1package agent
2
3import (
4 "bufio"
5 "context"
6 "encoding/json"
7 "errors"
8 "fmt"
9 "io"
10 "os"
11 "strings"
12 "sync"
13 "time"
14
15 "github.com/bnema/zut/packages/agent/extensions"
16 "github.com/bnema/zut/packages/agent/extproto"
17 "github.com/bnema/zut/packages/agent/modes"
18 "github.com/bnema/zut/packages/agent/tools"
19 "github.com/bnema/zut/packages/core"
20 "github.com/bnema/zut/packages/provider"
21)
22
23// runRPCMode implements the JSON-over-stdin/stdout RPC protocol.
24//
25// Wire format: one JSON object per line in both directions.
26//
27// Commands (stdin):
28//
29// {"id":"1","type":"prompt","message":"hello","images":[]}
30// {"id":"2","type":"abort"}
31// {"id":"3","type":"compact"}
32// {"id":"4","type":"get_state"}
33// {"id":"5","type":"set_model","model":"claude-opus-4-5"}
34// {"id":"6","type":"set_reasoning","reasoning":"max"}
35// {"id":"6","type":"get_messages"}
36// {"id":"7","type":"clear"}
37// {"id":"8","type":"get_models"}
38//
39// Responses (stdout): {"type":"response","id":"1","command":"prompt","success":true}
40// Events (stdout): one JSON object per AgentEvent (same schema as --json mode).
41// Extensions may additionally emit ext_status with extension/key/level/text,
42// ext_widget with extension/id/position/title/lines, and ext_widget_clear
43// with extension/id when loaded extensions publish persistent UI state.
44//
45// Auth: if $ZUTCORE_RPC_TOKEN is set, the first command must be
46// {"type":"hello","token":"..."} or the connection is closed.
47func runRPCMode(ctx context.Context, args Args, version string) error {
48 if args.NoYolo {
49 fmt.Fprintln(os.Stderr, "warning: --no-yolo has no effect in rpc mode (no interactive prompt available); tools will run without confirmation")
50 }
51 r, err := Resolve(args, true)
52 if err != nil {
53 return err
54 }
55
56 // Construct the RPC event sink before extension startup. Alerts can
57 // arrive as soon as an extension is ready; rpcServer queues them until
58 // the command loop has started and, when configured, authenticated.
59 server := &rpcServer{
60 ctx: ctx,
61 args: args,
62 provider: r.Provider,
63 model: r.Model,
64 out: os.Stdout,
65 version: version,
66 }
67
68 // Extensions: same lifecycle as interactive mode, minus the
69 // host-hooks integration. Notify/Display calls from extensions
70 // emit RPC events instead of TUI lines so any consumer can react.
71 extHooks := &rpcExtHooks{server: server}
72 extMgr := extensions.New(ZutHome(), r.CWD, version, r.Provider, r.Model, extHooks)
73 for _, e := range extMgr.LoadExplicit(ctx, args.Exts) {
74 fmt.Fprintln(os.Stderr, "extension load:", e)
75 }
76 if !args.NoExt {
77 for _, e := range extMgr.Discover(ctx) {
78 fmt.Fprintln(os.Stderr, "extension load:", e)
79 }
80 }
81 extMgr.WaitForReady(3 * time.Second)
82 defer extMgr.Stop(2 * time.Second)
83 r.MergeExtensionTools(&extToolAdapter{mgr: extMgr})
84
85 ag := r.NewAgent()
86 defer closeAgentLSP(ag)
87 server.agent = ag
88 ag.BeforeToolExecute = func(call provider.ToolCallBlock) (bool, string, json.RawMessage) {
89 r := extMgr.InterceptToolCall(ctx, call.ID, call.Name, call.Arguments)
90 if r.Block {
91 return false, r.Reason, nil
92 }
93 return true, "", r.ModifiedArgs
94 }
95 ag.BeforeTurnContext = func(turnCtx context.Context, step int) (bool, string, string) {
96 r := extMgr.InterceptTurnStart(turnCtx, step)
97 return !r.Block, r.Reason, r.Context
98 }
99 ag.BeforeAssistantMessage = func(text string) (bool, string, string) {
100 r := extMgr.InterceptAssistantMessage(ctx, text)
101 if r.Block {
102 return false, r.Reason, ""
103 }
104 return true, "", r.ReplaceText
105 }
106 ag.OnEvent = func(ev core.AgentEvent) { fanoutAgentEvent(extMgr, ev) }
107
108 // /reload-ext hot-reload callback (also triggered via rpc
109 // `reload_ext` if/when added). Rebuilds the tool registry on the
110 // current agent so freshly-registered extension tools become
111 // callable without restarting the rpc process.
