1// Package extensions implements the host side of zot's subprocess
2// extension protocol. The Manager discovers extensions in well-known
3// directories, spawns each one, completes the hello handshake, and
4// routes slash commands to the right extension.
5//
6// Each extension is its own process, communicating with zot over its
7// stdin/stdout in newline-delimited JSON. Stderr is redirected to a
8// per-extension log file under $ZOT_HOME/logs/. Crashing one
9// extension does not affect the others or the host.
10//
11// See docs/extensions.md for the user-facing reference and
12// packages/agent/extproto for the wire format.
13package extensions
14
15import (
16 "bufio"
17 "context"
18 "encoding/json"
19 "errors"
20 "fmt"
21 "io"
22 "os"
23 "os/exec"
24 "path/filepath"
25 "sort"
26 "strings"
27 "sync"
28 "syscall"
29 "time"
30
31 "github.com/patriceckhart/zot/packages/agent/extproto"
32 "github.com/patriceckhart/zot/packages/tui"
33)
34
35// Manifest is the extension.json file shipped alongside an
36// extension's executable. It tells zot how to launch the extension
37// and provides display metadata.
38type Manifest struct {
39 Name string `json:"name"`
40 Version string `json:"version,omitempty"`
41 Exec string `json:"exec"` // executable path, relative to manifest dir
42 Args []string `json:"args,omitempty"` // extra argv passed to exec
43 Language string `json:"language,omitempty"` // informational ("go", "python", "typescript", ...)
44 Enabled *bool `json:"enabled,omitempty"` // nil = enabled
45 Description string `json:"description,omitempty"`
46}
47
48// IsEnabled returns the manifest's effective enabled state. Default
49// is true so adding a new extension folder Just Works without an
50// extra zot ext enable command.
51func (m Manifest) IsEnabled() bool {
52 return m.Enabled == nil || *m.Enabled
53}
54
55// Extension is a running extension subprocess and the metadata zot
56// tracks about it.
57type Extension struct {
58 Manifest Manifest
59 Dir string // absolute path to extension directory
60 LogPath string
61
62 cmd *exec.Cmd
63 stdin io.WriteCloser
64 stdout io.ReadCloser
65 logFile *os.File
66 helloAck bool
67 commands []extproto.RegisterCommandFromExt
68 tools []extproto.RegisterToolFromExt
69
70 // readyCh is closed when the extension sends a ReadyFromExt
71 // frame, or when the host gives up waiting (registrationGrace).
72 readyCh chan struct{}
73 readyOnce sync.Once
74
75 // pending command invocations waiting on a CommandResponseFromExt
76 // keyed by the id we sent in CommandInvokedFromHost.
77 // pendingTool is the same idea for tool calls.
78 // pendingIntercept is the same idea for event_intercept calls.
79 mu sync.Mutex
80 pending map[string]chan extproto.CommandResponseFromExt
81 pendingTool map[string]chan extproto.ToolResultFromExt
82 pendingIntercept map[string]chan extproto.EventInterceptResponseFromExt
83
84 // lastFrameTime is updated by the read loop on every frame it
85 // processes. Used by the auto-ready idle watchdog so legacy
86 // extensions (no `ready` frame) don't pin the WaitForReady wait
87 // to its full grace.
88 lastFrameTime time.Time
89
90 // eventSubs and interceptSubs are the sets of event names this
91 // extension subscribed to via SubscribeFromExt. Used by
92 // EmitEvent / InterceptToolCall to filter recipients.
93 eventSubs map[string]struct{}
94 interceptSubs map[string]struct{}
95
96 autoReady bool
97 readyTimedOut bool
98 diagnostics []string
99}
100
101// HostHooks is the small interface the manager calls back into the
102// running TUI through. Decouples extensions from interactive.go.
103type HostHooks interface {
104 // Notify pushes an ext-originated status message into the chat.
105 // Level is one of "info", "warn", "error", "success".
106 Notify(extName, level, message string)
107
108 // Submit feeds text as if the user had typed and pressed enter,
109 // running it through the agent loop.
110 Submit(text string)
111
112 // SubmitSlash runs text as a slash command in the TUI as if the
113 // user had typed it (text must start with '/'). Unlike Submit it
114 // does NOT run text through the model. Wired to the spontaneous
115 // submit_slash frame from extensions; ignored when the host is
116 // not interactive.
117 SubmitSlash(text string)
118
119 // Insert places text at the cursor in the editor.
120 Insert(text string)
121
122 // Display appends a one-shot styled note to the chat without
123 // invoking the model and without writing to the transcript.
124 Display(extName, text string)
125
126 // ClearNotes removes any notes previously pushed by extName via
127 // Notify/Display so transient status lines do not stack forever.
128 ClearNotes(extName string)
129
130 OpenPanel(extName string, spec extproto.PanelSpec)
131 UpdatePanel(extName, panelID, title string, lines []string, footer string)
132 ClosePanel(extName, panelID string)
133}
134
135type commandRegistration struct {
136 ext *Extension
137 name string
138}
139
140// Manager owns every extension subprocess for the lifetime of zot.
141type Manager struct {
142 zotHome string
143 cwd string
144 zotVersion string
145 provider string
146 model string
147 hooks HostHooks
148
149 mu sync.RWMutex
150 ext map[string]*Extension // keyed by manifest name
151
152 // commandIndex maps a lower-cased slash-command name (without the
153 // leading /) to its canonical registration. First-come-first-served:
154 // later registrations that differ only by case are dropped with a warning.
155 commandIndex map[string]commandRegistration
156
157 // toolIndex maps an extension-defined tool name to its owning
158 // extension. Same first-come-first-served rule as commandIndex.
159 toolIndex map[string]*Extension
160
161 // explicitPaths remembers ad-hoc paths passed via --ext so
162 // Reload can respawn them alongside the discovered set.
