1package subagents
2
3import (
4 "context"
5 "encoding/json"
6 "fmt"
7 "os"
8 "path/filepath"
9 "strings"
10 "testing"
11 "time"
12)
13
14// TestSubagentWorkerArgs locks in the exact flag set the subprocess
15// runner uses to start a subagent agent in daemon mode. Past
16// regressions in this area:
17//
18// - "--no-sess" instead of "--no-session" (old print-mode runner):
19// every spawned agent died with "unknown flag" before it could
20// talk to the model.
21//
22// - Forgetting --cwd: the child resolved tools against the parent
23// zut's working directory, defeating the whole point of the
24// worktree isolation.
25//
26// - Forgetting --session: a daemon-mode agent without a session
27// file would lose context between follow-up turns, making
28// "send another message" mostly useless.
29//
30// The test asserts the load-bearing pieces are present in plausible
31// positions. If a flag is renamed, update both the runner and this
32// test so we notice immediately.
33func TestCredentialStdinHelperProcess(t *testing.T) {
34 if os.Getenv("ZUT_SUBAGENT_CREDENTIAL_HELPER") != "1" {
35 return
36 }
37 if os.Getenv("ZUT_SUBAGENT_CREDENTIAL_STDIN") != "1" {
38 os.Exit(2)
39 }
40 var credential Credential
41 if err := json.NewDecoder(os.Stdin).Decode(&credential); err != nil {
42 os.Exit(3)
43 }
44 if credential.Value != "inherited-secret" || credential.Method != "apikey" {
45 os.Exit(4)
46 }
47 for _, arg := range os.Args {
48 if strings.Contains(arg, credential.Value) {
49 os.Exit(5)
50 }
51 }
52 for _, env := range os.Environ() {
53 if strings.Contains(env, credential.Value) {
54 os.Exit(6)
55 }
56 }
57 fmt.Println(`{"type":"agent_stopped","data":{"reason":"completed"}}`)
58 os.Exit(0)
59}
60
61func TestWorkerEnvironmentRedactsProviderSecrets(t *testing.T) {
62 t.Setenv("OPENAI_API_KEY", "openai-secret")
63 t.Setenv("CUSTOM_PROVIDER_API_KEY", "custom-secret")
64 t.Setenv("HF_TOKEN", "hf-secret")
65 t.Setenv("GITHUB_TOKEN", "github-secret")
66 t.Setenv("CUSTOM_PROVIDER_TOKEN", "token-secret")
67 t.Setenv("CUSTOM_PROVIDER_KEY", "key-secret")
68 t.Setenv("AWS_ACCESS_KEY_ID", "aws-access-secret")
69 t.Setenv("GOOGLE_APPLICATION_CREDENTIALS", "/tmp/service-account.json")
70 t.Setenv("ZUT_SUBAGENT_TEST_VALUE", "not-secret")
71 env := strings.Join(workerEnvironment("openai"), "\n")
72 for _, secret := range []string{
73 "OPENAI_API_KEY=openai-secret",
74 "CUSTOM_PROVIDER_API_KEY=custom-secret",
75 "HF_TOKEN=hf-secret",
76 "GITHUB_TOKEN=github-secret",
77 "CUSTOM_PROVIDER_TOKEN=token-secret",
78 "CUSTOM_PROVIDER_KEY=key-secret",
79 "AWS_ACCESS_KEY_ID=aws-access-secret",
80 "GOOGLE_APPLICATION_CREDENTIALS=/tmp/service-account.json",
81 } {
82 if strings.Contains(env, secret) {
83 t.Fatalf("worker environment leaked %s", secret)
84 }
85 }
86 if !strings.Contains(env, "ZUT_SUBAGENT_TEST_VALUE=not-secret") {
87 t.Fatal("worker environment dropped unrelated variables")
88 }
89
90 t.Setenv("AWS_BEARER_TOKEN_BEDROCK", "bedrock-secret")
91 bedrockEnv := strings.Join(workerEnvironment("amazon-bedrock"), "\n")
92 if !strings.Contains(bedrockEnv, "AWS_BEARER_TOKEN_BEDROCK=bedrock-secret") {
93 t.Fatal("Bedrock ambient credential chain was unexpectedly removed")
94 }
