+        // resume=true for an unmapped command → original args (graceful fallback)
+        let mut shell = spec.clone();
+        shell.command = "bash".into();
+        let resumed_shell = resume_spec(&shell, true, &cfg);
+        assert_eq!(resumed_shell.args, shell.args);
+    }
 }

crates/spaceshd/src/snapshot_store.rs

@@ -0,0 +1,162 @@
+use std::path::PathBuf;
+use spacesh_core::snapshot::Snapshot;
+use spacesh_proto::SurfaceId;
+
+/// Stores one visible-screen snapshot per surface as `<dir>/<surface_id>.json`.
+pub trait SnapshotStore: Send + Sync {
+    fn save(&self, sid: &SurfaceId, snap: &Snapshot);
+    fn load(&self, sid: &SurfaceId) -> Option<Snapshot>;
+    fn remove(&self, sid: &SurfaceId);
+}
+
+/// Writer command: persist or delete a surface's snapshot. Shared by the
+/// router ticker, the close/remove paths, and each actor's on-exit dump, so a
+/// single channel type flows everywhere.
+pub enum SnapshotMsg {
+    Save(SurfaceId, Snapshot),
+    Remove(SurfaceId),
+}
+
+/// A no-op store for tests and contexts that do not persist snapshots.
+pub struct NullSnapshotStore;
+impl SnapshotStore for NullSnapshotStore {
+    fn save(&self, _sid: &SurfaceId, _snap: &Snapshot) {}
+    fn load(&self, _sid: &SurfaceId) -> Option<Snapshot> { None }
+    fn remove(&self, _sid: &SurfaceId) {}
+}
+
+/// JSON file store. Filenames are the surface id (e.g. `s_1f.json`); ids are
+/// `^[a-z]_[0-9a-f]+$` so they are always safe path components.
+pub struct JsonSnapshotStore {
+    dir: PathBuf,
+}
+
+impl JsonSnapshotStore {
+    pub fn new(dir: PathBuf) -> Self {
+        let _ = std::fs::create_dir_all(&dir);
+        Self { dir }
+    }
+    fn path(&self, sid: &SurfaceId) -> PathBuf {
+        self.dir.join(format!("{}.json", sid.0))
+    }
+}
+
+impl SnapshotStore for JsonSnapshotStore {
+    fn save(&self, sid: &SurfaceId, snap: &Snapshot) {
+        let path = self.path(sid);
+        let tmp = path.with_extension("json.tmp");
+        let Ok(bytes) = serde_json::to_vec(snap) else { return };
+        if std::fs::write(&tmp, &bytes).is_err() { return; }
+        if std::fs::File::open(&tmp).and_then(|f| f.sync_all()).is_err() { return; }
+        let _ = std::fs::rename(&tmp, &path);
+    }
+    fn load(&self, sid: &SurfaceId) -> Option<Snapshot> {
+        let bytes = std::fs::read(self.path(sid)).ok()?;
+        serde_json::from_slice(&bytes).ok()
+    }
+    fn remove(&self, sid: &SurfaceId) {
+        let _ = std::fs::remove_file(self.path(sid));
+    }
+}
+
+/// Spawn the writer task; returns the sender used by the router and actors.
+pub fn spawn_writer(store: std::sync::Arc<dyn SnapshotStore>) -> tokio::sync::mpsc::UnboundedSender<SnapshotMsg> {
+    let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel::<SnapshotMsg>();
+    tokio::spawn(async move {
+        while let Some(msg) = rx.recv().await {
+            match msg {
+                SnapshotMsg::Save(sid, snap) => store.save(&sid, &snap),
+                SnapshotMsg::Remove(sid) => store.remove(&sid),
+            }
+        }
+    });
+    tx
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn tmp_dir(name: &str) -> PathBuf {
+        let n = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).unwrap().as_nanos();
+        let p = std::env::temp_dir().join(format!("spacesh-snap-{name}-{n}"));
+        std::fs::create_dir_all(&p).unwrap();
+        p
+    }
+
+    fn sample() -> Snapshot {
+        Snapshot { ansi: "\u{1b}[mhello".into(), cols: 80, rows: 24, cursor_row: 1, cursor_col: 6 }
+    }
+
+    #[test]
+    fn save_then_load_round_trips() {
+        let dir = tmp_dir("roundtrip");
+        let store = JsonSnapshotStore::new(dir.clone());
+        let sid = SurfaceId("s_1".into());
+        store.save(&sid, &sample());
+        assert_eq!(store.load(&sid), Some(sample()));
+        let _ = std::fs::remove_dir_all(dir);
+    }
+
+    #[test]
+    fn missing_loads_none() {
+        let store = JsonSnapshotStore::new(tmp_dir("missing"));
+        assert_eq!(store.load(&SurfaceId("s_none".into())), None);
+    }
+
+    #[test]
+    fn corrupt_loads_none() {
+        let dir = tmp_dir("corrupt");
+        let store = JsonSnapshotStore::new(dir.clone());
+        let sid = SurfaceId("s_2".into());
+        std::fs::write(dir.join("s_2.json"), b"{ not json").unwrap();
+        assert_eq!(store.load(&sid), None);
+        let _ = std::fs::remove_dir_all(dir);
+    }
+
+    #[test]
+    fn remove_deletes_file() {
+        let dir = tmp_dir("remove");
+        let store = JsonSnapshotStore::new(dir.clone());
+        let sid = SurfaceId("s_3".into());
+        store.save(&sid, &sample());
+        assert!(store.load(&sid).is_some());
+        store.remove(&sid);
+        assert_eq!(store.load(&sid), None);
+        let _ = std::fs::remove_dir_all(dir);
+    }
+
+    #[test]
+    fn null_store_is_inert() {
+        let store = NullSnapshotStore;
+        let sid = SurfaceId("s_4".into());
+        store.save(&sid, &sample());
+        assert_eq!(store.load(&sid), None);
+        store.remove(&sid);
+    }
+
+    #[tokio::test]
+    async fn writer_saves_and_removes() {
+        let dir = tmp_dir("writer");
+        let store: std::sync::Arc<dyn SnapshotStore> = std::sync::Arc::new(JsonSnapshotStore::new(dir.clone()));
+        let tx = spawn_writer(store.clone());
+        let sid = SurfaceId("s_w".into());
+
+        tx.send(SnapshotMsg::Save(sid.clone(), sample())).unwrap();
+        let mut saved = None;
+        for _ in 0..50 {
+            if let Some(s) = store.load(&sid) { saved = Some(s); break; }
+            tokio::time::sleep(std::time::Duration::from_millis(10)).await;
+        }
+        assert_eq!(saved, Some(sample()));
+
+        tx.send(SnapshotMsg::Remove(sid.clone())).unwrap();
+        let mut gone = false;
+        for _ in 0..50 {
+            if store.load(&sid).is_none() { gone = true; break; }
+            tokio::time::sleep(std::time::Duration::from_millis(10)).await;
+        }
+        assert!(gone, "writer should have removed the snapshot file");
+        let _ = std::fs::remove_dir_all(dir);
+    }
+}