163 explicitPaths []string
164
165 // onReload, if set, is invoked after a successful Reload. Used
166 // by the host so it can rebuild the agent's tool registry with
167 // the freshly-registered extension tools.
168 onReload func()
169}
170
171// New constructs an empty Manager. Call Discover to populate it from
172// the on-disk extension directories.
173func New(zotHome, cwd, zotVersion, provider, model string, hooks HostHooks) *Manager {
174 return &Manager{
175 zotHome: zotHome,
176 cwd: cwd,
177 zotVersion: zotVersion,
178 provider: provider,
179 model: model,
180 hooks: hooks,
181 ext: map[string]*Extension{},
182 commandIndex: map[string]commandRegistration{},
183 toolIndex: map[string]*Extension{},
184 }
185}
186
187// Discover scans the global and project extension dirs and starts
188// every extension whose manifest is enabled. Spawns happen in
189// parallel so a slow runtime (e.g. `npx tsx` cold-start, ~1.5s)
190// doesn't block other extensions from starting. Returns a slice of
191// errors encountered (one per extension); a single bad extension
192// doesn't abort the rest.
193func (m *Manager) Discover(ctx context.Context) []error {
194 type loadJob struct {
195 dir string
196 }
197 var jobs []loadJob
198 seenDirs := map[string]bool{} // dedup by basename so project wins
199 for _, dir := range m.searchDirs() {
200 entries, err := os.ReadDir(dir)
201 if err != nil {
202 continue // missing directory is fine
203 }
204 for _, e := range entries {
205 if !e.IsDir() {
206 continue
207 }
208 if seenDirs[e.Name()] {
209 continue // higher-priority location already queued
210 }
211 seenDirs[e.Name()] = true
212 jobs = append(jobs, loadJob{dir: filepath.Join(dir, e.Name())})
213 }
214 }
215
216 var wg sync.WaitGroup
217 errCh := make(chan error, len(jobs))
218 for _, j := range jobs {
219 wg.Add(1)
220 go func(extDir string) {
221 defer wg.Done()
222 if err := m.loadOne(ctx, extDir); err != nil {
223 errCh <- fmt.Errorf("%s: %w", extDir, err)
224 }
225 }(j.dir)
226 }
227 wg.Wait()
228 close(errCh)
229
230 var errs []error
231 for e := range errCh {
232 errs = append(errs, e)
233 }
234 return errs
235}
236
237// searchDirs returns the directories the discoverer walks, in
238// priority order: project-local first (so a project can override
239// global behavior), then global.
240func (m *Manager) searchDirs() []string {
241 var dirs []string
242 if m.cwd != "" {
243 dirs = append(dirs, filepath.Join(m.cwd, ".zot", "extensions"))
244 }
245 if m.zotHome != "" {
246 dirs = append(dirs, filepath.Join(m.zotHome, "extensions"))
247 }
248 return dirs
249}
250
251// loadOne reads a single extension's manifest and, if enabled,
252// spawns its subprocess + completes the hello handshake.
253func (m *Manager) loadOne(ctx context.Context, dir string) error {
254 manifestPath := filepath.Join(dir, "extension.json")
255 raw, err := os.ReadFile(manifestPath)
256 if err != nil {
257 return fmt.Errorf("read manifest: %w", err)
258 }
259 var mf Manifest
260 if err := json.Unmarshal(raw, &mf); err != nil {
261 return fmt.Errorf("parse manifest: %w", err)
262 }
263 if mf.Name == "" {
264 return errors.New("manifest: name is required")
265 }
266 hasTheme := HasExtensionTheme(dir)
267 if mf.Exec == "" && !hasTheme {
268 return errors.New("manifest: exec is required")
269 }
270 if !mf.IsEnabled() {
271 // Quietly skip disabled extensions; zot ext list will show them.
272 return nil
273 }
274
275 m.mu.RLock()
276 _, dup := m.ext[mf.Name]
277 m.mu.RUnlock()
278 if dup {
279 // Project-local copy already won; ignore the global one.
280 return nil
281 }
282
283 ext := &Extension{
284 Manifest: mf,
285 Dir: dir,
286 readyCh: make(chan struct{}),
287 pending: map[string]chan extproto.CommandResponseFromExt{},
288 pendingTool: map[string]chan extproto.ToolResultFromExt{},
289 pendingIntercept: map[string]chan extproto.EventInterceptResponseFromExt{},
290 eventSubs: map[string]struct{}{},
291 interceptSubs: map[string]struct{}{},
292 }
293 if mf.Exec != "" {
294 if err := m.spawn(ctx, ext); err != nil {
295 return err
296 }
297 } else {
298 ext.readyOnce.Do(func() { close(ext.readyCh) })
299 }
300
301 m.mu.Lock()
302 m.ext[mf.Name] = ext
303 // Note: ext.commands and ext.tools may be empty here — they're
304 // populated by the read loop as register_* frames arrive after
305 // hello. Indexing happens in the read loop too. Discover()'s
306 // caller can WaitForReady() before relying on the registries.
307 m.mu.Unlock()
308 return nil
309}
310
311// LoadExplicit loads each path as an ad-hoc extension. Used for
312// `zot --ext <path>` so extension authors can iterate on a working
313// copy without having to `zot ext install` after every change.
314//
315// Loaded BEFORE Discover so explicit paths win on name conflicts
316// against installed extensions. Spawns happen in parallel like the
317// regular discovery path; errors are returned per path.
318// HasExtensionTheme reports whether an extension directory contains a theme
319// file in one of the supported extension-owned locations.
