forked from bots-garden/ori
| ✨ Introduce new feature(s): ACP web client — Go backend (agent, bridge, httpserver, mockagent) + React SPA (Zed-like agent panel), tests, quality gate PASS | 1 | // Package bridge relays events between the ACP agent session and the |
| 2 | // browsers connected over WebSocket. | |
| 3 | // | |
| 4 | // It is the meeting point of two asynchronous worlds: the agent streams | |
| 5 | // updates and blocks on permission requests; browsers connect, disconnect and | |
| 6 | // answer at human speed. The bridge broadcasts agent events to every client, | |
| 7 | // replays the session history to late joiners, and routes each permission | |
| 8 | // decision back to the exact request waiting for it. | |
| 9 | package bridge | |
| 10 | ||
| 11 | import ( | |
| 12 | "context" | |
| 13 | "encoding/json" | |
| 14 | "fmt" | |
| 15 | "log/slog" | |
| 16 | "strconv" | |
| 17 | "sync" | |
| 18 | ||
| 19 | acp "github.com/coder/acp-go-sdk" | |
| 20 | ) | |
| 21 | ||
| 22 | // Prompter is the slice of the agent session the bridge drives. It is | |
| 23 | // satisfied by *agent.Session and by test fakes. | |
| 24 | type Prompter interface { | |
| 25 | ID() string | |
| 26 | PromptText(ctx context.Context, text string) (acp.StopReason, error) | |
| 27 | Cancel(ctx context.Context) error | |
| 28 | } | |
| 29 | ||
| 30 | // maxHistory bounds the number of events replayed to late-joining clients; | |
| 31 | // beyond it the oldest events are dropped (the live broadcast is unaffected). | |
| 32 | const maxHistory = 4096 | |
| 33 | ||
| 34 | // clientBuffer is each subscriber's outgoing queue; a client too slow to | |
| 35 | // drain it is disconnected rather than allowed to stall the whole bridge. | |
| 36 | const clientBuffer = 256 | |
| 37 | ||
| 38 | type permissionDecision struct { | |
| 39 | optionId string | |
| 40 | cancelled bool | |
| 41 | } | |
| 42 | ||
| 43 | // Bridge broadcasts agent events to WebSocket clients and routes their | |
| 44 | // responses back to the agent. The zero value is not usable; call New. | |
| 45 | // | |
| 46 | // Example: | |
| 47 | // | |
| 48 | // b := bridge.New(slog.Default()) | |
| 49 | // session, _ := agent.Start(ctx, agent.Options{Handler: b, ...}) | |
| 50 | // b.SetSession(session) | |
| 51 | // http.Handle("GET /ws", b.WebSocketHandler()) | |
| 52 | type Bridge struct { | |
| 53 | logger *slog.Logger | |
| 54 | ||
| 55 | mu sync.Mutex | |
| 56 | session Prompter | |
| 57 | subscribers map[chan []byte]struct{} | |
| 58 | history [][]byte | |
| 59 | turnActive bool | |
| 60 | nextRequest int | |
| 61 | pendingPerms map[string]chan permissionDecision | |
| 62 | // pendingRequests keeps the encoded permission_request events so they | |
| 63 | // can be replayed to clients that connect while a request is open. | |
| 64 | pendingRequests map[string][]byte | |
| 65 | } | |
| 66 | ||
| 67 | // New creates an empty bridge; attach the agent session with SetSession. | |
| 68 | func New(logger *slog.Logger) *Bridge { | |
| 69 | if logger == nil { | |
| 70 | logger = slog.Default() | |
| 71 | } | |
| 72 | return &Bridge{ | |
| 73 | logger: logger, | |
| 74 | subscribers: make(map[chan []byte]struct{}), | |
| 75 | pendingPerms: make(map[string]chan permissionDecision), | |
| 76 | pendingRequests: make(map[string][]byte), | |
| 77 | } | |
| 78 | } | |
| 79 | ||
| 80 | // SetSession attaches the agent session the bridge drives. It must be called | |
| 81 | // before the first client message arrives. | |
| 82 | func (b *Bridge) SetSession(session Prompter) { | |
| 83 | b.mu.Lock() | |
| 84 | defer b.mu.Unlock() | |
| 85 | b.session = session | |
| 86 | } | |
| 87 | ||
| 88 | // Subscribe registers a new client. It returns the client's event channel, | |
| 89 | // the events to replay so the client catches up (history, then any pending | |
| 90 | // permission requests), and an unsubscribe function. | |
| 91 | func (b *Bridge) Subscribe() (events <-chan []byte, replay [][]byte, unsubscribe func()) { | |
| 92 | ch := make(chan []byte, clientBuffer) | |