112 adapter := &extToolAdapter{mgr: extMgr}
113 extMgr.SetOnReload(func() {
114 resolved, err := Resolve(args, true)
115 if err != nil {
116 return
117 }
118 resolved.MergeExtensionTools(adapter)
119 oldTools := ag.SetPromptConfig(resolved.SystemPrompt, resolved.ToolRegistry)
120 if closeErr := tools.CloseLSPManagers(oldTools); closeErr != nil {
121 fmt.Fprintln(os.Stderr, "reload LSP cleanup:", closeErr)
122 }
123 })
124
125 extMgr.EmitSessionEvent("session_start", nil, nil)
126
127 return server.run(os.Stdin)
128}
129
130// rpcExtHooks implements extensions.HostHooks for the headless RPC
131// loop. Notify and Display surface as `event` frames so any RPC
132// client can render them; Submit and Insert are no-ops because the
133// RPC loop has no editor and the prompt comes from the client.
134type rpcExtHooks struct {
135 server *rpcServer
136}
137
138func (h *rpcExtHooks) Notify(extName, level, message string) {
139 if h.server != nil {
140 h.server.writeExtensionEvent(map[string]any{
141 "type": "ext_notify",
142 "extension": extName,
143 "level": level,
144 "message": message,
145 })
146 }
147}
148func (h *rpcExtHooks) Alert(extName string, alert extproto.AlertRequest) {
149 if h.server != nil {
150 h.server.writeExtensionEvent(map[string]any{
151 "type": "ext_alert",
152 "extension": extName,
153 "kind": alert.Kind,
154 "reason": alert.Reason,
155 })
156 }
157}
158func (h *rpcExtHooks) Display(extName, text string) {
159 if h.server != nil {
160 h.server.writeExtensionEvent(map[string]any{
161 "type": "ext_display",
162 "extension": extName,
163 "text": text,
164 })
165 }
166}
167func (h *rpcExtHooks) ClearNotes(extName string) {
168 if h.server != nil {
169 h.server.writeExtensionEvent(map[string]any{
170 "type": "ext_clear_notes",
171 "extension": extName,
172 })
173 }
174}
175func (h *rpcExtHooks) Submit(string) {} // ignored in rpc mode
176func (h *rpcExtHooks) SubmitSlash(string) {} // ignored in rpc mode
177func (h *rpcExtHooks) Insert(string) {} // ignored in rpc mode
178func (h *rpcExtHooks) OpenPanel(string, extproto.PanelSpec) {}
179func (h *rpcExtHooks) UpdatePanel(string, string, string, []string, string) {}
180func (h *rpcExtHooks) ClosePanel(string, string) {}
181func (h *rpcExtHooks) SetStatus(extName, key, level, text string) {
182 if h.server != nil {
183 h.server.writeExtensionEvent(map[string]any{
184 "type": "ext_status", "extension": extName, "key": key, "level": level, "text": text,
185 })
186 }
187}
188func (h *rpcExtHooks) SetWidget(extName, id, position, title string, lines []string) {
189 if h.server != nil {
190 h.server.writeExtensionEvent(map[string]any{
191 "type": "ext_widget", "extension": extName, "id": id, "position": position, "title": title, "lines": lines,
192 })
193 }
194}
195func (h *rpcExtHooks) ClearWidget(extName, id string) {
196 if h.server != nil {
197 h.server.writeExtensionEvent(map[string]any{
198 "type": "ext_widget_clear", "extension": extName, "id": id,
199 })
200 }
201}
202
203func (h *rpcExtHooks) ClearExtensionChrome(extName string) {
204 if h.server != nil {
205 h.server.writeExtensionEvent(map[string]any{
206 "type": "ext_chrome_clear", "extension": extName,
207 })
208 }
209}
210
211const maxPendingRPCExtEvents = 64
212
213type rpcServer struct {
214 ctx context.Context
215 args Args
216 agent *core.Agent
217 provider string
218 model string
219 out io.Writer
220 version string
221
222 writeMu sync.Mutex
223 turnMu sync.Mutex // serialises one prompt at a time
224 activeCancel context.CancelFunc
225 started bool
226 authed bool
227 pendingExtEvents [][]byte
228
229 // inFlight tracks long-running command goroutines so run() can
230 // wait for them before returning when stdin closes. Without this,
231 // piping a single 'prompt' command into 'zut rpc' would race the
232 // process exit against the agent loop and the prompt would never
233 // produce output.