320func HasExtensionTheme(dir string) bool {
321 for _, file := range []string{"theme.json", filepath.Join("themes", "theme.json")} {
322 if _, err := os.Stat(filepath.Join(dir, file)); err == nil {
323 return true
324 }
325 }
326 return false
327}
328
329func (m *Manager) ThemeOptions() []tui.ThemeOption {
330 m.mu.RLock()
331 defer m.mu.RUnlock()
332 names := make([]string, 0, len(m.ext))
333 for name := range m.ext {
334 names = append(names, name)
335 }
336 sort.Strings(names)
337 var out []tui.ThemeOption
338 for _, name := range names {
339 ext := m.ext[name]
340 for _, file := range []string{"theme.json", "themes/theme.json"} {
341 path := filepath.Join(ext.Dir, file)
342 value := path
343 if opt, ok := tui.ThemeOptionFromFile(path, value, "extension "+ext.Manifest.Name); ok {
344 out = append(out, opt)
345 }
346 }
347 }
348 return out
349}
350
351func (m *Manager) LoadExplicit(ctx context.Context, paths []string) []error {
352 if len(paths) == 0 {
353 return nil
354 }
355
356 var wg sync.WaitGroup
357 errCh := make(chan error, len(paths))
358 absPaths := make([]string, 0, len(paths))
359 for _, p := range paths {
360 abs, err := filepath.Abs(p)
361 if err != nil {
362 errCh <- fmt.Errorf("%s: %w", p, err)
363 continue
364 }
365 absPaths = append(absPaths, abs)
366 wg.Add(1)
367 go func(extDir string) {
368 defer wg.Done()
369 if err := m.loadOne(ctx, extDir); err != nil {
370 errCh <- fmt.Errorf("%s: %w", extDir, err)
371 }
372 }(abs)
373 }
374 wg.Wait()
375 close(errCh)
376
377 m.mu.Lock()
378 m.explicitPaths = append(m.explicitPaths, absPaths...)
379 m.mu.Unlock()
380
381 var errs []error
382 for e := range errCh {
383 errs = append(errs, e)
384 }
385 return errs
386}
387
388// SetOnReload registers a callback fired after a successful Reload.
389// Hosts use it to rebuild the agent's tool registry with freshly-
390// registered extension tools.
391func (m *Manager) SetOnReload(fn func()) {
392 m.mu.Lock()
393 m.onReload = fn
394 m.mu.Unlock()
395}
396
397// ReloadStats summarises the outcome of Reload.
398type ReloadStats struct {
399 Stopped int // how many old processes were torn down
400 Loaded int // how many new processes reached spawn
401 Ready int // how many of those signalled ready in time
402 Errors []error // non-fatal per-extension errors
403}
404
405// Reload tears down every running extension, re-reads the manifests
406// from disk, respawns everyone (including the --ext paths remembered
407// from LoadExplicit), waits up to grace for ready signals, and
408// invokes the SetOnReload callback so the host can rebuild its tool
409// registry. The manager's internal maps are cleared before the
410// new load to ensure a clean slate.
411//
412// Safe to call concurrently with normal host operations: the lock is
413// released between stop and respawn so pending InvokeTool / Invoke
414// calls on the old processes get a clean error as their stdin
415// closes.
416func (m *Manager) Reload(ctx context.Context, grace time.Duration) ReloadStats {
417 stats := ReloadStats{}
418
419 // Snapshot and remember the explicit paths before we wipe state.
420 m.mu.Lock()
421 old := m.ext
422 explicit := append([]string(nil), m.explicitPaths...)
423 stats.Stopped = len(old)
424 m.ext = map[string]*Extension{}
425 m.commandIndex = map[string]commandRegistration{}
426 m.toolIndex = map[string]*Extension{}
427 m.explicitPaths = nil
428 callback := m.onReload
429 m.mu.Unlock()
430
431 // Graceful stop of the old set (reuses Stop's shutdown logic,
432 // but Stop re-reads m.ext which is now empty, so we replicate
433 // the small shutdown loop here on the snapshot).
434 oldExts := make([]*Extension, 0, len(old))
435 for _, ext := range old {
436 oldExts = append(oldExts, ext)
437 }
438 stopExtensions(oldExts, grace)
439
440 // Fresh load. Explicit paths first (they still win on conflict).
441 if errs := m.LoadExplicit(ctx, explicit); len(errs) > 0 {
442 stats.Errors = append(stats.Errors, errs...)
443 }
444 if errs := m.Discover(ctx); len(errs) > 0 {
445 stats.Errors = append(stats.Errors, errs...)
446 }
447
448 m.mu.RLock()
449 stats.Loaded = len(m.ext)
450 m.mu.RUnlock()
451
452 // Wait for ready frames. Use the same 3s grace zot uses at
453 // startup so the reload feels no slower than a cold boot.
454 readyDeadline := time.Now().Add(grace)
455 if time.Until(readyDeadline) < 3*time.Second {
456 readyDeadline = time.Now().Add(3 * time.Second)
457 }
458 m.WaitForReady(time.Until(readyDeadline))
459
460 m.mu.RLock()
461 for _, ext := range m.ext {
462 select {
463 case <-ext.readyCh:
464 stats.Ready++
465 default:
466 }
467 }
468 m.mu.RUnlock()
469
470 if callback != nil {
471 callback()
472 }
473
474 return stats
475}
476
477// WaitForReady blocks until every loaded extension has signalled
478// ReadyFromExt, or the grace period expires for the slowest one.
479//
480// Waits run in parallel: total time is max(per-extension wait), not
481// sum. Without this, a single slow extension (e.g. `npx tsx` cold)
482// would gate every other extension's wait too and zot startup would
483// scale linearly with the number of slow runtimes installed.
484//
485// Call after Discover and before relying on tool registrations.
