12 "github.com/stretchr/testify/require"
15 func TestTurnRunsToolsUntilTheFinalAnswer(t *testing.T) {
16 server := newFakeResponses(t)
17 server.reply = func(w http.ResponseWriter, call fakeCall) {
19 writeEvents(w, completedEvent([]any{
20 reasoningItem("rs_1", "opaque thought"),
21 functionCallItem("fc_1", "call_1", "shell", `{"command":"echo hi"}`),
27 writeEvents(w, completedEvent([]any{messageItem("It said hi.")}, 120, 100, 5))
30 test := newTestLoop(t, server)
31 _, err := test.loop.Submit("Run echo hi.")
32 require.NoError(t, err)
33 test.events.WaitFor(t, "idle", 1)
34 require.Equal(t, 2, server.Calls())
36 first := server.Call(0).body
37 second := server.Call(1).body
38 instructions, err := test.agent.Prompt()
39 require.NoError(t, err)
40 require.Equal(t, instructions, first["instructions"])
41 require.Equal(t, test.agent.config.Model, first["model"])
43 // The second request begins with the whole first one.
44 firstInput := first["input"].([]any)
45 secondInput := second["input"].([]any)
46 require.Equal(t, firstInput, secondInput[:len(firstInput)])
47 require.Equal(t, []any{
48 map[string]any{"type": "reasoning", "encrypted_content": "opaque thought", "summary": []any{}},
49 map[string]any{"type": "function_call", "call_id": "call_1", "name": "shell", "arguments": `{"command":"echo hi"}`},
50 map[string]any{"type": "function_call_output", "call_id": "call_1", "output": "hi\n"},
51 }, secondInput[len(firstInput):])
53 require.Equal(t, []string{
55 `tool: shell {"command":"echo hi"}`,
59 require.Equal(t, []string{"user", "busy", "tool", "echo", "reply", "idle"}, test.events.Kinds())
62 func TestFirstRequestHoldsTheViewTheStateAndTheMessage(t *testing.T) {
63 server := newFakeResponses(t)
64 test := newTestLoop(t, server)
65 _, err := test.memory.Append("user", "an earlier message")
66 require.NoError(t, err)
68 _, err = test.loop.Submit("The new message.")
69 require.NoError(t, err)
70 test.events.WaitFor(t, "idle", 1)
72 require.Equal(t, []string{
73 "<chat>\n0+1|user: an earlier message\n</chat>\n",
74 "Now: Thursday 2026-01-01 12:00 +00:00\n",
76 }, inputParts(t, server.Call(0)))
79 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
80 server := newFakeResponses(t)
81 arrived := make(chan struct{})
82 release := make(chan struct{})
83 server.reply = func(w http.ResponseWriter, call fakeCall) {
87 writeEvents(w, completedEvent([]any{functionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
92 writeEvents(w, completedEvent([]any{messageItem("done")}, 100, 0, 5))
95 test := newTestLoop(t, server)
96 _, err := test.loop.Submit("first")
97 require.NoError(t, err)
100 _, err = test.loop.Submit("second, during the turn")
101 require.NoError(t, err)
104 test.events.WaitFor(t, "idle", 1)
105 input := server.Call(1).body["input"].([]any)
106 require.Equal(t, map[string]any{
108 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
109 }, input[len(input)-1])
111 time.Sleep(50 * time.Millisecond)
112 require.Equal(t, 2, server.Calls())
113 require.Equal(t, 1, test.events.Count("busy"))
116 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
117 server := newFakeResponses(t)
118 arrived := make(chan struct{})
119 release := make(chan struct{})
120 server.reply = func(w http.ResponseWriter, call fakeCall) {
121 if call.number == 0 {
126 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
129 test := newTestLoop(t, server)
130 _, err := test.loop.Submit("first")
131 require.NoError(t, err)
134 _, err = test.loop.Submit("second")
135 require.NoError(t, err)
138 test.events.WaitFor(t, "idle", 2)
139 require.Equal(t, 2, server.Calls())
141 parts := inputParts(t, server.Call(1))
142 require.Equal(t, "second", parts[len(parts)-1])
143 require.Contains(t, parts[0], "|reply: answer\n")
146 func TestTurnWaitsForTheSummaries(t *testing.T) {
147 server := newFakeResponses(t)
148 summarizer := newFakeSummarizer()
149 summarizer.gate = make(chan struct{})
150 test := newTestLoopWith(t, server, summarizer)
152 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
153 require.NoError(t, err)
