14 "github.com/stretchr/testify/require"
17 func TestTurnRunsToolsUntilTheFinalAnswer(t *testing.T) {
18 server := newModelFake(t)
19 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
21 modelWriteEvents(w, modelCompletedEvent([]any{
22 modelReasoningItem("rs_1", "opaque thought"),
23 modelFunctionCallItem("fc_1", "call_1", "shell", `{"command":"echo hi"}`),
29 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("It said hi.")}, 120, 100, 5))
32 test := newTestTurnLoop(t, server)
33 _, err := test.loop.Submit(terminalSource, "Run echo hi.")
34 require.NoError(t, err)
35 test.events.WaitFor(t, "idle", 1)
36 require.Equal(t, 2, server.Calls())
38 first := server.Call(0).body
39 second := server.Call(1).body
40 instructions, err := test.agent.Prompt()
41 require.NoError(t, err)
42 require.Equal(t, instructions, first["instructions"])
43 require.Equal(t, test.agent.config.Model, first["model"])
45 // The second request begins with the whole first one.
46 firstInput := first["input"].([]any)
47 secondInput := second["input"].([]any)
48 require.Equal(t, firstInput, secondInput[:len(firstInput)])
49 require.Equal(t, []any{
50 map[string]any{"type": "reasoning", "encrypted_content": "opaque thought", "summary": []any{}},
51 map[string]any{"type": "function_call", "call_id": "call_1", "name": "shell", "arguments": `{"command":"echo hi"}`},
52 map[string]any{"type": "function_call_output", "call_id": "call_1", "output": "hi\n"},
53 }, secondInput[len(firstInput):])
55 require.Equal(t, []string{
57 `tool: shell {"command":"echo hi"}`,
61 require.Equal(t, []string{"user", "busy", "tool", "echo", "reply", "idle"}, test.events.Kinds())
65 for _, record := range modelReadUsage(t, test.usagePath) {
66 turns = append(turns, record.Turn)
69 require.Equal(t, []string{"first", "next"}, turns)
72 func TestFirstRequestHoldsTheViewTheStateAndTheMessage(t *testing.T) {
73 server := newModelFake(t)
74 test := newTestTurnLoop(t, server)
75 _, err := test.memory.Append("user", "an earlier message")
76 require.NoError(t, err)
78 _, err = test.loop.Submit(terminalSource, "The new message.")
79 require.NoError(t, err)
80 test.events.WaitFor(t, "idle", 1)
82 require.Equal(t, []string{
83 "<chat>\n0+1|user: an earlier message\n",
84 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
86 }, turnInputTexts(t, server.Call(0)))
89 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
90 server := newModelFake(t)
91 arrived := make(chan struct{})
92 release := make(chan struct{})
93 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
97 modelWriteEvents(w, modelCompletedEvent([]any{modelFunctionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
102 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("done")}, 100, 0, 5))
105 test := newTestTurnLoop(t, server)
106 _, err := test.loop.Submit(terminalSource, "first")
107 require.NoError(t, err)
111 _, err = test.loop.Submit(terminalSource, "second, during the turn")
112 require.NoError(t, err)
115 test.events.WaitFor(t, "idle", 1)
117 input := server.Call(1).body["input"].([]any)
118 require.Equal(t, map[string]any{
120 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
121 }, input[len(input)-1])
123 time.Sleep(50 * time.Millisecond)
124 require.Equal(t, 2, server.Calls())
125 require.Equal(t, 1, test.events.Count("busy"))
128 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
129 server := newModelFake(t)
130 arrived := make(chan struct{})
131 release := make(chan struct{})
132 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
133 if call.number == 0 {
138 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
141 test := newTestTurnLoop(t, server)
142 _, err := test.loop.Submit(terminalSource, "first")
143 require.NoError(t, err)
147 _, err = test.loop.Submit(terminalSource, "second")
148 require.NoError(t, err)
151 test.events.WaitFor(t, "idle", 2)
152 require.Equal(t, 2, server.Calls())
154 texts := turnInputTexts(t, server.Call(1))
155 require.Equal(t, "second", texts[len(texts)-1])
156 require.Contains(t, strings.Join(texts, ""), "|reply: answer\n")
159 func TestTurnWaitsForTheSummaries(t *testing.T) {
160 server := newModelFake(t)
161 summarizer := newFakeSummarizer()
162 summarizer.gate = make(chan struct{})
163 test := newTestTurnLoopWith(t, server, summarizer)
165 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
166 require.NoError(t, err)
168 _, err = test.loop.Submit(terminalSource, "go on")
169 require.NoError(t, err)
171 time.Sleep(50 * time.Millisecond)
172 require.Equal(t, 0, server.Calls())
174 close(summarizer.gate)
175 test.events.WaitFor(t, "idle", 1)
176 require.Equal(t, 1, server.Calls())
177 require.NotContains(t, turnInputTexts(t, server.Call(0))[0], viewPending)
180 func TestInterruptEndsTheTurn(t *testing.T) {
181 server := newModelFake(t)
182 arrived := make(chan struct{})
183 release := make(chan struct{})
184 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
185 if call.number == 0 {
190 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
193 test := newTestTurnLoop(t, server)
194 _, err := test.loop.Submit(terminalSource, "a long task")
195 require.NoError(t, err)
198 require.True(t, test.loop.Interrupt())