486func (m *Manager) WaitForReady(grace time.Duration) {
487 m.mu.RLock()
488 exts := make([]*Extension, 0, len(m.ext))
489 for _, e := range m.ext {
490 exts = append(exts, e)
491 }
492 m.mu.RUnlock()
493
494 deadline := time.After(grace)
495 var wg sync.WaitGroup
496 for _, ext := range exts {
497 wg.Add(1)
498 go func(ext *Extension) {
499 defer wg.Done()
500 select {
501 case <-ext.readyCh:
502 case <-deadline:
503 ext.mu.Lock()
504 ext.readyTimedOut = true
505 ext.diagnostics = append(ext.diagnostics, "timed out waiting for ready frame")
506 ext.mu.Unlock()
507 fmt.Fprintf(ext.logFile, "[zot] timed out waiting for ready frame; proceeding\n")
508 ext.readyOnce.Do(func() { close(ext.readyCh) })
509 }
510 }(ext)
511 }
512 wg.Wait()
513}
514
515// spawn launches the subprocess, hooks up pipes, logs stderr, and
516// runs the synchronous portion of the hello handshake. Asynchronous
517// frames are processed in a goroutine started here.
518func (m *Manager) spawn(ctx context.Context, ext *Extension) error {
519 logsDir := filepath.Join(m.zotHome, "logs")
520 _ = os.MkdirAll(logsDir, 0o755)
521 logPath := filepath.Join(logsDir, "ext-"+ext.Manifest.Name+".log")
522 logFile, err := os.OpenFile(logPath, os.O_CREATE|os.O_APPEND|os.O_WRONLY, 0o644)
523 if err != nil {
524 return fmt.Errorf("open log: %w", err)
525 }
526 ext.LogPath = logPath
527 ext.logFile = logFile
528 fmt.Fprintf(logFile, "\n[zot] starting %s/%s at %s\n", ext.Manifest.Name, ext.Manifest.Version, time.Now().Format(time.RFC3339))
529
530 // Exec resolution rules:
531 // - absolute path: used as-is.
532 // - starts with "." (./ or ../): resolved relative to ext.Dir.
533 // - bare name (no path separator): looked up via $PATH so
534 // "node", "npx", "python3",
535 // "tsx" etc. work without
536 // forcing absolute paths.
537 // - other relative form (foo/bar): resolved relative to ext.Dir.
538 execPath := ext.Manifest.Exec
539 switch {
540 case filepath.IsAbs(execPath):
541 // keep
542 case strings.HasPrefix(execPath, "."+string(filepath.Separator)) ||
543 strings.HasPrefix(execPath, ".."+string(filepath.Separator)) ||
544 execPath == "." || execPath == "..":
545 execPath = filepath.Join(ext.Dir, execPath)
546 case strings.ContainsRune(execPath, filepath.Separator):
547 execPath = filepath.Join(ext.Dir, execPath)
548 default:
549 // bare name: leave as-is for exec.LookPath via exec.Command.
550 }
551 cmd := exec.CommandContext(ctx, execPath, ext.Manifest.Args...)
552 cmd.Dir = ext.Dir
553 cmd.Stderr = logFile
554 isolateExtensionProcess(cmd)
555
556 var stdin io.WriteCloser
557 var stdout io.ReadCloser
558 started := false
559 cleanup := true
560 defer func() {
561 if !cleanup {
562 return
563 }
564 if stdin != nil {
565 _ = stdin.Close()
566 }
567 if stdout != nil {
568 _ = stdout.Close()
569 }
570 if started {
571 _ = cmd.Process.Kill()
572 _ = cmd.Wait()
573 }
574 _ = logFile.Close()
575 }()
576
577 stdin, err = cmd.StdinPipe()
578 if err != nil {
579 return fmt.Errorf("stdin pipe (stderr log: %s): %w", logPath, err)
580 }
581 stdout, err = cmd.StdoutPipe()
582 if err != nil {
583 return fmt.Errorf("stdout pipe (stderr log: %s): %w", logPath, err)
584 }
585 if err := cmd.Start(); err != nil {
586 return fmt.Errorf("spawn (stderr log: %s): %w", logPath, err)
587 }
588 started = true
589 ext.cmd = cmd
590 ext.stdin = stdin
591 ext.stdout = stdout
592
593 // Hello handshake. Read the extension's HelloFromExt synchronously
594 // so we can fail fast on a broken extension; everything after is
595 // processed in the read goroutine.
596 scanner := bufio.NewScanner(stdout)
597 scanner.Buffer(make([]byte, 0, 64*1024), 4*1024*1024)
598 if !scanner.Scan() {
599 if err := scanner.Err(); err != nil {
600 return fmt.Errorf("read hello (stderr log: %s): %w", logPath, err)
601 }
602 return fmt.Errorf("extension exited before hello (stderr log: %s)", logPath)
603 }
604 var hello extproto.HelloFromExt
605 if err := json.Unmarshal(scanner.Bytes(), &hello); err != nil {
606 return fmt.Errorf("parse hello (stderr log: %s): %w", logPath, err)
607 }
608 if hello.Type != "hello" || hello.Name == "" {
609 return fmt.Errorf("first frame must be hello (got %q; stderr log: %s)", hello.Type, logPath)
610 }
611 // Trust the manifest's name; ignore mismatch from the hello.
612 ext.helloAck = true
613
614 ack, _ := extproto.Encode(extproto.HelloAckFromHost{
615 Type: "hello_ack",
616 ProtocolVersion: extproto.ProtocolVersion,
617 ZotVersion: m.zotVersion,
618 Provider: m.provider,
619 Model: m.model,
620 CWD: m.cwd,
621 ExtensionDir: ext.Dir,
622 DataDir: ext.Dir,
623 })
624 if _, err := stdin.Write(ack); err != nil {
625 return fmt.Errorf("send hello_ack (stderr log: %s): %w", logPath, err)
626 }
627
628 cleanup = false
629
630 // Spin up the read loop now that the handshake is done.