crates/spaceshd/src/surface.rs

@@ -7,6 +7,7 @@ use spacesh_proto::workspace::SurfaceSpec;
 use spacesh_pty::{PtyHandle, SpawnSpec};
 use tokio::sync::{broadcast, mpsc, oneshot};
 use tokio::time::{Duration, Instant};
+use crate::snapshot_store::SnapshotMsg;
 
 /// Spawn (or restart) a surface actor from a persisted spec. Injects
 /// SPACESH_SURFACE_ID into the child env, mirroring `new_surface`.
@@ -24,6 +25,7 @@ pub fn spawn_from_spec(
     hooks_active: bool,
     state_tx: mpsc::UnboundedSender<(SurfaceId, SurfaceState)>,
     exit_tx: mpsc::UnboundedSender<(SurfaceId, i32)>,
+    snapshot_tx: mpsc::UnboundedSender<SnapshotMsg>,
 ) -> std::io::Result<SurfaceHandle> {
     let mut env = vec![("SPACESH_SURFACE_ID".to_string(), id.0.clone())];
     env.extend(extra_env);
@@ -35,7 +37,7 @@ pub fn spawn_from_spec(
         rows: spec.rows,
         env,
     };
-    Ok(spawn_surface_deferred(id, workspace_id, spawn_spec, spec.cols, spec.rows, hooks_active, state_tx, exit_tx))
+    Ok(spawn_surface_deferred(id, workspace_id, spawn_spec, spec.cols, spec.rows, hooks_active, state_tx, exit_tx, snapshot_tx))
 }
 