95 if !strings.Contains(bedrockEnv, "AWS_ACCESS_KEY_ID=aws-access-secret") {
96 t.Fatal("Bedrock AWS access-key credential was unexpectedly removed")
97 }
98}
99
100func TestExecRunnerTransfersCredentialOnlyOnStdin(t *testing.T) {
101 t.Setenv("ZUT_SUBAGENT_CREDENTIAL_HELPER", "1")
102 root := t.TempDir()
103 a := &Agent{
104 ID: "credential-test",
105 Dir: root,
106 Provider: "custom",
107 SessionPath: filepath.Join(root, "session.json"),
108 InboxPath: filepath.Join(root, "inbox"),
109 EventLogPath: filepath.Join(root, "events.jsonl"),
110 }
111 calls := 0
112 r := &execRunner{
113 agent: a,
114 Command: []string{os.Args[0], "-test.run=^TestCredentialStdinHelperProcess$"},
115 resolveCredential: func(context.Context, string) (Credential, error) {
116 calls++
117 return Credential{Value: "inherited-secret", Method: "apikey"}, nil
118 },
119 }
120 if err := r.Run(context.Background(), agentSink{a: a}); err != nil {
121 t.Fatal(err)
122 }
123 if calls != 1 {
124 t.Fatalf("credential resolver called %d times, want 1", calls)
125 }
126}
127
128func TestApplyEventPreservesMultilineRolesAndFinalAssistant(t *testing.T) {
129 a := &Agent{}
130 sink := agentSink{a: a}
131 applyEventToSink(NewEvent("user_message", map[string]any{"content": []any{
132 map[string]any{"type": "text", "text": "first line\n\nlast line"},
133 }}), sink)
134 applyEventToSink(NewEvent("assistant_message", map[string]any{"content": []any{
135 map[string]any{"type": "text", "text": "complete\nanswer"},
136 }}), sink)
137
138 snap := a.Snapshot()
139 wantLines := []string{"user: first line", "user: ", "user: last line", "complete", "answer"}
140 if strings.Join(snap.Lines, "\n") != strings.Join(wantLines, "\n") {
141 t.Fatalf("transcript = %#v, want %#v", snap.Lines, wantLines)
142 }
143 if snap.LastAssistant != "complete\nanswer" {
144 t.Fatalf("last assistant = %q", snap.LastAssistant)
145 }
146}
147
148func TestSubagentWorkerArgs(t *testing.T) {
149 args := subagentWorkerArgs(subagentWorkerArgsOpts{
150 Exe: "/path/to/zut",
151 Dir: "/tmp/worktree",
152 SessionPath: "/tmp/state/session.json",
153 InboxPath: "/tmp/state/in.sock",
154 Task: "do the thing",
155 Reasoning: "high",
156 Subagent: "reviewer",
157 })
158 if len(args) < 7 {
159 t.Fatalf("argv unexpectedly short: %v", args)
160 }
161 if args[0] != "/path/to/zut" {
162 t.Fatalf("argv[0] = %q; want the binary path", args[0])
163 }
164 // The task must come last so anything that looks flag-like in
165 // the task body doesn't get interpreted as a flag.
166 if args[len(args)-1] != "do the thing" {
167 t.Fatalf("task should be last positional; got %v", args)
168 }
169
170 mustHave := map[string]string{
171 "--subagent-worker": "/tmp/state/in.sock",
172 "--session": "/tmp/state/session.json",
173 "--cwd": "/tmp/worktree",
174 "--reasoning": "high",
175 "--subagent": "reviewer",
176 }
177 for flag, value := range mustHave {
178 i := indexOf(args, flag)
179 if i < 0 {
180 t.Errorf("argv missing %q: %v", flag, args)
181 continue
182 }
183 if i+1 >= len(args) || args[i+1] != value {
184 t.Errorf("argv %q value = %q; want %q", flag, safeAt(args, i+1), value)
185 }
186 }
187
188 // Reject prior bad flags explicitly so a future revert is caught.