631 go m.readLoop(ext, scanner)
632
633 // Compatibility shim: extensions built against the phase-1 SDK
634 // don't send a ready frame. Watch the read loop's frame arrival
635 // rate; if nothing's arrived for readyIdleWindow we treat the
636 // extension as ready so WaitForReady doesn't burn the full grace
637 // on every startup. Newer extensions still trigger the explicit
638 // path on their own ready frame.
639 go m.assumeReadyAfterIdle(ext)
640
641 return nil
642}
643
644// readyIdleWindow is how long the manager waits for a frame after
645// hello before assuming an extension that doesn't send `ready` is
646// nevertheless ready. 250ms is enough for any well-behaved native
647// binary to flush its register frames; slow runtimes (npx tsx) flush
648// even faster once they've started, so this rarely affects them.
649const readyIdleWindow = 250 * time.Millisecond
650
651func (m *Manager) assumeReadyAfterIdle(ext *Extension) {
652 ext.mu.Lock()
653 last := ext.lastFrameTime
654 ext.mu.Unlock()
655 for {
656 select {
657 case <-ext.readyCh:
658 return
659 case <-time.After(readyIdleWindow):
660 }
661 ext.mu.Lock()
662 current := ext.lastFrameTime
663 ext.mu.Unlock()
664 if current.Equal(last) {
665 // No new frame in the idle window. Treat as ready.
666 ext.readyOnce.Do(func() {
667 ext.mu.Lock()
668 ext.autoReady = true
669 ext.diagnostics = append(ext.diagnostics, "no ready frame; auto-ready after idle")
670 ext.mu.Unlock()
671 fmt.Fprintf(ext.logFile, "[zot] no ready frame; auto-readying after idle (legacy SDK?)\n")
672 close(ext.readyCh)
673 })
674 return
675 }
676 last = current
677 }
678}
679
680// readLoop processes every frame the extension sends after hello.
681// Returns when stdout closes.
682func (m *Manager) readLoop(ext *Extension, scanner *bufio.Scanner) {
683 defer func() {
684 // On close, drop every command + tool this extension owned so
685 // future invocations don't dangle. The subprocess is gone; we
686 // won't hear back about its commands or tool calls anymore.
687 m.mu.Lock()
688 for name, registration := range m.commandIndex {
689 if registration.ext == ext {
690 delete(m.commandIndex, name)
691 }
692 }
693 for name, owner := range m.toolIndex {
694 if owner == ext {
695 delete(m.toolIndex, name)
696 }
697 }
698 m.mu.Unlock()
699 ext.readyOnce.Do(func() { close(ext.readyCh) })
700 fmt.Fprintf(ext.logFile, "[zot] extension %s read loop exited at %s\n", ext.Manifest.Name, time.Now().Format(time.RFC3339))
701 }()
702
703 for scanner.Scan() {
704 line := scanner.Bytes()
705 ext.mu.Lock()
706 ext.lastFrameTime = time.Now()
707 ext.mu.Unlock()
708 var frame extproto.Frame
709 if err := json.Unmarshal(line, &frame); err != nil {
710 ext.recordDiagnostic(fmt.Sprintf("malformed json frame: %v", err))
711 fmt.Fprintf(ext.logFile, "[zot] malformed json from extension: %v\n", err)
712 continue
713 }
714 switch frame.Type {
715 case "register_command":
716 var rc extproto.RegisterCommandFromExt
717 if err := json.Unmarshal(line, &rc); err == nil {
718 // Both ext.commands and m.commandIndex are read by
719 // the public Commands() / HasCommand() helpers under
720 // m.mu, so the writes have to take the same lock to
721 // keep the race detector happy.
722 m.mu.Lock()
723 ext.commands = append(ext.commands, rc)
724 key := strings.ToLower(rc.Name)
725 _, shadowed := m.commandIndex[key]
726 if !shadowed {
727 m.commandIndex[key] = commandRegistration{ext: ext, name: rc.Name}
728 }
729 m.mu.Unlock()
730 if shadowed {
731 ext.recordDiagnostic(fmt.Sprintf("command /%s conflicts with another extension registering the same name", rc.Name))
732 }
733 }
734 case "register_tool":
735 var rt extproto.RegisterToolFromExt
736 if err := json.Unmarshal(line, &rt); err != nil {
737 fmt.Fprintf(ext.logFile, "[zot] bad register_tool frame: %v\n", err)
738 continue
739 }
740 // Validate the schema parses as JSON. If not, refuse to
741 // register — a broken schema confuses the model.