 const BROADCAST_CAP: usize = 1024;
@@ -53,6 +55,8 @@ pub enum SurfaceMsg {
     Attach { reply: oneshot::Sender<broadcast::Receiver<Vec<u8>>> },
     /// Attach with snapshot: subscribe AND capture the grid in one actor turn.
     AttachSnapshot { reply: oneshot::Sender<(Snapshot, broadcast::Receiver<Vec<u8>>)> },
+    /// On-demand snapshot without subscribing; bool = dirty since last snapshot.
+    Snapshot { reply: oneshot::Sender<(Snapshot, bool)> },
     Close,
 }
 
@@ -76,10 +80,11 @@ pub fn spawn_surface(
     hooks_active: bool,
     state_tx: mpsc::UnboundedSender<(SurfaceId, SurfaceState)>,
     exit_tx: mpsc::UnboundedSender<(SurfaceId, i32)>,
+    snapshot_tx: mpsc::UnboundedSender<SnapshotMsg>,
 ) -> SurfaceHandle {
     let (tx, rx) = mpsc::channel::<SurfaceMsg>(64);
     let (bcast, _) = broadcast::channel::<Vec<u8>>(BROADCAST_CAP);
-    tokio::spawn(run_actor(id.clone(), pty, cols, rows, hooks_active, bcast, rx, state_tx, exit_tx, Vec::new()));
+    tokio::spawn(run_actor(id.clone(), pty, cols, rows, hooks_active, bcast, rx, state_tx, exit_tx, snapshot_tx, Vec::new()));
     SurfaceHandle { id, workspace_id, tx }
 }
 
@@ -97,6 +102,7 @@ pub fn spawn_surface_deferred(
     hooks_active: bool,
     state_tx: mpsc::UnboundedSender<(SurfaceId, SurfaceState)>,
     exit_tx: mpsc::UnboundedSender<(SurfaceId, i32)>,
+    snapshot_tx: mpsc::UnboundedSender<SnapshotMsg>,
 ) -> SurfaceHandle {
     let (tx, mut rx) = mpsc::channel::<SurfaceMsg>(64);
     let (bcast, _) = broadcast::channel::<Vec<u8>>(BROADCAST_CAP);
@@ -122,6 +128,10 @@ pub fn spawn_surface_deferred(
                         let snap = snapshot_ansi(&GridSurface::new(cols, rows));
                         let _ = reply.send((snap, sub));
                     }
+                    Some(SurfaceMsg::Snapshot { reply }) => {
+                        let snap = snapshot_ansi(&GridSurface::new(cols, rows));
+                        let _ = reply.send((snap, false));
+                    }
                     Some(SurfaceMsg::Close) | None => break false,
                 }
             }
@@ -135,7 +145,7 @@ pub fn spawn_surface_deferred(
         spec.cols = cols;
         spec.rows = rows;
         match PtyHandle::spawn(spec) {
-            Ok(pty) => run_actor(actor_id, pty, cols, rows, hooks_active, bcast, rx, state_tx, exit_tx, prebuf).await,
+            Ok(pty) => run_actor(actor_id, pty, cols, rows, hooks_active, bcast, rx, state_tx, exit_tx, snapshot_tx, prebuf).await,
             Err(_) => { let _ = exit_tx.send((actor_id, 127)); }
         }
     });
@@ -156,6 +166,7 @@ async fn run_actor(
     mut rx: mpsc::Receiver<SurfaceMsg>,
     state_tx: mpsc::UnboundedSender<(SurfaceId, SurfaceState)>,
     exit_tx: mpsc::UnboundedSender<(SurfaceId, i32)>,
+    snapshot_tx: mpsc::UnboundedSender<SnapshotMsg>,
     prebuffered_input: Vec<u8>,
 ) {
     let actor_id = id.clone();
@@ -173,6 +184,7 @@ async fn run_actor(
         // (hooks active, or any OSC 133 marker observed).
         let mut deterministic = hooks_active;
         let mut last_state = SurfaceState::Idle;
+        let mut dirty = false;
 
