Let agents show and drive Page view tabs on the iPhone

docs/iphone.md

@@ -2,7 +2,9 @@
 
 The iPhone app is the same `Byre` target as the Mac app, built for iOS: platform differences are `#if os(macOS)` / `#if os(iOS)` blocks, with the small shared aliases (`PlatformImage`, `PlatformColor`, `PlatformFont`, `PlatformView`, `touchOnly`, `deviceKind`) in `mac/Sources/Byre/Platform.swift`. Only the layout is the phone's own (`Phone.swift`): a tab bar with Lead, Tasks and More. Tasks lists every project's active tasks with the Mac's own task rows; a task opens pushed, with the Mac's task bar and tab strip on top and the tab's view below in place of the Mac's attached sheet. Zoomed images and diagrams cover the screen. More holds Notes, Workflows, Status and Settings. Every action that focuses a task or project (a row, a link, a notification, the switcher) bumps `Store.reveals`, which brings the Tasks tab forward with the task open; the server's own focus changes don't.
 
-Left out on the phone: the Server tab and running byre locally, the client connection (`Remote`: the phone is never a client for agents' client tools, so `list_clients` doesn't show it), the Sparkle updater, and keyboard shortcuts. Notifications come from the app's own event stream, so they only show while the app is open; push notifications would need APNs on the server.
+The phone holds the client connection (`Remote`) like a Mac, so agents can show it pages and files and drive its Page view (`show_page`, `show_file`, `drive_page`); it answers every other client request with an error. It sends `Byre-Device: iPhone` (Macs send `Mac`), and the server signs commits and runs `byre client-exec` without `--client` only on Macs.
+
+Left out on the phone: the Server tab and running byre locally, the shell, file, hands and signing client requests, the Sparkle updater, and keyboard shortcuts. Notifications come from the app's own event stream, so they only show while the app is open; push notifications would need APNs on the server.
 
 ## Building
 

internal/agents/clients.go

@@ -36,7 +36,8 @@ func (m *Manager) clientTools(l *live) []mcp.Tool {
 			Description: "List the Macs with byre's app open: the human's own machines, not the one byre (and your worktree) runs on. " +
 				"Each has its name, which the other client tools take as client, the id of that running app (pass it as client instead when two apps are open on one Mac), the human using it, and when the human last did something in byre from it. " +
 				"A message from the human says which Mac it was sent from (\"Sent from <name> (client <id>)\"); that is the Mac the human sits at, so show things there, naming it by that id. " +
-				"The most recently active one is listed first.",
+				"The most recently active one is listed first. " +
+				"The human's iPhone is listed too, when byre's app is open on it; it takes only show_page, show_file and drive_page, and the other client tools fail there.",
 			InputSchema: json.RawMessage(`{"type": "object", "properties": {}}`),
 			Handler: func(ctx context.Context, raw json.RawMessage) (string, error) {
 				return m.listClients(), nil
@@ -289,7 +290,7 @@ func (m *Manager) listClients() string {
 	now := time.Now()
 	b := &strings.Builder{}
 	for i, r := range rows {
-		fmt.Fprintf(b, "- %s (%s, id %s): ", r.Host, r.Login, r.ID)
+		fmt.Fprintf(b, "- %s (%s, %s, id %s): ", r.Host, r.Device, r.Login, r.ID)
 		if r.LastActionAt == nil {
 			b.WriteString("no action since byre started")
 		} else {

internal/agents/kb/client-exec.md

@@ -11,7 +11,7 @@ byre client-exec [--client NAME] [--timeout SECS] -- CMD...
 `byre` is `/Applications/Byre.app/Contents/Helpers/byre` on every Mac with the app.
 
 - It runs CMD as `client_bash` does: zsh as a login shell in the home directory, in the app's session, as the human. That session is what GUI-only commands need.
-- Without `--client` it picks the most recently active client, the one `list_clients` lists first.
+- Without `--client` it picks the most recently active Mac, skipping the iPhone, which runs no commands.
 - Each argument reaches the client as one word, so shell syntax needs `-- zsh -c '...'`.
 - The command's stdout and stderr come back separately and byte for byte, once it ends, written to client-exec's own stdout and stderr.
 - Exit status: the command's own; 75 when no client is connected, it didn't acknowledge the request within 2s, or it disconnected; 124 when `--timeout` (default 30s) killed it. A one-line `byre client-exec: ...` on stderr says why.