742 if len(rt.Schema) > 0 {
743 var tmp any
744 if err := json.Unmarshal(rt.Schema, &tmp); err != nil {
745 ext.recordDiagnostic(fmt.Sprintf("tool %q schema is not valid json: %v", rt.Name, err))
746 fmt.Fprintf(ext.logFile, "[zot] tool %q: schema is not valid json (%v); skipped\n", rt.Name, err)
747 continue
748 }
749 }
750 m.mu.Lock()
751 ext.tools = append(ext.tools, rt)
752 _, shadowed := m.toolIndex[rt.Name]
753 if !shadowed {
754 m.toolIndex[rt.Name] = ext
755 }
756 m.mu.Unlock()
757 if shadowed {
758 ext.recordDiagnostic(fmt.Sprintf("tool %s conflicts with another extension registering the same name", rt.Name))
759 }
760 case "ready":
761 ext.readyOnce.Do(func() { close(ext.readyCh) })
762 case "subscribe":
763 var sub extproto.SubscribeFromExt
764 if err := json.Unmarshal(line, &sub); err == nil {
765 ext.mu.Lock()
766 for _, ev := range sub.Events {
767 ext.eventSubs[ev] = struct{}{}
768 }
769 for _, ev := range sub.Intercept {
770 switch ev {
771 case "tool_call", "turn_start", "assistant_message":
772 ext.interceptSubs[ev] = struct{}{}
773 }
774 }
775 ext.mu.Unlock()
776 }
777 case "event_intercept_response":
778 var er extproto.EventInterceptResponseFromExt
779 if err := json.Unmarshal(line, &er); err == nil {
780 ext.mu.Lock()
781 ch, ok := ext.pendingIntercept[er.ID]
782 if ok {
783 delete(ext.pendingIntercept, er.ID)
784 }
785 ext.mu.Unlock()
786 if ok {
787 select {
788 case ch <- er:
789 default:
790 }
791 }
792 }
793 case "tool_result":
794 var tr extproto.ToolResultFromExt
795 if err := json.Unmarshal(line, &tr); err == nil {
796 ext.mu.Lock()
797 ch, ok := ext.pendingTool[tr.ID]
798 if ok {
799 delete(ext.pendingTool, tr.ID)
800 }
801 ext.mu.Unlock()
802 if ok {
803 select {
804 case ch <- tr:
805 default:
806 }
807 }
808 }
809 case "notify":
810 var n extproto.NotifyFromExt
811 if err := json.Unmarshal(line, &n); err == nil {
812 m.hooks.Notify(ext.Manifest.Name, n.Level, n.Message)
813 }
814 case "clear_notes":
815 m.hooks.ClearNotes(ext.Manifest.Name)
816 case "submit":
817 var s extproto.SubmitFromExt
818 if err := json.Unmarshal(line, &s); err == nil {
819 text := strings.TrimSpace(s.Text)
820 if text != "" {
821 m.hooks.Submit(text)
822 }
823 }
824 case "submit_slash":
825 // Spontaneous request to invoke a slash command in the
826 // TUI. Refused unless the payload looks like a slash
827 // command so a misbehaving extension can't sneak a model
828 // prompt through this path.
829 var s extproto.SubmitSlashFromExt
830 if err := json.Unmarshal(line, &s); err == nil {
831 text := strings.TrimSpace(s.Text)
832 if strings.HasPrefix(text, "/") {
833 m.hooks.SubmitSlash(text)
834 } else {
835 fmt.Fprintf(ext.logFile, "[zot] submit_slash refused (not a slash command): %q\n", s.Text)
836 }
837 }
838 case "command_response":
839 var cr extproto.CommandResponseFromExt
840 if err := json.Unmarshal(line, &cr); err == nil {
841 ext.mu.Lock()
842 ch, ok := ext.pending[cr.ID]
843 if ok {
844 delete(ext.pending, cr.ID)
845 }
846 ext.mu.Unlock()
847 if ok {
848 select {
849 case ch <- cr:
850 default:
851 }
852 }
853 }
854 case "open_panel":
855 var op extproto.OpenPanelFromExt
856 if err := json.Unmarshal(line, &op); err == nil {
857 m.hooks.OpenPanel(ext.Manifest.Name, op.Panel)
858 }
859 case "panel_render":
860 var pr extproto.PanelRenderFromExt
861 if err := json.Unmarshal(line, &pr); err == nil {
862 m.hooks.UpdatePanel(ext.Manifest.Name, pr.PanelID, pr.Title, pr.Lines, pr.Footer)
863 }
864 case "panel_close":
865 var pc extproto.PanelCloseFromExt
866 if err := json.Unmarshal(line, &pc); err == nil {
867 m.hooks.ClosePanel(ext.Manifest.Name, pc.PanelID)
868 }
869 case "shutdown_ack":
870 // Caller of Stop is waiting on the process exit, not this frame.
871 default:
872 ext.recordDiagnostic(fmt.Sprintf("unknown frame type %q", frame.Type))
873 fmt.Fprintf(ext.logFile, "[zot] unknown frame type %q\n", frame.Type)
874 }
875 }
876}
877
878func (ext *Extension) recordDiagnostic(msg string) {
879 ext.mu.Lock()
880 defer ext.mu.Unlock()
881 ext.diagnostics = append(ext.diagnostics, msg)
882}
883
884// RegisteredCommandDiagnostic describes one command an extension attempted
885// to register and whether it owns the active dispatch slot.
886type RegisteredCommandDiagnostic struct {
887 Name string
888 Description string
889 Active bool
890}
891
892// RegisteredToolDiagnostic describes one tool an extension attempted to
893// register and whether it owns the active dispatch slot.
894type RegisteredToolDiagnostic struct {
895 Name string
896 Description string
897 Active bool
898}
899
900// ExtensionDiagnostic is a stable snapshot for diagnostic commands.
901type ExtensionDiagnostic struct {
902 Name string
903 Version string
904 Dir string
905 LogPath string
906 ThemeOnly bool
907 Ready bool
908 AutoReady bool
909 ReadyTimedOut bool
910 Commands []RegisteredCommandDiagnostic
911 Tools []RegisteredToolDiagnostic
912 Messages []string
913}
914
915// Diagnostics returns a snapshot of loaded extension state for commands such
916// as `zot ext doctor`. It is read-only and does not change manager behavior.
917func (m *Manager) Diagnostics() []ExtensionDiagnostic {
918 m.mu.RLock()
919 defer m.mu.RUnlock()
920 names := make([]string, 0, len(m.ext))
921 for name := range m.ext {
922 names = append(names, name)
923 }
924 sort.Strings(names)
925 out := make([]ExtensionDiagnostic, 0, len(names))
926 for _, name := range names {
927 ext := m.ext[name]
928 ext.mu.Lock()
929 msgs := append([]string(nil), ext.diagnostics...)
