13 "github.com/stretchr/testify/require"
16 func TestTurnRunsToolsUntilTheFinalAnswer(t *testing.T) {
17 server := newModelFake(t)
18 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
20 modelWriteEvents(w, modelCompletedEvent([]any{
21 modelReasoningItem("rs_1", "opaque thought"),
22 modelFunctionCallItem("fc_1", "call_1", "shell", `{"command":"echo hi"}`),
28 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("It said hi.")}, 120, 100, 5))
31 test := newTestTurnLoop(t, server)
32 _, err := test.loop.Submit(terminalSource, "Run echo hi.")
33 require.NoError(t, err)
34 test.events.WaitFor(t, "idle", 1)
35 require.Equal(t, 2, server.Calls())
37 first := server.Call(0).body
38 second := server.Call(1).body
39 instructions, err := test.agent.Prompt()
40 require.NoError(t, err)
41 require.Equal(t, instructions, first["instructions"])
42 require.Equal(t, test.agent.config.Model, first["model"])
44 // The second request begins with the whole first one.
45 firstInput := first["input"].([]any)
46 secondInput := second["input"].([]any)
47 require.Equal(t, firstInput, secondInput[:len(firstInput)])
48 require.Equal(t, []any{
49 map[string]any{"type": "reasoning", "encrypted_content": "opaque thought", "summary": []any{}},
50 map[string]any{"type": "function_call", "call_id": "call_1", "name": "shell", "arguments": `{"command":"echo hi"}`},
51 map[string]any{"type": "function_call_output", "call_id": "call_1", "output": "hi\n"},
52 }, secondInput[len(firstInput):])
54 require.Equal(t, []string{
56 `tool: shell {"command":"echo hi"}`,
60 require.Equal(t, []string{"user", "busy", "tool", "echo", "reply", "idle"}, test.events.Kinds())
64 for _, record := range modelReadUsage(t, test.usagePath) {
65 turns = append(turns, record.Turn)
68 require.Equal(t, []string{"first", "next"}, turns)
71 func TestFirstRequestHoldsTheViewTheStateAndTheMessage(t *testing.T) {
72 server := newModelFake(t)
73 test := newTestTurnLoop(t, server)
74 _, err := test.memory.Append("user", "an earlier message")
75 require.NoError(t, err)
77 _, err = test.loop.Submit(terminalSource, "The new message.")
78 require.NoError(t, err)
79 test.events.WaitFor(t, "idle", 1)
81 require.Equal(t, []string{
82 "<chat>\n0+1|user: an earlier message\n",
83 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
85 }, turnInputTexts(t, server.Call(0)))
88 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
89 server := newModelFake(t)
90 arrived := make(chan struct{})
91 release := make(chan struct{})
92 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
96 modelWriteEvents(w, modelCompletedEvent([]any{modelFunctionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
101 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("done")}, 100, 0, 5))
104 test := newTestTurnLoop(t, server)
105 _, err := test.loop.Submit(terminalSource, "first")
106 require.NoError(t, err)
110 _, err = test.loop.Submit(terminalSource, "second, during the turn")
111 require.NoError(t, err)
114 test.events.WaitFor(t, "idle", 1)
116 input := server.Call(1).body["input"].([]any)
117 require.Equal(t, map[string]any{
119 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
120 }, input[len(input)-1])
122 time.Sleep(50 * time.Millisecond)
123 require.Equal(t, 2, server.Calls())
124 require.Equal(t, 1, test.events.Count("busy"))
127 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
128 server := newModelFake(t)
129 arrived := make(chan struct{})
130 release := make(chan struct{})
131 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
132 if call.number == 0 {
137 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
140 test := newTestTurnLoop(t, server)
141 _, err := test.loop.Submit(terminalSource, "first")
142 require.NoError(t, err)
146 _, err = test.loop.Submit(terminalSource, "second")
147 require.NoError(t, err)
150 test.events.WaitFor(t, "idle", 2)
151 require.Equal(t, 2, server.Calls())
153 texts := turnInputTexts(t, server.Call(1))
154 require.Equal(t, "second", texts[len(texts)-1])
155 require.Contains(t, strings.Join(texts, ""), "|reply: answer\n")
158 func TestTurnWaitsForTheSummaries(t *testing.T) {
159 server := newModelFake(t)
160 summarizer := newFakeSummarizer()
161 summarizer.gate = make(chan struct{})
162 test := newTestTurnLoopWith(t, server, summarizer)
164 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
165 require.NoError(t, err)
167 _, err = test.loop.Submit(terminalSource, "go on")
168 require.NoError(t, err)
170 time.Sleep(50 * time.Millisecond)
171 require.Equal(t, 0, server.Calls())
173 close(summarizer.gate)
174 test.events.WaitFor(t, "idle", 1)
175 require.Equal(t, 1, server.Calls())
176 require.NotContains(t, turnInputTexts(t, server.Call(0))[0], viewPending)
179 func TestInterruptEndsTheTurn(t *testing.T) {
180 server := newModelFake(t)
181 arrived := make(chan struct{})
182 release := make(chan struct{})
183 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
184 if call.number == 0 {
189 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
192 test := newTestTurnLoop(t, server)
193 _, err := test.loop.Submit(terminalSource, "a long task")
194 require.NoError(t, err)
197 require.True(t, test.loop.Interrupt())
198 test.events.WaitFor(t, "idle", 1)