189 joined := strings.Join(args, " ")
190 for _, bad := range []string{"--print", "--no-sess ", "--no-session"} {
191 if strings.Contains(joined, bad) {
192 t.Fatalf("argv contains stale/wrong flag %q: %s", bad, joined)
193 }
194 }
195}
196
197func TestResolveSubagentExecutableUsesCurrentWhenAvailable(t *testing.T) {
198 got, err := resolveSubagentExecutable(context.Background(), "/current/bin/zut", "/old/bin/zut", func(_ context.Context, candidate string) (string, error) {
199 if candidate == "/current/bin/zut" {
200 return candidate, nil
201 }
202 return "", os.ErrNotExist
203 })
204 if err != nil {
205 t.Fatal(err)
206 }
207 if got != "/current/bin/zut" {
208 t.Fatalf("executable = %q, want current executable", got)
209 }
210}
211
212func TestResolveSubagentExecutableFallsBackToPath(t *testing.T) {
213 got, err := resolveSubagentExecutable(context.Background(), "/removed/.local/bin/zut", "/removed/.local/bin/zut", func(_ context.Context, candidate string) (string, error) {
214 if candidate == "zut" {
215 return "/active/go/bin/zut", nil
216 }
217 return "", os.ErrNotExist
218 })
219 if err != nil {
220 t.Fatal(err)
221 }
222 if got != "/active/go/bin/zut" {
223 t.Fatalf("executable = %q, want PATH fallback", got)
224 }
225}
226
227func TestResolveSubagentExecutableCancelsOtherLookupsAfterFirstSuccess(t *testing.T) {
228 slowStarted := make(chan struct{})
229 slowCanceled := make(chan struct{})
230 got, err := resolveSubagentExecutable(context.Background(), "/current/bin/zut", "/old/bin/zut", func(ctx context.Context, candidate string) (string, error) {
231 switch candidate {
232 case "/current/bin/zut":
233 close(slowStarted)
234 <-ctx.Done()
235 close(slowCanceled)
236 return "", ctx.Err()
237 case "zut":
238 <-slowStarted
239 return "/active/go/bin/zut", nil
240 default:
241 return "", os.ErrNotExist
242 }
243 })
244 if err != nil {
245 t.Fatal(err)
246 }
247 if got != "/active/go/bin/zut" {
248 t.Fatalf("executable = %q, want first successful lookup", got)
249 }
250 select {
251 case <-slowCanceled:
252 case <-time.After(time.Second):
253 t.Fatal("slower lookup was not canceled")
254 }
255}
256
257func TestSubagentWorkerArgsPropagatesProviderConnectionSettings(t *testing.T) {
258 args := subagentWorkerArgs(subagentWorkerArgsOpts{
259 Exe: "/zut",
260 Dir: "/work",
261 SessionPath: "/state/session.json",
262 InboxPath: "/state/inbox.sock",
263 BaseURL: "https://gateway.example.test/v1",
264 InsecureTLS: true,
265 })
266 if i := indexOf(args, "--base-url"); i < 0 || safeAt(args, i+1) != "https://gateway.example.test/v1" {
267 t.Fatalf("argv = %v, want --base-url with inherited endpoint", args)
268 }
269 if !containsArg(args, "--insecure") {
270 t.Fatalf("argv = %v, want inherited --insecure", args)
271 }
272 for _, arg := range args {
273 if strings.Contains(arg, "api-key") || strings.Contains(arg, "secret") {
274 t.Fatalf("argv contains credential material: %v", args)
275 }
276 }
277}
278
279func TestDefaultChildArgsPropagatesProviderConnectionSettings(t *testing.T) {
280 a := &Agent{
281 Dir: "/work",
282 BaseURL: "https://gateway.example.test/v1",
283 InsecureTLS: true,
284 }
285 args := defaultChildArgs("/zut", a, "/state/session.json", "/state/inbox.sock")
286 if i := indexOf(args, "--base-url"); i < 0 || safeAt(args, i+1) != a.BaseURL {
287 t.Fatalf("argv = %v, want Agent.BaseURL propagated", args)
288 }
289 if !containsArg(args, "--insecure") {
290 t.Fatalf("argv = %v, want Agent.InsecureTLS propagated", args)
291 }
292}
293
294// TestSubagentWorkerArgsEmptyTaskOmitsPositional makes sure that when the
295// agent is being adopted (no fresh task) we don't pass an empty
296// positional which the arg parser would treat as a real prompt.