         loop {
             // Copy the deadline into an owned local so the timer future doesn't
@@ -202,8 +214,15 @@ async fn run_actor(
                             // this snapshot. Broadcasting here would double-render on reattach.
                             let sub = bcast.subscribe();
                             let snap = snapshot_ansi(&grid);
+                            dirty = false;
                             let _ = reply.send((snap, sub));
                         }
+                        Some(SurfaceMsg::Snapshot { reply }) => {
+                            let snap = snapshot_ansi(&grid);
+                            let was_dirty = dirty;
+                            dirty = false;
+                            let _ = reply.send((snap, was_dirty));
+                        }
                         Some(SurfaceMsg::Close) | None => { pty.kill(); break; }
                     }
                 }
@@ -211,26 +230,31 @@ async fn run_actor(
                     match chunk {
                         Some(bytes) => {
                             pending.extend_from_slice(&bytes);
+                            dirty = true;
                             if flush_deadline.is_none() {
                                 flush_deadline = Some(Instant::now() + FLUSH_INTERVAL);
                             }
                             if pending.len() >= FLUSH_BYTES {
-                                flush(&mut pending, &mut grid, &mut osc, &mut deterministic, &mut last_state, &detect_id, &bcast, &state_tx);
+                                let replies = flush(&mut pending, &mut grid, &mut osc, &mut deterministic, &mut last_state, &detect_id, &bcast, &state_tx);
+                                if !replies.is_empty() { let _ = pty.write_input(&replies); }
                                 flush_deadline = None;
                             }
                         }
                         None => {
-                            flush(&mut pending, &mut grid, &mut osc, &mut deterministic, &mut last_state, &detect_id, &bcast, &state_tx);
+                            let _ = flush(&mut pending, &mut grid, &mut osc, &mut deterministic, &mut last_state, &detect_id, &bcast, &state_tx);
                             break;
                         }
                     }
                 }
                 _ = timer => {
-                    flush(&mut pending, &mut grid, &mut osc, &mut deterministic, &mut last_state, &detect_id, &bcast, &state_tx);
+                    let replies = flush(&mut pending, &mut grid, &mut osc, &mut deterministic, &mut last_state, &detect_id, &bcast, &state_tx);
+                    if !replies.is_empty() { let _ = pty.write_input(&replies); }
                     flush_deadline = None;
                 }
             }
         }
+        let final_snap = snapshot_ansi(&grid);
+        let _ = snapshot_tx.send(SnapshotMsg::Save(actor_id.clone(), final_snap));
         let code = pty.wait();
         let _ = exit_tx.send((actor_id, code));
     }
@@ -238,6 +262,8 @@ async fn run_actor(
 
 /// Feed pending bytes into the grid, run detectors, broadcast output, and emit a
 /// state change (if any). No-op when pending is empty.
+/// Returns escape-sequence replies the terminal model produced (DA/DSR answers) that
+/// the caller must write back to the PTY. Empty when there's nothing to feed or reply.
 #[allow(clippy::too_many_arguments)]
 fn flush(
     pending: &mut Vec<u8>,
@@ -248,9 +274,9 @@ fn flush(
     id: &SurfaceId,
     bcast: &broadcast::Sender<Vec<u8>>,
     state_tx: &mpsc::UnboundedSender<(SurfaceId, SurfaceState)>,
-) {
+) -> Vec<u8> {
     if pending.is_empty() {
-        return;
+        return Vec::new();
     }
     // Deterministic source: OSC 133 markers in this chunk.
     // Emit each distinct state transition immediately so no marker is dropped
@@ -265,6 +291,9 @@ fn flush(
         }
     }
     grid.feed(&pending[..]);
+    // Answers to device-attribute / status queries the model just parsed; the actor
+    // writes these back to the PTY so query-blocking programs (fish) don't time out.
+    let replies = grid.take_replies();
     // Best-effort fallback only when no deterministic source is active.
     if !had_osc && !*deterministic {
         if let Some(st) = FallbackScanner::scan(&grid.tail_text(6)) {
@@ -275,6 +304,7 @@ fn flush(
         }
     }
     let _ = bcast.send(std::mem::take(pending));
+    replies
 }
 