930 autoReady := ext.autoReady
931 readyTimedOut := ext.readyTimedOut
932 ext.mu.Unlock()
933
934 diag := ExtensionDiagnostic{
935 Name: ext.Manifest.Name,
936 Version: ext.Manifest.Version,
937 Dir: ext.Dir,
938 LogPath: ext.LogPath,
939 ThemeOnly: ext.Manifest.Exec == "",
940 AutoReady: autoReady,
941 ReadyTimedOut: readyTimedOut,
942 Messages: msgs,
943 }
944 select {
945 case <-ext.readyCh:
946 diag.Ready = true
947 default:
948 }
949 for _, c := range ext.commands {
950 registration, ok := m.commandIndex[strings.ToLower(c.Name)]
951 diag.Commands = append(diag.Commands, RegisteredCommandDiagnostic{
952 Name: c.Name,
953 Description: c.Description,
954 Active: ok && registration.ext == ext && registration.name == c.Name,
955 })
956 }
957 for _, t := range ext.tools {
958 diag.Tools = append(diag.Tools, RegisteredToolDiagnostic{
959 Name: t.Name,
960 Description: t.Description,
961 Active: m.toolIndex[t.Name] == ext,
962 })
963 }
964 out = append(out, diag)
965 }
966 return out
967}
968
969// Commands returns a snapshot of every (extension, command) pair
970// currently registered. Used by the slash autocomplete + /help.
971func (m *Manager) Commands() []CommandInfo {
972 m.mu.RLock()
973 defer m.mu.RUnlock()
974 var out []CommandInfo
975 for _, ext := range m.ext {
976 for _, c := range ext.commands {
977 registration, ok := m.commandIndex[strings.ToLower(c.Name)]
978 if !ok || registration.ext != ext || registration.name != c.Name {
979 continue
980 }
981 out = append(out, CommandInfo{
982 Extension: ext.Manifest.Name,
983 Name: c.Name,
984 Description: c.Description,
985 })
986 }
987 }
988 return out
989}
990
991// CommandInfo is one extension-registered slash command, surfaced to
992// the rest of zot for display purposes.
993type CommandInfo struct {
994 Extension string
995 Name string
996 Description string
997}
998
999// ToolInfo is one extension-registered tool. Used by the agent's
1000// build step to materialise core.Tool wrappers.
1001type ToolInfo struct {
1002 Extension string
1003 Name string
1004 Description string
1005 Schema json.RawMessage
1006 Deferred bool
1007}
1008
1009// Tools returns a snapshot of every (extension, tool) pair currently
1010// registered. Used at agent-build time to fold extension tools into
1011// the runtime tool registry.
1012func (m *Manager) Tools() []ToolInfo {
1013 m.mu.RLock()
1014 defer m.mu.RUnlock()
1015 var out []ToolInfo
1016 for _, ext := range m.ext {
1017 for _, t := range ext.tools {
1018 out = append(out, ToolInfo{
1019 Extension: ext.Manifest.Name,
1020 Name: t.Name,
1021 Description: t.Description,
1022 Schema: t.Schema,
1023 Deferred: t.Deferred,
1024 })
1025 }
1026 }
1027 return out
1028}
1029
1030// HasTool reports whether name is registered by any extension.
1031func (m *Manager) HasTool(name string) bool {
1032 m.mu.RLock()
1033 defer m.mu.RUnlock()
1034 _, ok := m.toolIndex[name]
1035 return ok
1036}
1037
1038// InvokeTool sends a tool_call to the owning extension and waits for
1039// the matching tool_result. Used by the core.Tool wrapper that the
1040// agent registers per extension-defined tool.
1041func (m *Manager) InvokeTool(ctx context.Context, name string, args json.RawMessage, timeout time.Duration) (extproto.ToolResultFromExt, error) {
1042 m.mu.RLock()
1043 ext, ok := m.toolIndex[name]
1044 m.mu.RUnlock()
1045 if !ok {
1046 return extproto.ToolResultFromExt{}, fmt.Errorf("no extension registered for tool %q", name)
1047 }
1048
1049 id := newCorrelationID()
1050 ch := make(chan extproto.ToolResultFromExt, 1)
1051 ext.mu.Lock()
1052 ext.pendingTool[id] = ch
1053 ext.mu.Unlock()
1054
1055 frame, _ := extproto.Encode(extproto.ToolCallFromHost{
1056 Type: "tool_call",
1057 ID: id,
1058 Name: name,
1059 Args: args,
1060 })
1061 if _, err := ext.stdin.Write(frame); err != nil {
1062 ext.mu.Lock()
1063 delete(ext.pendingTool, id)
1064 ext.mu.Unlock()
1065 return extproto.ToolResultFromExt{}, fmt.Errorf("write: %w", err)
1066 }
1067
1068 select {
1069 case resp := <-ch:
1070 return resp, nil
1071 case <-time.After(timeout):
1072 ext.mu.Lock()
1073 delete(ext.pendingTool, id)
1074 ext.mu.Unlock()
1075 return extproto.ToolResultFromExt{}, fmt.Errorf("timeout waiting for %s/%s", ext.Manifest.Name, name)
1076 case <-ctx.Done():
1077 ext.mu.Lock()
1078 delete(ext.pendingTool, id)
1079 ext.mu.Unlock()
1080 return extproto.ToolResultFromExt{}, ctx.Err()
1081 }
1082}
1083
1084// HasCommand reports whether name is registered by any extension.
1085func (m *Manager) HasCommand(name string) bool {
1086 m.mu.RLock()
1087 defer m.mu.RUnlock()
1088 _, ok := m.commandIndex[strings.ToLower(name)]
1089 return ok
1090}
1091
1092func (m *Manager) CommandOwner(name string) string {
1093 m.mu.RLock()
1094 defer m.mu.RUnlock()
1095 if registration, ok := m.commandIndex[strings.ToLower(name)]; ok && registration.ext != nil {
1096 return registration.ext.Manifest.Name
1097 }
1098 return ""
1099}
1100
1101// Invoke fires the named slash command's handler in the owning
1102// extension and waits up to timeout for the response. Returns the
1103// extension's CommandResponse so the caller can act on the action
1104// (prompt / insert / display).