297func TestSubagentWorkerArgsEmptyTaskOmitsPositional(t *testing.T) {
298 args := subagentWorkerArgs(subagentWorkerArgsOpts{
299 Exe: "/zut", Dir: "/wt", SessionPath: "/s.json", InboxPath: "/in.sock",
300 })
301 for _, a := range args {
302 if a == "" {
303 t.Fatalf("argv contains an empty positional: %v", args)
304 }
305 }
306 // last arg should be a real flag value, not a stray positional
307 if a := args[len(args)-1]; strings.HasPrefix(a, "--") {
308 t.Fatalf("argv ends on a flag with no value: %v", args)
309 }
310 for _, flag := range []string{"--reasoning", "--subagent"} {
311 if indexOf(args, flag) >= 0 {
312 t.Fatalf("empty child options should omit %s: %v", flag, args)
313 }
314 }
315}
316
317// TestDefaultChildArgsSpawnIncludesTask pins the spawn shape: a
318// fresh (non-resuming) Agent produces argv that ends with the
319// original task as a positional, so the child runs it as the
320// initial user turn.
321func TestDefaultChildArgsSpawnIncludesTask(t *testing.T) {
322 a := &Agent{Dir: "/wt", Task: "do thing"}
323 args := defaultChildArgs("/zut", a, "/s.json", "/in.sock")
324 if got := args[len(args)-1]; got != "do thing" {
325 t.Fatalf("spawn argv last = %q; want %q\n%v", got, "do thing", args)
326 }
327}
328
329// TestDefaultChildArgsResumeOmitsTask is the regression for the
330// "agent busy; send 'cancel' first" error: when an Agent is being
331// resumed (Resuming==true), the child argv MUST NOT include the
332// original Task as a positional. Otherwise the child fires the task
333// as a fresh user turn on every resume, racing with whatever the
334// user types next via the inbox.
335func TestDefaultChildArgsResumeOmitsTask(t *testing.T) {
336 a := &Agent{Dir: "/wt", Task: "do thing", Resuming: true}
337 args := defaultChildArgs("/zut", a, "/s.json", "/in.sock")
338 for _, v := range args {
339 if v == "do thing" {
340 t.Fatalf("resume argv contains the task; it would re-fire as a duplicate turn\n%v", args)
341 }
342 }
343 // And no trailing positional at all: every final token must be a
344 // recognized boolean flag or a value belonging to a preceding flag,
345 // never a stray empty string or the original task.