 #[cfg(test)]
@@ -299,7 +329,8 @@ mod tests {
         let pty = PtyHandle::spawn(spec("printf HELLO; sleep 0.3")).unwrap();
         let (state_tx, _state_rx) = mpsc::unbounded_channel();
         let (exit_tx, _exit_rx) = mpsc::unbounded_channel();
-        let handle = spawn_surface(SurfaceId("s_1".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx);
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let handle = spawn_surface(SurfaceId("s_1".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx, snap_tx);
 
         let (reply_tx, reply_rx) = oneshot::channel();
         handle.tx.send(SurfaceMsg::Attach { reply: reply_tx }).await.unwrap();
@@ -324,7 +355,8 @@ mod tests {
         let pty = PtyHandle::spawn(spec("exit 7")).unwrap();
         let (state_tx, _state_rx) = mpsc::unbounded_channel();
         let (exit_tx, mut exit_rx) = mpsc::unbounded_channel();
-        let _handle = spawn_surface(SurfaceId("s_2".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx);
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let _handle = spawn_surface(SurfaceId("s_2".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx, snap_tx);
         let (sid, code) = tokio::time::timeout(tokio::time::Duration::from_secs(3), exit_rx.recv())
             .await.unwrap().unwrap();
         assert_eq!(sid, SurfaceId("s_2".into()));
@@ -337,7 +369,8 @@ mod tests {
         let pty = PtyHandle::spawn(spec("printf SNAPME; sleep 0.5")).unwrap();
         let (state_tx, _state_rx) = mpsc::unbounded_channel();
         let (exit_tx, _exit_rx) = mpsc::unbounded_channel();
-        let handle = spawn_surface(SurfaceId("s_s".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx);
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let handle = spawn_surface(SurfaceId("s_s".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx, snap_tx);
 
         // Give the child time to write and the actor time to flush into the grid.
         tokio::time::sleep(tokio::time::Duration::from_millis(200)).await;
@@ -359,7 +392,8 @@ mod tests {
         };
         let (state_tx, _state_rx) = mpsc::unbounded_channel();
         let (exit_tx, _rx) = mpsc::unbounded_channel();
-        let handle = spawn_from_spec(SurfaceId("s_r".into()), WorkspaceId("w_1".into()), &spec, vec![], false, state_tx, exit_tx).unwrap();
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let handle = spawn_from_spec(SurfaceId("s_r".into()), WorkspaceId("w_1".into()), &spec, vec![], false, state_tx, exit_tx, snap_tx).unwrap();
         let (reply_tx, reply_rx) = oneshot::channel();
         handle.tx.send(SurfaceMsg::Attach { reply: reply_tx }).await.unwrap();
         let mut sub = reply_rx.await.unwrap();
@@ -404,7 +438,8 @@ mod tests {
         let _serial = crate::test_support::serial();
         let (state_tx, _s) = mpsc::unbounded_channel();
         let (exit_tx, _e) = mpsc::unbounded_channel();
-        let handle = spawn_surface_deferred(SurfaceId("s_d".into()), WorkspaceId("w_1".into()), stty_spec(), 80, 24, false, state_tx, exit_tx);
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let handle = spawn_surface_deferred(SurfaceId("s_d".into()), WorkspaceId("w_1".into()), stty_spec(), 80, 24, false, state_tx, exit_tx, snap_tx);
 
         let (rtx, rrx) = oneshot::channel();
         handle.tx.send(SurfaceMsg::Attach { reply: rtx }).await.unwrap();
@@ -421,7 +456,8 @@ mod tests {
         let _serial = crate::test_support::serial();
         let (state_tx, _s) = mpsc::unbounded_channel();
         let (exit_tx, _e) = mpsc::unbounded_channel();
-        let handle = spawn_surface_deferred(SurfaceId("s_f".into()), WorkspaceId("w_1".into()), stty_spec(), 80, 24, false, state_tx, exit_tx);
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let handle = spawn_surface_deferred(SurfaceId("s_f".into()), WorkspaceId("w_1".into()), stty_spec(), 80, 24, false, state_tx, exit_tx, snap_tx);
 