234 inFlight sync.WaitGroup
235}
236
237// run reads NDJSON commands from in and dispatches them. Returns when
238// in is closed AND every in-flight long-running command (prompt /
239// compact) has finished, so a quick `echo cmd | zut rpc` invocation
240// still produces full output before the process exits.
241func (s *rpcServer) run(in io.Reader) error {
242 requireToken := os.Getenv("ZUTCORE_RPC_TOKEN") != ""
243 s.writeMu.Lock()
244 s.started = true
245 s.authed = !requireToken
246 if s.authed {
247 s.flushPendingExtEventsLocked()
248 }
249 s.writeMu.Unlock()
250
251 sc := bufio.NewScanner(in)
252 sc.Buffer(make([]byte, 0, 64*1024), 16*1024*1024)
253 for sc.Scan() {
254 line := strings.TrimSpace(sc.Text())
255 if line == "" {
256 continue
257 }
258 var head struct {
259 Type string `json:"type"`
260 ID string `json:"id,omitempty"`
261 }
262 if err := json.Unmarshal([]byte(line), &head); err != nil {
263 s.writeError("", "", fmt.Sprintf("malformed json: %v", err))
264 continue
265 }
266 if !s.isAuthed() {
267 if head.Type != "hello" {
268 s.writeError(head.ID, head.Type, "auth required: send hello with token first")
269 continue
270 }
271 var hello struct {
272 Token string `json:"token"`
273 }
274 _ = json.Unmarshal([]byte(line), &hello)
275 if hello.Token != os.Getenv("ZUTCORE_RPC_TOKEN") {
276 s.writeError(head.ID, head.Type, "invalid token")
277 return fmt.Errorf("rpc: bad auth token")
278 }
279 s.authenticate(head.ID, head.Type)
280 continue
281 }
282 s.dispatch(head.Type, head.ID, []byte(line))
283 }
284 err := sc.Err()
285 s.inFlight.Wait()
286 return err
287}
288
289func resolveRPCModel(providerName, modelID string) (provider.Model, error) {
290 model, err := provider.FindModel(providerName, modelID)
291 if err == nil {
292 return model, nil
293 }
294 if providerName == provider.ProviderOpenCodeGo && provider.AcceptsUnlistedModels(providerName) {
295 return provider.DynamicOpenCodeGoModel(providerName, modelID, ""), nil
296 }
297 return provider.Model{}, err
298}
299
300// dispatch routes a command. Long-running commands (prompt, compact)
301// run on their own goroutine so the read loop stays responsive.
302func (s *rpcServer) dispatch(cmd, id string, raw []byte) {
303 switch cmd {
304 case "hello":
305 s.writeResponse(id, cmd, map[string]any{
306 "protocol_version": 1,
307 "version": s.version,
308 "provider": s.provider,
309 "model": s.model,
310 })
311 case "prompt":
312 var req struct {
313 Message string `json:"message"`
314 Images []struct {
315 MimeType string `json:"mime_type"`
316 Data []byte `json:"data"`
317 } `json:"images"`
318 }
319 if err := json.Unmarshal(raw, &req); err != nil {
320 s.writeError(id, cmd, err.Error())
321 return
322 }
323 s.inFlight.Add(1)
324 go func() {
325 defer s.inFlight.Done()
326 s.runPrompt(id, req.Message, req.Images)
327 }()
328
329 case "abort":
330 if c := s.takeCancel(); c != nil {
331 c()
332 }
333 s.writeResponse(id, cmd, nil)
334
335 case "compact":
336 s.inFlight.Add(1)
337 go func() {
338 defer s.inFlight.Done()
339 s.runCompact(id)
340 }()
341
342 case "get_state":
343 s.writeResponse(id, cmd, s.snapshotState())
344
345 case "get_messages":
346 s.writeResponse(id, cmd, map[string]any{
347 "messages": messagesToJSON(s.agent.Messages()),
348 })
349
350 case "clear":
351 s.agent.SetMessages(nil)
352 s.agent.SetPlan(nil)
353 s.writeResponse(id, cmd, nil)
354
355 case "set_model":
356 if !s.turnMu.TryLock() {
357 s.writeError(id, cmd, "runtime is busy")
358 return
359 }
360 defer s.turnMu.Unlock()
361
362 var req struct {
363 Model string `json:"model"`
364 }
365 if err := json.Unmarshal(raw, &req); err != nil {
366 s.writeError(id, cmd, err.Error())
367 return
368 }
369 model := strings.TrimSpace(req.Model)
370 if model == "" {
371 s.writeError(id, cmd, "model must not be empty")