346 if got := args[len(args)-1]; got == "" {
347 t.Fatalf("resume argv ends with an empty positional: %v", args)
348 }
349}
350
351func TestUpdateAgentFromEventIgnoresProviderToolLoopTurnEnd(t *testing.T) {
352 a := &Agent{}
353 a.setTurnState(TurnRunning, "turn-1")
354 persisted := 0
355 a.persistFn = func(*Agent) { persisted++ }
356
357 updateAgentFromEvent(a, NewEvent("turn_end", map[string]any{"stop": "tool_use"}))
358
359 if got := a.TurnState(); got != TurnRunning {
360 t.Fatalf("turn state = %q, want %q", got, TurnRunning)
361 }
362 if got := a.CurrentTurnID(); got != "turn-1" {
363 t.Fatalf("turn ID = %q, want turn-1", got)
364 }
365 if persisted != 0 {
366 t.Fatalf("persist called %d times for provider/tool-loop turn_end, want 0", persisted)
367 }
368}
369
370func TestUpdateAgentFromEventPersistsTerminalTurnEndWithStep(t *testing.T) {
371 a := &Agent{}
372 persisted := 0
373 a.persistFn = func(*Agent) { persisted++ }
374
375 updateAgentFromEvent(a, NewEvent("turn_end", map[string]any{"step": float64(2), "error": "boom"}))
376
377 if got := a.TurnState(); got != TurnFailed {
378 t.Fatalf("turn state = %q, want %q", got, TurnFailed)
379 }
380 if persisted != 1 {
381 t.Fatalf("persist called %d times for terminal turn_end, want 1", persisted)
382 }
383}
384
385func TestNotifyPromptTurnEndIgnoresProviderToolLoopTurnEnd(t *testing.T) {
386 called := make(chan struct{}, 1)
387 a := &Agent{}
388 a.SetOnTurnEnd(func(step int, errMsg string) {
389 called <- struct{}{}
390 })
391
392 notifyPromptTurnEnd(a, NewEvent("turn_end", map[string]any{"stop": "tool_use"}))
393
394 select {
395 case <-called:
396 t.Fatal("OnTurnEnd fired for provider/tool-loop turn_end without step")
397 case <-time.After(50 * time.Millisecond):
398 }
399}
400
401func TestNotifyPromptTurnEndFiresForDaemonPromptCompletion(t *testing.T) {
402 type got struct {
403 step int
404 err string
405 }
406 called := make(chan got, 1)
407 a := &Agent{}
408 a.SetOnTurnEnd(func(step int, errMsg string) {
409 called <- got{step: step, err: errMsg}
410 })
411
412 notifyPromptTurnEnd(a, NewEvent("turn_end", map[string]any{"step": float64(2), "error": "boom"}))
413
414 select {
415 case g := <-called:
416 if g.step != 2 || g.err != "boom" {
417 t.Fatalf("callback = (%d, %q); want (2, boom)", g.step, g.err)
418 }
419 case <-time.After(time.Second):
420 t.Fatal("OnTurnEnd did not fire for prompt-level turn_end with step")
421 }
422}
423
424func TestSetOnTurnEndReplaysPendingNoticesInArrivalOrder(t *testing.T) {
425 a := &Agent{}
426 notifyPromptTurnEnd(a, NewEvent("turn_end", map[string]any{"step": float64(1)}))
427 notifyPromptTurnEnd(a, NewEvent("turn_end", map[string]any{"step": float64(2), "error": "boom"}))
428
429 type notice struct {
430 step int
431 err string
432 }
433 var got []notice
434 a.SetOnTurnEnd(func(step int, errMsg string) {
435 got = append(got, notice{step: step, err: errMsg})
436 })
437
438 want := []notice{{step: 1}, {step: 2, err: "boom"}}
439 if len(got) != len(want) {
440 t.Fatalf("replayed %d notices, want %d: %+v", len(got), len(want), got)
441 }
442 for i := range want {
443 if got[i] != want[i] {
444 t.Errorf("replayed notice %d = %+v, want %+v", i, got[i], want[i])
445 }
446 }
447}
448
449func indexOf(xs []string, want string) int {
450 for i, x := range xs {
451 if x == want {
452 return i
453 }
454 }
455 return -1
456}
457
458func safeAt(xs []string, i int) string {
459 if i < 0 || i >= len(xs) {
460 return ""
461 }
462 return xs[i]
463}