1105func (m *Manager) Invoke(ctx context.Context, name, args string, timeout time.Duration) (extproto.CommandResponseFromExt, error) {
1106 m.mu.RLock()
1107 registration, ok := m.commandIndex[strings.ToLower(name)]
1108 m.mu.RUnlock()
1109 if !ok {
1110 return extproto.CommandResponseFromExt{}, fmt.Errorf("no extension registered for /%s", name)
1111 }
1112 ext := registration.ext
1113 name = registration.name
1114
1115 id := newCorrelationID()
1116 ch := make(chan extproto.CommandResponseFromExt, 1)
1117 ext.mu.Lock()
1118 ext.pending[id] = ch
1119 ext.mu.Unlock()
1120
1121 frame, _ := extproto.Encode(extproto.CommandInvokedFromHost{
1122 Type: "command_invoked",
1123 ID: id,
1124 Name: name,
1125 Args: args,
1126 })
1127 if _, err := ext.stdin.Write(frame); err != nil {
1128 ext.mu.Lock()
1129 delete(ext.pending, id)
1130 ext.mu.Unlock()
1131 return extproto.CommandResponseFromExt{}, fmt.Errorf("write: %w", err)
1132 }
1133
1134 select {
1135 case resp := <-ch:
1136 return resp, nil
1137 case <-time.After(timeout):
1138 ext.mu.Lock()
1139 delete(ext.pending, id)
1140 ext.mu.Unlock()
1141 return extproto.CommandResponseFromExt{}, fmt.Errorf("timeout waiting for %s/%s", ext.Manifest.Name, name)
1142 case <-ctx.Done():
1143 ext.mu.Lock()
1144 delete(ext.pending, id)
1145 ext.mu.Unlock()
1146 return extproto.CommandResponseFromExt{}, ctx.Err()
1147 }
1148}
1149
1150// Stop cleanly terminates every extension. Sends ShutdownFromHost,
1151// waits up to gracePeriod for each subprocess to exit, then SIGTERMs
1152// (and SIGKILLs after another second) the holdouts.
1153func (m *Manager) SendPanelKey(extName, panelID, key, text string) error {
1154 m.mu.RLock()
1155 ext, ok := m.ext[extName]
1156 m.mu.RUnlock()
1157 if !ok {
1158 return fmt.Errorf("no extension %q", extName)
1159 }
1160 frame, _ := extproto.Encode(extproto.PanelKeyFromHost{Type: "panel_key", PanelID: panelID, Key: key, Text: text})
1161 _, err := ext.stdin.Write(frame)
1162 return err
1163}
1164
1165func (m *Manager) SendPanelClose(extName, panelID string) error {
1166 m.mu.RLock()
1167 ext, ok := m.ext[extName]
1168 m.mu.RUnlock()
1169 if !ok {
1170 return fmt.Errorf("no extension %q", extName)
1171 }
1172 frame, _ := extproto.Encode(extproto.PanelCloseFromHost{Type: "panel_close", PanelID: panelID})
1173 _, err := ext.stdin.Write(frame)
1174 return err
1175}
1176
1177func (m *Manager) Stop(gracePeriod time.Duration) {
1178 m.mu.RLock()
1179 exts := make([]*Extension, 0, len(m.ext))
1180 for _, e := range m.ext {
1181 exts = append(exts, e)
1182 }
1183 m.mu.RUnlock()
1184 stopExtensions(exts, gracePeriod)
1185}
1186
1187func stopExtensions(exts []*Extension, gracePeriod time.Duration) {
1188 for _, ext := range exts {
1189 if ext.stdin == nil {
1190 continue
1191 }
1192 if frame, err := extproto.Encode(extproto.ShutdownFromHost{Type: "shutdown"}); err == nil {
1193 _, _ = ext.stdin.Write(frame)
1194 }
1195 _ = ext.stdin.Close()
1196 }
1197
1198 deadline := time.Now().Add(gracePeriod)
1199 for _, ext := range exts {
1200 if ext.cmd == nil {
1201 if ext.logFile != nil {
1202 _ = ext.logFile.Close()
1203 }
1204 continue
1205 }
1206 remaining := time.Until(deadline)
1207 if remaining <= 0 {
1208 remaining = 100 * time.Millisecond
1209 }
1210 done := make(chan struct{})
1211 go func() { _ = ext.cmd.Wait(); close(done) }()
1212 select {
1213 case <-done:
1214 case <-time.After(remaining):
1215 _ = ext.cmd.Process.Signal(syscall.SIGTERM)
1216 select {
1217 case <-done:
1218 case <-time.After(time.Second):
1219 _ = ext.cmd.Process.Kill()
1220 <-done
1221 }
1222 }
1223 if ext.logFile != nil {
1224 _ = ext.logFile.Close()
1225 }
1226 }
1227}
1228
1229// All returns every extension currently tracked, enabled or not.
1230// Used by `zot ext list`.
1231func (m *Manager) All() []*Extension {
1232 m.mu.RLock()
1233 defer m.mu.RUnlock()
1234 out := make([]*Extension, 0, len(m.ext))
1235 for _, e := range m.ext {
1236 out = append(out, e)
1237 }
1238 return out
1239}
1240
1241// newCorrelationID returns a short non-cryptographic id. We don't
1242// need uniqueness across processes, just within the lifetime of one
1243// extension's pending map.
1244func newCorrelationID() string {
1245 return strings.ReplaceAll(time.Now().Format("150405.000000"), ".", "")
1246}