372 return
373 }
374 metadata, err := resolveRPCModel(s.provider, model)
375 if err != nil {
376 s.writeError(id, cmd, err.Error())
377 return
378 }
379 if setter, ok := s.agent.Client.(provider.ModelMetadataSetter); ok {
380 setter.SetModelMetadata(metadata)
381 }
382 if metadata.ContextWindow > 0 {
383 s.agent.ContextWindow = metadata.ContextWindow
384 }
385 if metadata.MaxOutput > 0 {
386 s.agent.MaxTokens = metadata.MaxOutput
387 }
388 s.agent.Model = model
389 s.model = model
390 s.writeResponse(id, cmd, map[string]any{"model": model})
391
392 case "set_reasoning":
393 var req struct {
394 Reasoning string `json:"reasoning"`
395 }
396 if err := json.Unmarshal(raw, &req); err != nil {
397 s.writeError(id, cmd, err.Error())
398 return
399 }
400 reasoning := provider.NormalizeReasoning(req.Reasoning)
401 switch reasoning {
402 case "", "minimum", "low", "medium", "high", "xhigh", "max":
403 default:
404 s.writeError(id, cmd, "reasoning must be off|minimum|low|medium|high|xhigh|max")
405 return
406 }
407 s.agent.Reasoning = reasoning
408 s.writeResponse(id, cmd, map[string]any{"reasoning": reasoning})
409
410 case "get_models":
411 out := []map[string]any{}
412 for _, m := range provider.ModelsForProvider(s.provider) {
413 out = append(out, map[string]any{
414 "id": m.ID,
415 "provider": m.Provider,
416 "context_window": m.ContextWindow,
417 "max_output": m.MaxOutput,
418 "reasoning": m.Reasoning,
419 })
420 }
421 s.writeResponse(id, cmd, map[string]any{"models": out})
422
423 case "ping":
424 s.writeResponse(id, cmd, map[string]any{"pong": true})
425
426 default:
427 s.writeError(id, cmd, "unknown command")
428 }
429}
430
431// runPrompt executes a single prompt turn and streams events out.
432// Holds turnMu so a second concurrent prompt blocks until this one
433// finishes; the user can abort with the abort command.
434func (s *rpcServer) runPrompt(id, message string, images []struct {
435 MimeType string `json:"mime_type"`
436 Data []byte `json:"data"`
437}) {
438 s.turnMu.Lock()
439 defer s.turnMu.Unlock()
440
441 subCtx, cancel := context.WithCancel(s.ctx)
442 s.setCancel(cancel)
443 defer s.setCancel(nil)
444
445 s.writeResponse(id, "prompt", map[string]any{"started": true})
446
447 imgs := make([]provider.ImageBlock, 0, len(images))
448 for _, im := range images {
449 imgs = append(imgs, provider.ImageBlock{MimeType: im.MimeType, Data: im.Data})
450 }
451
452 err := s.agent.Prompt(subCtx, message, imgs, func(ev core.AgentEvent) {
453 // EvDone is emitted by the agent loop and we re-emit our own
454 // 'done' below; suppressing it here avoids duplicate frames.
455 if _, ok := ev.(core.EvDone); ok {
456 return
457 }
458 s.writeEvent(modes.EventToJSON(ev))
459 })
460 // Don't emit a stand-alone error event for cancellation; the prior
461 // turn_end with stop=aborted already carries that signal.
462 if err != nil && !errors.Is(err, context.Canceled) {
463 s.writeEvent(map[string]any{"type": "error", "message": err.Error()})
464 }
465 s.writeEvent(map[string]any{"type": "done"})
466}
467
468// runCompact mirrors runPrompt but for compaction.
469func (s *rpcServer) runCompact(id string) {
470 s.turnMu.Lock()
471 defer s.turnMu.Unlock()
472
473 subCtx, cancel := context.WithCancel(s.ctx)
474 s.setCancel(cancel)
475 defer s.setCancel(nil)
476
477 s.writeResponse(id, "compact", map[string]any{"started": true})
478 summary, err := s.agent.CompactWithEvents(subCtx, 4, func(ev core.AgentEvent) {
479 s.writeEvent(modes.EventToJSON(ev))
480 })
481 if err != nil {
482 if !errors.Is(err, context.Canceled) {
483 s.writeEvent(map[string]any{"type": "error", "message": err.Error()})
484 }
485 return
486 }
487 s.writeEvent(map[string]any{
488 "type": "compact_done",
489 "summary": summary,
490 })
491}
492
493// snapshotState builds the get_state response.