155 _, err = test.loop.Submit("go on")
156 require.NoError(t, err)
158 time.Sleep(50 * time.Millisecond)
159 require.Equal(t, 0, server.Calls())
161 close(summarizer.gate)
162 test.events.WaitFor(t, "idle", 1)
163 require.Equal(t, 1, server.Calls())
164 require.NotContains(t, inputParts(t, server.Call(0))[0], pending)
167 func TestInterruptEndsTheTurn(t *testing.T) {
168 server := newFakeResponses(t)
169 arrived := make(chan struct{})
170 release := make(chan struct{})
171 server.reply = func(w http.ResponseWriter, call fakeCall) {
172 if call.number == 0 {
177 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
180 test := newTestLoop(t, server)
181 _, err := test.loop.Submit("a long task")
182 require.NoError(t, err)
185 require.True(t, test.loop.Interrupt())
186 test.events.WaitFor(t, "idle", 1)
189 require.Equal(t, "interrupted", test.events.Last("error"))
190 require.Equal(t, []string{"user: a long task"}, test.Log(t))
191 require.False(t, test.loop.Interrupt())
193 _, err = test.loop.Submit("again")
194 require.NoError(t, err)
195 test.events.WaitFor(t, "idle", 2)
196 require.Equal(t, "reply: answer", test.Log(t)[2])
199 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
200 server := newFakeResponses(t)
201 server.reply = func(w http.ResponseWriter, call fakeCall) {
202 http.Error(w, "busy", http.StatusServiceUnavailable)
205 test := newTestLoop(t, server)
206 _, err := test.loop.Submit("hello")
207 require.NoError(t, err)
208 test.events.WaitFor(t, "idle", 1)
210 time.Sleep(50 * time.Millisecond)
211 require.Equal(t, 1, server.Calls())
212 require.Contains(t, test.events.Last("error"), "HTTP 503")
215 // testLoop is a loop that runs in the background of a test, with its parts.
216 type testLoop struct {
220 events *eventRecorder
223 // Log returns the logged messages as "kind: text".
224 func (l *testLoop) Log(t *testing.T) []string {
228 for id := range l.memory.Len() {
229 message, err := l.memory.Message(id)
230 require.NoError(t, err)
232 lines = append(lines, message.Kind+": "+message.Text)
238 // eventRecorder keeps the events of a loop.
239 type eventRecorder struct {
244 func (r *eventRecorder) Add(event Event) {
246 defer r.mutex.Unlock()
248 r.events = append(r.events, event)
251 func (r *eventRecorder) Kinds() []string {
253 defer r.mutex.Unlock()
255 kinds := make([]string, len(r.events))
256 for n, event := range r.events {
257 kinds[n] = event.Kind
263 func (r *eventRecorder) Count(kind string) int {
265 for _, each := range r.Kinds() {
274 // Last returns the text of the last event of kind.
275 func (r *eventRecorder) Last(kind string) string {
277 defer r.mutex.Unlock()
279 for n := len(r.events) - 1; n >= 0; n-- {
280 if r.events[n].Kind == kind {
281 return r.events[n].Text
288 // WaitFor waits until count events of kind came.
289 func (r *eventRecorder) 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 newTestLoop(t *testing.T, server *fakeResponses) *testLoop {
300 return newTestLoopWith(t, server, newFakeSummarizer())
303 // newTestLoopWith returns a running loop on a new agent whose builds use
304 // summarizer and whose commands run in a temporary directory.
305 func newTestLoopWith(t *testing.T, server *fakeResponses, summarizer Summarizer) *testLoop {
308 agent, err := openAgent(t.TempDir())
309 require.NoError(t, err)
311 agent.config.Workdir = t.TempDir()
312 clock := newTestClock()
313 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer)
314 require.NoError(t, err)
319 events: &eventRecorder{},
322 test.loop = newLoop(agent, memory, newTestClient(t, server), clock.Now, test.events.Add)
323 ctx, cancel := context.WithCancel(t.Context())
324 done := make(chan error)
326 done <- test.loop.Run(ctx)
331 require.True(t, errors.Is(<-done, context.Canceled))
332 require.NoError(t, memory.Close())
338 func reasoningItem(id, encrypted string) map[string]any {
339 return map[string]any{
342 "encrypted_content": encrypted,
347 func functionCallItem(id, callID, name, arguments string) map[string]any {
348 return map[string]any{
349 "type": "function_call",
353 "arguments": arguments,