199 test.events.WaitFor(t, "idle", 1)
202 require.Equal(t, "interrupted", test.events.Last("error"))
203 require.Equal(t, []string{"user: a long task"}, test.Log(t))
204 require.False(t, test.loop.Interrupt())
206 _, err = test.loop.Submit(terminalSource, "again")
207 require.NoError(t, err)
208 test.events.WaitFor(t, "idle", 2)
209 require.Equal(t, "reply: answer", test.Log(t)[2])
210 require.Empty(t, test.logs.String())
213 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
214 server := newModelFake(t)
215 server.reply = func(w http.ResponseWriter, _ ModelFakeCall) {
216 http.Error(w, "busy", http.StatusServiceUnavailable)
219 test := newTestTurnLoop(t, server)
220 _, err := test.loop.Submit(terminalSource, "hello")
221 require.NoError(t, err)
222 test.events.WaitFor(t, "idle", 1)
224 time.Sleep(50 * time.Millisecond)
225 require.Equal(t, 1, server.Calls())
226 require.Contains(t, test.events.Last("error"), "HTTP 503")
227 require.Equal(t, 1, strings.Count(test.logs.String(), "\n"))
228 require.Contains(t, test.logs.String(), "turn failed: ")
229 require.Contains(t, test.logs.String(), "HTTP 503")
230 require.NotContains(t, test.logs.String(), "hello")
233 func TestViewMessagesOnlyGrowBetweenTurns(t *testing.T) {
234 server := newModelFake(t)
235 test := newTestTurnLoop(t, server)
236 for n, text := range []string{"one", "two", "three"} {
237 _, err := test.loop.Submit(terminalSource, text)
238 require.NoError(t, err)
239 test.events.WaitFor(t, "idle", n+1)
242 first := server.Call(1).body["input"].([]any)
243 second := server.Call(2).body["input"].([]any)
245 require.Len(t, first, 4)
246 require.Len(t, second, 5)
247 require.Equal(t, first[:2], second[:2])
248 require.Equal(t, []string{
250 "0+1|user: one\n1+1|reply: a line\n",
251 "2+1|user: two\n3+1|reply: a line\n",
252 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
254 }, turnInputTexts(t, server.Call(2)))
257 // TurnTestLoop is a loop that runs in the background of a test, with its parts.
258 type TurnTestLoop struct {
262 events *TurnEventRecorder
267 // Log returns the logged messages as "kind: text".
268 func (l *TurnTestLoop) Log(t *testing.T) []string {
273 for id := range l.memory.Len() {
274 message, err := l.memory.Message(id)
275 require.NoError(t, err)
277 lines = append(lines, message.Kind+": "+message.Text)
283 // TurnEventRecorder keeps the events of a loop.
284 type TurnEventRecorder struct {
289 // WaitFor waits until count events of kind came.
290 func (r *TurnEventRecorder) WaitFor(t *testing.T, kind string, count int) {
293 require.Eventually(t, func() bool {
294 return r.Count(kind) >= count
295 }, 10*time.Second, time.Millisecond)
298 func (r *TurnEventRecorder) Kinds() []string {
300 defer r.mutex.Unlock()
302 kinds := make([]string, len(r.events))
303 for n, event := range r.events {
304 kinds[n] = event.Kind
310 func (r *TurnEventRecorder) Count(kind string) int {
312 for _, each := range r.Kinds() {
321 // Last returns the text of the last event of kind.
322 func (r *TurnEventRecorder) Last(kind string) string {
323 return r.LastEvent(kind).Text
326 // LastEvent returns the last event of kind.
327 func (r *TurnEventRecorder) LastEvent(kind string) Event {
329 defer r.mutex.Unlock()
331 for n := len(r.events) - 1; n >= 0; n-- {
332 if r.events[n].Kind == kind {
340 func (r *TurnEventRecorder) Add(event Event) {
342 defer r.mutex.Unlock()
344 r.events = append(r.events, event)
347 func newTestTurnLoop(t *testing.T, server *ModelFake) *TurnTestLoop {
350 return newTestTurnLoopWith(t, server, newFakeSummarizer())
353 // turnInputTexts returns the text of each user message of the input of call.
354 func turnInputTexts(t *testing.T, call ModelFakeCall) []string {
359 Role string `json:"role"`
361 Text string `json:"text"`
366 require.NoError(t, json.Unmarshal(call.raw, &body))
370 for _, item := range body.Input {
371 if item.Role != "user" {
375 var text strings.Builder
377 for _, part := range item.Content {
378 text.WriteString(part.Text)
381 texts = append(texts, text.String())
387 // newTestTurnLoopWith returns a running loop on a new agent whose builds use
388 // summarizer and whose commands run in a temporary directory.
389 func newTestTurnLoopWith(t *testing.T, server *ModelFake, summarizer Summarizer) *TurnTestLoop {
392 agent, err := openAgent(t.TempDir())
393 require.NoError(t, err)
395 agent.config.Workdir = t.TempDir()
396 clock := newTestClock()
397 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer, discardLog())
398 require.NoError(t, err)
400 test := &TurnTestLoop{
403 events: &TurnEventRecorder{},
405 usagePath: filepath.Join(t.TempDir(), "usage.jsonl"),
408 client := newModelClient(server.URL, ModelAPIKey(modelTestKey), agent.CacheKey(), test.usagePath)
409 whatsapp, _, _ := whatsappStartTest(t, t.TempDir(), clock)
410 test.loop, err = newTurnLoop(agent, memory, newTestSubagents(t), whatsapp, newTestSchedules(t, clock.Now), client, clock.Now, log.New(test.logs, "", 0), test.events.Add)
411 require.NoError(t, err)
413 ctx, cancel := context.WithCancel(t.Context())
414 done := make(chan error)
416 done <- test.loop.Run(ctx)
421 require.ErrorIs(t, <-done, context.Canceled)
422 require.NoError(t, memory.Close())