internal/agents/sign.go

@@ -15,7 +15,7 @@ const signTimeout = 2 * time.Minute
 func (m *Manager) signer(ctx context.Context, user api.User) func(payload []byte) ([]byte, error) {
 	return func(payload []byte) ([]byte, error) {
 		mine := []clients.Client{}
-		for _, c := range m.clients.Recent() {
+		for _, c := range m.clients.RecentMacs() {
 			if c.Login == user.Login {
 				mine = append(mine, c)
 			}

internal/clients/clients.go

@@ -33,6 +33,7 @@ type Client struct {
 	Login        string
 	Host         string
 	Commit       string
+	Device       string
 	ConnectedAt  time.Time
 	LastActionAt *time.Time
 	Sync         *api.ClientSync
@@ -109,7 +110,7 @@ func (r *Registry) Synced(id, login, host, commit string, sync api.ClientSync) {
 	r.entry(id, login, host, commit, time.Now()).Sync = &sync
 }
 
-func (r *Registry) Serve(ctx context.Context, ws *websocket.Conn, id, login, host, commit string) {
+func (r *Registry) Serve(ctx context.Context, ws *websocket.Conn, id, login, host, commit, device string) {
 	ws.SetReadLimit(readLimit)
 	c := &conn{ws: ws, done: make(chan struct{}), pending: map[string]chan api.ClientReply{}}
 
@@ -118,12 +119,13 @@ func (r *Registry) Serve(ctx context.Context, ws *websocket.Conn, id, login, hos
 	e := r.entry(id, login, host, commit, now)
 	old := e.conn
 	e.conn = c
+	e.Device = device
 	e.ConnectedAt = now
 	r.mu.Unlock()
 	if old != nil {
 		old.ws.Close(websocket.StatusGoingAway, "replaced by a newer connection from this client")
 	}
-	slog.Info("client connected", "id", id, "host", host, "login", login, "commit", commit)
+	slog.Info("client connected", "id", id, "host", host, "device", device, "login", login, "commit", commit)
 	r.changed()
 
 	defer func() {
@@ -219,6 +221,16 @@ func (r *Registry) Recent() []Client {
 	return rows
 }
 
+func (r *Registry) RecentMacs() []Client {
+	out := []Client{}
+	for _, c := range r.Recent() {
+		if c.Device == "Mac" {
+			out = append(out, c)
+		}
+	}
+	return out
+}
+
 func (r *Registry) Call(ctx context.Context, name string, req api.ClientRequest) (api.ClientReply, error) {
 	r.mu.Lock()
 	matches := []entry{}

internal/server/clients.go

@@ -35,7 +35,7 @@ func (s *Server) ConnectClient(w http.ResponseWriter, r *http.Request) {
 	}
 	defer ws.CloseNow()
 
-	s.clients.Serve(r.Context(), ws, id, actor(r), host, r.Header.Get("Byre-Commit"))
+	s.clients.Serve(r.Context(), ws, id, actor(r), host, r.Header.Get("Byre-Commit"), r.Header.Get("Byre-Device"))
 }
 
 func (s *Server) SetClientSync(w http.ResponseWriter, r *http.Request) {
@@ -84,7 +84,7 @@ func (s *Server) ExecClient(w http.ResponseWriter, r *http.Request) {
 		name = *req.Client
 	}
 	if name == "" {
-		recent := s.clients.Recent()
+		recent := s.clients.RecentMacs()
 		if len(recent) == 0 {
 			writeError(w, http.StatusServiceUnavailable, errors.New("no client Macs are connected"))
 			return

mac/Sources/Byre/API.swift

@@ -118,6 +118,7 @@ final class API: Sendable {
         config.httpAdditionalHeaders = [
             "Byre-Client": clientID,
             "Byre-Commit": Bundle.main.object(forInfoDictionaryKey: "ByreCommit") as! String,
+            "Byre-Device": deviceKind,
             "Byre-Host": deviceName,
         ]
         config.proxyConfigurations = [tailnetNode.proxy()]

mac/Sources/Byre/Pages.swift

@@ -278,6 +278,17 @@ final class Pages {
         }
         return png
     }
+    #else
+    func capture(_ view: WKWebView) async throws -> Data {
+        guard view.window != nil, !view.isHidden else {
+            throw ApiError(status: 0, message: "the tab isn't on screen: the human has the page view closed or another tab showing")
+        }
+        let image = try await view.takeSnapshot(configuration: nil)
+        guard let png = image.pngData() else {
+            throw ApiError(status: 0, message: "couldn't encode the capture as PNG")
+        }
+        return png
+    }
     #endif
 
     func close() {

mac/Sources/Byre/Remote.swift

@@ -7,12 +7,13 @@ typealias ClientRequest = Components.Schemas.ClientRequest
 typealias ClientReply = Components.Schemas.ClientReply
 typealias ClientSync = Components.Schemas.ClientSync
 