         let (rtx, rrx) = oneshot::channel();
         handle.tx.send(SurfaceMsg::Attach { reply: rtx }).await.unwrap();
@@ -437,7 +473,8 @@ mod tests {
         let pty = PtyHandle::spawn(spec("printf '\\033]133;C\\007'; printf working; printf '\\033]133;D;0\\007'; sleep 0.3")).unwrap();
         let (state_tx, mut state_rx) = mpsc::unbounded_channel();
         let (exit_tx, _exit_rx) = mpsc::unbounded_channel();
-        let _h = spawn_surface(SurfaceId("s_o".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx);
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let _h = spawn_surface(SurfaceId("s_o".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx, snap_tx);
         let mut seen = Vec::new();
         let deadline = tokio::time::Instant::now() + tokio::time::Duration::from_secs(2);
         while tokio::time::Instant::now() < deadline {
@@ -449,4 +486,45 @@ mod tests {
         assert!(seen.contains(&SurfaceState::Work), "states: {seen:?}");
         assert!(seen.contains(&SurfaceState::Done), "states: {seen:?}");
     }
+
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn snapshot_msg_returns_grid_and_tracks_dirty() {
+        let _serial = crate::test_support::serial();
+        let pty = PtyHandle::spawn(spec("printf DIRTYME; sleep 0.4")).unwrap();
+        let (state_tx, _s) = mpsc::unbounded_channel();
+        let (exit_tx, _e) = mpsc::unbounded_channel();
+        let (snap_tx, _snap_rx) = mpsc::unbounded_channel();
+        let handle = spawn_surface(SurfaceId("s_1".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx, snap_tx);
+
+        tokio::time::sleep(Duration::from_millis(150)).await;
+        let (reply_tx, reply_rx) = oneshot::channel();
+        handle.tx.send(SurfaceMsg::Snapshot { reply: reply_tx }).await.unwrap();
+        let (snap, dirty) = reply_rx.await.unwrap();
+        assert!(snap.ansi.contains("DIRTYME"), "snapshot: {:?}", snap.ansi);
+        assert!(dirty, "first snapshot after output should be dirty");
+
+        let (reply_tx, reply_rx) = oneshot::channel();
+        handle.tx.send(SurfaceMsg::Snapshot { reply: reply_tx }).await.unwrap();
+        let (_snap2, dirty2) = reply_rx.await.unwrap();
+        assert!(!dirty2, "second snapshot with no new output should be clean");
+    }
+
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn final_snapshot_sent_on_exit() {
+        let _serial = crate::test_support::serial();
+        let pty = PtyHandle::spawn(spec("printf BYE")).unwrap(); // exits immediately
+        let (state_tx, _s) = mpsc::unbounded_channel();
+        let (exit_tx, _e) = mpsc::unbounded_channel();
+        let (snap_tx, mut snap_rx) = mpsc::unbounded_channel();
+        let _handle = spawn_surface(SurfaceId("s_x".into()), WorkspaceId("w_1".into()), pty, 80, 24, false, state_tx, exit_tx, snap_tx);
+
+        let msg = tokio::time::timeout(Duration::from_secs(2), snap_rx.recv()).await.unwrap().unwrap();
+        match msg {
+            crate::snapshot_store::SnapshotMsg::Save(sid, snap) => {
+                assert_eq!(sid.0, "s_x");
+                assert!(snap.ansi.contains("BYE"), "final snapshot: {:?}", snap.ansi);
+            }
+            _ => panic!("expected a Save message on exit"),
+        }
+    }
 }

scripts/bump_version.mjs

@@ -0,0 +1,26 @@
+#!/usr/bin/env node
+// Bump the patch version in lockstep for the GUI (app/src-tauri/tauri.conf.json)
+// and the daemon/CLI (root Cargo.toml [workspace.package].version), so the app's
+// update check and the daemon's reported version never drift apart.
+import { readFileSync, writeFileSync } from "node:fs";
+
+const TAURI_CONF = "app/src-tauri/tauri.conf.json";
+const CARGO_TOML = "Cargo.toml";
+
+const conf = JSON.parse(readFileSync(TAURI_CONF, "utf8"));
+const parts = conf.version.split(".").map((n) => parseInt(n, 10) || 0);
+parts[2] = (parts[2] || 0) + 1;
+const next = parts.join(".");
+
+conf.version = next;
+writeFileSync(TAURI_CONF, JSON.stringify(conf, null, 2) + "\n");
+
+let cargo = readFileSync(CARGO_TOML, "utf8");
+// Replace only the version inside [workspace.package], not dependency versions.
+cargo = cargo.replace(
+  /(\[workspace\.package\][\s\S]*?version\s*=\s*")[^"]+(")/,
+  `$1${next}$2`
+);
+writeFileSync(CARGO_TOML, cargo);
+
+console.log(`version → ${next}  (GUI + daemon)`);