| 93 | b.mu.Lock() | |
| 94 | defer b.mu.Unlock() | |
| 95 | ||
| 96 | b.subscribers[ch] = struct{}{} | |
| 97 | ||
| 98 | sessionId := "" | |
| 99 | if b.session != nil { | |
| 100 | sessionId = b.session.ID() | |
| 101 | } | |
| 102 | replay = append(replay, encode(Outgoing{Type: OutgoingHello, SessionId: sessionId, TurnActive: b.turnActive})) | |
| 103 | replay = append(replay, b.history...) | |
| 104 | for _, raw := range b.pendingRequests { | |
| 105 | replay = append(replay, raw) | |
| 106 | } | |
| 107 | ||
| 108 | return ch, replay, func() { | |
| 109 | b.mu.Lock() | |
| 110 | defer b.mu.Unlock() | |
| 111 | if _, ok := b.subscribers[ch]; ok { | |
| 112 | delete(b.subscribers, ch) | |
| 113 | close(ch) | |
| 114 | } | |
| 115 | } | |
| 116 | } | |
| 117 | ||
| 118 | // HandleIncoming processes one message from a client. Errors are reported to | |
| 119 | // the clients as OutgoingError events, never returned: a malformed message | |
| 120 | // from one browser must not tear down the connection handling. | |
| 121 | func (b *Bridge) HandleIncoming(ctx context.Context, raw []byte) { | |
| 122 | var msg Incoming | |
| 123 | if err := json.Unmarshal(raw, &msg); err != nil { | |
| 124 | b.broadcast(encode(Outgoing{Type: OutgoingError, Message: "malformed message: " + err.Error()}), false) | |
| 125 | return | |
| 126 | } | |
| 127 | ||
| 128 | switch msg.Type { | |
| 129 | case IncomingPrompt: | |
| 130 | b.startTurn(msg.Text) | |
| 131 | case IncomingCancel: | |
| 132 | b.cancelTurn(ctx) | |
| 133 | case IncomingPermissionResponse: | |
| 134 | b.resolvePermission(msg) | |
| 135 | default: | |
| 136 | b.broadcast(encode(Outgoing{Type: OutgoingError, Message: fmt.Sprintf("unknown message type %q", msg.Type)}), false) | |
| 137 | } | |
| 138 | } | |
| 139 | ||
| 140 | // startTurn launches a prompt turn in the background; updates stream through | |
| 141 | // HandleSessionUpdate while it runs. | |
| 142 | func (b *Bridge) startTurn(text string) { | |
| 143 | b.mu.Lock() | |
| 144 | session := b.session | |
| 145 | if session == nil { | |
| 146 | b.mu.Unlock() | |
| 147 | b.broadcast(encode(Outgoing{Type: OutgoingError, Message: "no agent session"}), false) | |
| 148 | return | |
| 149 | } | |
| 150 | if b.turnActive { | |
| 151 | b.mu.Unlock() | |
| 152 | b.broadcast(encode(Outgoing{Type: OutgoingError, Message: "a turn is already running"}), false) | |
| 153 | return | |
| 154 | } | |
| 155 | b.turnActive = true | |
| 156 | b.mu.Unlock() | |
| 157 | ||
| 158 | b.broadcast(encode(Outgoing{Type: OutgoingUserMessage, Text: text}), true) | |
| 159 | b.broadcast(encode(Outgoing{Type: OutgoingTurnStarted}), true) | |
| 160 | ||
| 161 | go func() { | |
| 162 | // The turn outlives the WebSocket message that started it, so it | |
| 163 | // runs under its own context, ended by session/cancel only. | |
| 164 | stop, err := session.PromptText(context.Background(), text) | |
| 165 | ||
| 166 | b.mu.Lock() | |
| 167 | b.turnActive = false | |
| 168 | b.mu.Unlock() | |
| 169 | ||
| 170 | if err != nil { | |
| 171 | b.logger.Error("prompt failed", "error", err) | |
| 172 | b.broadcast(encode(Outgoing{Type: OutgoingError, Message: "prompt failed: " + err.Error()}), true) | |
| 173 | b.broadcast(encode(Outgoing{Type: OutgoingTurnEnded, StopReason: string(acp.StopReasonRefusal)}), true) | |
| 174 | return | |
| 175 | } | |
| 176 | b.broadcast(encode(Outgoing{Type: OutgoingTurnEnded, StopReason: string(stop)}), true) | |
| 177 | }() | |
| 178 | } | |
| 179 | ||
| 180 | func (b *Bridge) cancelTurn(ctx context.Context) { | |
| 181 | b.mu.Lock() | |
| 182 | session := b.session | |
| 183 | b.mu.Unlock() | |
| 184 | if session == nil { | |
| 185 | return | |
| 186 | } | |
| 187 | if err := session.Cancel(ctx); err != nil { | |
| 188 | b.logger.Error("cancel failed", "error", err) | |
| 189 | b.broadcast(encode(Outgoing{Type: OutgoingError, Message: "cancel failed: " + err.Error()}), false) | |
| 190 | } | |
| 191 | } | |
| 192 | ||
| 193 | func (b *Bridge) resolvePermission(msg Incoming) { | |