494func (s *rpcServer) snapshotState() map[string]any {
495 cum := s.agent.Cost()
496 return map[string]any{
497 "provider": s.provider,
498 "model": s.model,
499 "reasoning": s.agent.Reasoning,
500 "cwd": s.args.CWD,
501 "message_count": len(s.agent.Messages()),
502 "busy": s.activeCancel != nil,
503 "usage": map[string]any{
504 "input": cum.InputTokens,
505 "output": cum.OutputTokens,
506 "reasoning": rpcReasoningTokens(cum),
507 "cache_read": cum.CacheReadTokens,
508 "cache_write": cum.CacheWriteTokens,
509 "cache_measured_prompt": cum.CacheMeasuredPromptTokens,
510 "cache_measured_read": cum.CacheMeasuredReadTokens,
511 "cost_usd": cum.CostUSD,
512 },
513 }
514}
515
516func rpcReasoningTokens(usage provider.Usage) any {
517 if !usage.ReasoningTokensKnown {
518 return nil
519 }
520 return usage.ReasoningTokens
521}
522
523// ---- write helpers (single-line JSON, mutex-guarded) ----
524
525func (s *rpcServer) writeResponse(id, cmd string, data any) {
526 frame := map[string]any{
527 "type": "response",
528 "command": cmd,
529 "success": true,
530 }
531 if id != "" {
532 frame["id"] = id
533 }
534 if data != nil {
535 frame["data"] = data
536 }
537 s.write(frame)
538}
539
540func (s *rpcServer) writeError(id, cmd, msg string) {
541 frame := map[string]any{
542 "type": "response",
543 "command": cmd,
544 "success": false,
545 "error": msg,
546 }
547 if id != "" {
548 frame["id"] = id
549 }
550 s.write(frame)
551}
552
553func (s *rpcServer) writeEvent(payload map[string]any) {
554 s.write(payload)
555}
556
557func (s *rpcServer) writeExtensionEvent(payload map[string]any) {
558 b, err := json.Marshal(payload)
559 if err != nil {
560 return
561 }
562 s.writeMu.Lock()
563 defer s.writeMu.Unlock()
564 if !s.started || !s.authed {
565 if len(s.pendingExtEvents) < maxPendingRPCExtEvents {
566 s.pendingExtEvents = append(s.pendingExtEvents, b)
567 }
568 return
569 }
570 s.writeLocked(b)
571}
572
573func (s *rpcServer) flushPendingExtEventsLocked() {
574 for _, b := range s.pendingExtEvents {
575 s.writeLocked(b)
576 }
577 s.pendingExtEvents = nil
578}
579
580func (s *rpcServer) isAuthed() bool {
581 s.writeMu.Lock()
582 defer s.writeMu.Unlock()
583 return s.authed
584}
585
586func (s *rpcServer) authenticate(id, cmd string) {
587 frame := map[string]any{
588 "type": "response",
589 "command": cmd,
590 "success": true,
591 "data": map[string]any{
592 "protocol_version": 1,
593 "version": s.version,
594 "provider": s.provider,
595 "model": s.model,
596 },
597 }
598 if id != "" {
599 frame["id"] = id
600 }
601 b, err := json.Marshal(frame)
602 if err != nil {
603 return
604 }
605 s.writeMu.Lock()
606 defer s.writeMu.Unlock()
607 s.authed = true
608 s.writeLocked(b)
609 s.flushPendingExtEventsLocked()
610}
611
612func (s *rpcServer) write(v any) {
613 b, err := json.Marshal(v)
614 if err != nil {
615 return
616 }
617 s.writeMu.Lock()
618 defer s.writeMu.Unlock()
619 s.writeLocked(b)
620}
621
622func (s *rpcServer) writeLocked(b []byte) {
623 _, _ = s.out.Write(b)
624 _, _ = s.out.Write([]byte("\n"))
625}
626
627func (s *rpcServer) setCancel(c context.CancelFunc) {
628 s.writeMu.Lock()
629 s.activeCancel = c
630 s.writeMu.Unlock()
631}
632
633func (s *rpcServer) takeCancel() context.CancelFunc {
634 s.writeMu.Lock()
635 defer s.writeMu.Unlock()
636 c := s.activeCancel
637 s.activeCancel = nil
638 return c
639}
640
641// messagesToJSON serialises a transcript using the same schema as the
642// --json event mode for cross-format consistency.
643func messagesToJSON(msgs []provider.Message) []map[string]any {
644 out := make([]map[string]any, 0, len(msgs))
645 for _, m := range msgs {
646 out = append(out, map[string]any{
647 "role": string(m.Role),
648 "content": modes.ContentToJSON(m.Content),
649 "time": m.Time,
650 })
651 }
652 return out
653}