201 require.Equal(t, "interrupted", test.events.Last("error"))
202 require.Equal(t, []string{"user: a long task"}, test.Log(t))
203 require.False(t, test.loop.Interrupt())
205 _, err = test.loop.Submit(terminalSource, "again")
206 require.NoError(t, err)
207 test.events.WaitFor(t, "idle", 2)
208 require.Equal(t, "reply: answer", test.Log(t)[2])
209 require.Empty(t, test.logs.String())
212 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
213 server := newModelFake(t)
214 server.reply = func(w http.ResponseWriter, _ ModelFakeCall) {
215 http.Error(w, "busy", http.StatusServiceUnavailable)
218 test := newTestTurnLoop(t, server)
219 _, err := test.loop.Submit(terminalSource, "hello")
220 require.NoError(t, err)
221 test.events.WaitFor(t, "idle", 1)
223 time.Sleep(50 * time.Millisecond)
224 require.Equal(t, 1, server.Calls())
225 require.Contains(t, test.events.Last("error"), "HTTP 503")
226 require.Equal(t, 1, strings.Count(test.logs.String(), "\n"))
227 require.Contains(t, test.logs.String(), "<3>turn failed: ")
228 require.Contains(t, test.logs.String(), "HTTP 503")
229 require.NotContains(t, test.logs.String(), "hello")
232 func TestViewMessagesOnlyGrowBetweenTurns(t *testing.T) {
233 server := newModelFake(t)
234 test := newTestTurnLoop(t, server)
235 for n, text := range []string{"one", "two", "three"} {
236 _, err := test.loop.Submit(terminalSource, text)
237 require.NoError(t, err)
238 test.events.WaitFor(t, "idle", n+1)
241 first := server.Call(1).body["input"].([]any)
242 second := server.Call(2).body["input"].([]any)
244 require.Len(t, first, 4)
245 require.Len(t, second, 5)
246 require.Equal(t, first[:2], second[:2])
247 require.Equal(t, []string{
249 "0+1|user: one\n1+1|reply: a line\n",
250 "2+1|user: two\n3+1|reply: a line\n",
251 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
253 }, turnInputTexts(t, server.Call(2)))
256 // TurnTestLoop is a loop that runs in the background of a test, with its parts.
257 type TurnTestLoop struct {
261 events *TurnEventRecorder
266 // Log returns the logged messages as "kind: text".
267 func (l *TurnTestLoop) Log(t *testing.T) []string {
272 for id := range l.memory.Len() {
273 message, err := l.memory.Message(id)
274 require.NoError(t, err)
276 lines = append(lines, message.Kind+": "+message.Text)
282 // TurnEventRecorder keeps the events of a loop.
283 type TurnEventRecorder struct {
288 // WaitFor waits until count events of kind came.
289 func (r *TurnEventRecorder) WaitFor(t *testing.T, kind string, count int) {
292 require.Eventually(t, func() bool {
293 return r.Count(kind) >= count
294 }, 10*time.Second, time.Millisecond)
297 func (r *TurnEventRecorder) Kinds() []string {
299 defer r.mutex.Unlock()
301 kinds := make([]string, len(r.events))
302 for n, event := range r.events {
303 kinds[n] = event.Kind
309 func (r *TurnEventRecorder) Count(kind string) int {
311 for _, each := range r.Kinds() {
320 // Last returns the text of the last event of kind.
321 func (r *TurnEventRecorder) Last(kind string) string {
322 return r.LastEvent(kind).Text
325 // LastEvent returns the last event of kind.
326 func (r *TurnEventRecorder) LastEvent(kind string) Event {
328 defer r.mutex.Unlock()
330 for n := len(r.events) - 1; n >= 0; n-- {
331 if r.events[n].Kind == kind {
339 func (r *TurnEventRecorder) Add(event Event) {
341 defer r.mutex.Unlock()
343 r.events = append(r.events, event)
346 func newTestTurnLoop(t *testing.T, server *ModelFake) *TurnTestLoop {
349 return newTestTurnLoopWith(t, server, newFakeSummarizer())
352 // turnInputTexts returns the text of each user message of the input of call.
353 func turnInputTexts(t *testing.T, call ModelFakeCall) []string {
358 Role string `json:"role"`
360 Text string `json:"text"`
365 require.NoError(t, json.Unmarshal(call.raw, &body))
369 for _, item := range body.Input {
370 if item.Role != "user" {
374 var text strings.Builder
376 for _, part := range item.Content {
377 text.WriteString(part.Text)
380 texts = append(texts, text.String())
386 // newTestTurnLoopWith returns a running loop on a new agent whose builds use
387 // summarizer and whose commands run in a temporary directory.
388 func newTestTurnLoopWith(t *testing.T, server *ModelFake, summarizer Summarizer) *TurnTestLoop {
391 agent, err := openAgent(filepath.Join(t.TempDir(), "agent"))
392 require.NoError(t, err)
394 agent.config.Workdir = t.TempDir()
395 clock := newTestClock()
396 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer, discardLog())
397 require.NoError(t, err)
399 test := &TurnTestLoop{
402 events: &TurnEventRecorder{},
404 usagePath: filepath.Join(t.TempDir(), "usage.jsonl"),
407 client := newModelClient(server.URL, ModelAPIKey(modelTestKey), agent.CacheKey(), test.usagePath)
408 whatsapp, _, _ := whatsappStartTest(t, t.TempDir(), clock)
409 test.loop, err = newTurnLoop(agent, memory, newTestSubagents(t), whatsapp, newTestSchedules(t, clock.Now), client, clock.Now, daemonTestLogger(test.logs), test.events.Add)
410 require.NoError(t, err)
412 ctx, cancel := context.WithCancel(t.Context())
413 done := make(chan error)
415 done <- test.loop.Run(ctx)
420 require.ErrorIs(t, <-done, context.Canceled)
421 require.NoError(t, memory.Close())