| 194 | b.mu.Lock() | |
| 195 | waiting, ok := b.pendingPerms[msg.RequestId] | |
| 196 | if ok { | |
| 197 | delete(b.pendingPerms, msg.RequestId) | |
| 198 | delete(b.pendingRequests, msg.RequestId) | |
| 199 | } | |
| 200 | b.mu.Unlock() | |
| 201 | if !ok { | |
| 202 | // Either already answered by another client or unknown: ignore. | |
| 203 | return | |
| 204 | } | |
| 205 | waiting <- permissionDecision{optionId: msg.OptionId, cancelled: msg.Cancelled} | |
| 206 | b.broadcast(encode(Outgoing{Type: OutgoingPermissionResolved, RequestId: msg.RequestId}), true) | |
| 207 | } | |
| 208 | ||
| 209 | // HandleSessionUpdate implements agent.Handler: every ACP update is relayed | |
| 210 | // verbatim to the clients and recorded for replay. | |
| 211 | func (b *Bridge) HandleSessionUpdate(_ context.Context, notification acp.SessionNotification) { | |
| 212 | update, err := json.Marshal(notification.Update) | |
| 213 | if err != nil { | |
| 214 | b.logger.Error("marshal session update", "error", err) | |
| 215 | return | |
| 216 | } | |
| 217 | b.broadcast(encode(Outgoing{Type: OutgoingSessionUpdate, Update: update}), true) | |
| 218 | } | |
| 219 | ||
| 220 | // HandlePermissionRequest implements agent.Handler: the request is broadcast | |
| 221 | // to the clients and the call blocks until one of them answers or the agent | |
| 222 | // cancels the turn. | |
| 223 | func (b *Bridge) HandlePermissionRequest(ctx context.Context, request acp.RequestPermissionRequest) (acp.RequestPermissionResponse, error) { | |
| 224 | rawRequest, err := json.Marshal(request) | |
| 225 | if err != nil { | |
| 226 | return acp.RequestPermissionResponse{}, fmt.Errorf("marshal permission request: %w", err) | |
| 227 | } | |
| 228 | ||
| 229 | decisionCh := make(chan permissionDecision, 1) | |
| 230 | b.mu.Lock() | |
| 231 | b.nextRequest++ | |
| 232 | requestId := "perm-" + strconv.Itoa(b.nextRequest) | |
| 233 | event := encode(Outgoing{Type: OutgoingPermissionRequest, RequestId: requestId, Request: rawRequest}) | |
| 234 | b.pendingPerms[requestId] = decisionCh | |
| 235 | b.pendingRequests[requestId] = event | |
| 236 | b.mu.Unlock() | |
| 237 | ||
| 238 | b.broadcast(event, false) | |
| 239 | ||
| 240 | select { | |
| 241 | case decision := <-decisionCh: | |
| 242 | if decision.cancelled { | |
| 243 | return acp.RequestPermissionResponse{Outcome: acp.NewRequestPermissionOutcomeCancelled()}, nil | |
| 244 | } | |
| 245 | return acp.RequestPermissionResponse{Outcome: acp.NewRequestPermissionOutcomeSelected(acp.PermissionOptionId(decision.optionId))}, nil | |
| 246 | case <-ctx.Done(): | |
| 247 | // The agent cancelled the turn (or disconnected) while waiting. | |
| 248 | b.mu.Lock() | |
| 249 | delete(b.pendingPerms, requestId) | |
| 250 | delete(b.pendingRequests, requestId) | |
| 251 | b.mu.Unlock() | |
| 252 | b.broadcast(encode(Outgoing{Type: OutgoingPermissionResolved, RequestId: requestId}), true) | |
| 253 | return acp.RequestPermissionResponse{Outcome: acp.NewRequestPermissionOutcomeCancelled()}, nil | |
| 254 | } | |
| 255 | } | |
| 256 | ||
| 257 | // broadcast sends an event to every connected client, disconnecting any | |
| 258 | // client whose buffer is full. With record=true the event also joins the | |
| 259 | // replay history. | |
| 260 | func (b *Bridge) broadcast(event []byte, record bool) { | |
| 261 | b.mu.Lock() | |
| 262 | defer b.mu.Unlock() | |
| 263 | ||
| 264 | if record { | |
| 265 | b.history = append(b.history, event) | |
| 266 | if len(b.history) > maxHistory { | |
| 267 | b.history = b.history[len(b.history)-maxHistory:] | |
| 268 | } | |
| 269 | } | |
| 270 | ||
| 271 | for ch := range b.subscribers { | |
| 272 | select { | |
| 273 | case ch <- event: | |
| 274 | default: | |
| 275 | // The client stopped draining; drop it instead of blocking | |
| 276 | // the agent's event stream. | |
| 277 | delete(b.subscribers, ch) | |
| 278 | close(ch) | |
| 279 | b.logger.Warn("dropped a slow websocket client") | |
| 280 | } | |
| 281 | } | |
| 282 | } |