-#if os(macOS)
 private let retryAfter = Duration.seconds(2)
 private let checkEvery = Duration.seconds(10)
 private let staleAfter: TimeInterval = 25
-private let maxOutput = 64 << 10
 private let maxMessage = 256 << 20
+
+#if os(macOS)
+private let maxOutput = 64 << 10
 private let maxRead = 64 << 20
 private let copyTimeoutMs = 10 * 60 * 1000
 private let hands = Hands()
@@ -27,6 +28,7 @@ private let signScript = """
     esac
     exec "$program" --status-fd=2 -bsau "$(git config user.signingkey || echo "$1")" >"$2"
     """
+#endif
 
 @MainActor
 final class Remote {
@@ -157,6 +159,10 @@ final class Remote {
 
     private func perform(_ req: ClientRequest) async throws -> ClientReply {
         switch req.op {
+        #if os(iOS)
+        case .bash, .exec, .upload, .download, .readFile, .writeFile, .hands, .sign:
+            throw ApiError(status: 0, message: "an iPhone only shows and drives pages and files in byre's app (show_page, show_file, drive_page)")
+        #else
         case .bash:
             let home = FileManager.default.homeDirectoryForCurrentUser.path
             let (code, output) = try await spawn(["/bin/zsh", "-l", "-c", req.command ?? ""], in: home, timeoutMs: req.timeoutMs ?? 120_000)
@@ -190,6 +196,7 @@ final class Remote {
         case .sign:
             let (code, status, signature) = try await sign(Data(req.data?.data ?? []), as: req.ident ?? "", timeoutMs: req.timeoutMs ?? 120_000)
             return ClientReply(id: req.id, exitCode: code, output: status, data: Base64EncodedData(signature))
+        #endif
         case .openPage:
             if let address = req.url {
                 guard let url = URL(string: address) else {
@@ -219,6 +226,7 @@ final class Remote {
         }
     }
 
+    #if os(macOS)
     private func sign(_ payload: Data, as ident: String, timeoutMs: Int) async throws -> (Int, String, Data) {
         let fm = FileManager.default
         let dir = fm.temporaryDirectory.appending(path: "byre-sign-\(UUID().uuidString)", directoryHint: .isDirectory)
@@ -293,8 +301,10 @@ final class Remote {
         }
         try fm.moveItem(at: entries[0], to: target)
     }
+    #endif
 }
 
+#if os(macOS)
 private func expand(_ path: String) -> URL {
     URL(filePath: (path as NSString).expandingTildeInPath)
 }

mac/Sources/Byre/Startup.swift

@@ -254,13 +254,7 @@ final class Connection {
             store.startGlobalStream(after: status.lastEventSeq) { [weak self] in
                 self?.reconnected(status.startedAt)
             }
-            #if os(macOS)
             store.remote.start()
-            #else
-            if let sync = Servers.shared.sync {
-                store.reportSync(sync)
-            }
-            #endif
             forwards.sync(store.api, Set(store.activeTasks().map(\.task.id)))
             Notifier.shared.deliver()
             return true

mac/Sources/Byre/Store.swift

@@ -89,9 +89,7 @@ final class Store {
     let imageCache = ImageCache()
     let events = Listeners<ByreEvent>()
     let paneFocus: PaneFocus
-    #if os(macOS)
     let remote: Remote
-    #endif
     let forwards: Forwards
     let pages: Pages
 
@@ -115,10 +113,8 @@ final class Store {
         self.pages = pages
         leadFeed = Feed(shown: { true })
         paneFocus = PaneFocus(api: api)
-        #if os(macOS)
         remote = Remote(api: api)
         remote.store = self
-        #endif
         pages.popup = { [weak self] page in
             self?.openPage(page)
         }
@@ -144,9 +140,7 @@ final class Store {
         stopped = true
         globalStream?.stop()
         leadStream?.stop()
-        #if os(macOS)
         remote.stop()
-        #endif
         drafts?.flush()
         await paneFocus.leave()
     }