| 1 | //go:build live |
| 2 | |
| 3 | package agent |
| 4 | |
| 5 | import ( |
| 6 | "bytes" |
| 7 | "context" |
| 8 | "encoding/json" |
| 9 | "fmt" |
| 10 | "io" |
| 11 | "net/http" |
| 12 | "net/http/httptest" |
| 13 | "os" |
| 14 | "strings" |
| 15 | "sync" |
| 16 | "sync/atomic" |
| 17 | "testing" |
| 18 | "time" |
| 19 | |
| 20 | "reasonix/internal/event" |
| 21 | "reasonix/internal/provider" |
| 22 | "reasonix/internal/provider/openai" |
| 23 | "reasonix/internal/tool" |
| 24 | ) |
| 25 | |
| 26 | // TestLiveDeepSeekFlashMissingReasoningRecovery exercises the production agent |
| 27 | // against DeepSeek's official API while a local proxy removes reasoning_content |
| 28 | // from one or two real tool-call responses. It is credential-gated and excluded |
| 29 | // from ordinary CI; response text, tool arguments, and credentials are never |
| 30 | // logged or written to disk. |
| 31 | func TestLiveDeepSeekFlashMissingReasoningRecovery(t *testing.T) { |
| 32 | key := os.Getenv("DEEPSEEK_API_KEY") |
| 33 | if key == "" { |
| 34 | t.Skip("DEEPSEEK_API_KEY not set") |
| 35 | } |
| 36 | |
| 37 | for _, tc := range []struct { |
| 38 | name string |
| 39 | stripResponses int32 |
| 40 | }{ |
| 41 | {name: "transient", stripResponses: 1}, |
| 42 | {name: "persistent", stripResponses: 2}, |
| 43 | } { |
| 44 | t.Run(tc.name, func(t *testing.T) { |
| 45 | result := runLiveDeepSeekRecoveryScenario(t, key, tc.stripResponses, 1) |
| 46 | t.Logf("mode=%s upstream_requests=%d stripped_fields=%d executions=%d tool_turns=%d warnings=%d retry_attempts=%d recovered=%d replaced=%d fallbacks=%d", |
| 47 | tc.name, result.requests, result.strippedFields, result.executions, result.toolTurns, |
| 48 | result.warnings, result.retryAttempts, result.recovered, result.replaced, result.fallbacks) |
| 49 | if result.strippedFields == 0 { |
| 50 | t.Fatal("official response contained no reasoning field to strip") |
| 51 | } |
| 52 | if result.executions != 1 || result.toolTurns != 1 || result.requests != 2 { |
| 53 | t.Fatalf("executions/tool turns/requests = %d/%d/%d, want 1/1/2", result.executions, result.toolTurns, result.requests) |
| 54 | } |
| 55 | if result.warnings != 0 || result.retryAttempts != 0 || result.recovered != 0 || result.replaced != 0 || result.fallbacks != 0 { |
| 56 | t.Fatal("compatible empty reasoning must not trigger a strict recovery incident") |
| 57 | } |
| 58 | }) |
| 59 | } |
| 60 | } |
| 61 | |
| 62 | type liveRecoveryResult struct { |
| 63 | requests, strippedFields int |
| 64 | executions, toolTurns, warnings int |
| 65 | retryAttempts, recovered, replaced int |
| 66 | fallbacks int |
| 67 | identicalRetry bool |
| 68 | } |
| 69 | |
| 70 | func runLiveDeepSeekRecoveryScenario(t *testing.T, key string, stripResponses int32, attempt int) liveRecoveryResult { |
| 71 | t.Helper() |
| 72 | proxy := &liveReasoningStripProxy{stripResponses: stripResponses} |
| 73 | server := httptest.NewServer(proxy) |
| 74 | defer server.Close() |
| 75 | |
| 76 | prov, err := openai.New(provider.Config{ |
| 77 | Name: "deepseek-live-recovery", |
| 78 | BaseURL: server.URL, |
| 79 | Model: "deepseek-v4-flash", |
| 80 | APIKey: key, |
| 81 | Extra: map[string]any{ |
| 82 | "api_key_env": "DEEPSEEK_API_KEY", |
| 83 | "reasoning_protocol": "deepseek", |
| 84 | "thinking": "enabled", |
| 85 | "effort": "low", |
| 86 | }, |
| 87 | }) |
| 88 | if err != nil { |
| 89 | t.Fatalf("create live provider: %v", err) |
| 90 | } |
| 91 | |
| 92 | var executions atomic.Int32 |
| 93 | registry := tool.NewRegistry() |
| 94 | registry.Add(liveRecoveryEchoTool{executions: &executions}) |
| 95 | sink := &recordSink{} |
| 96 | a := New(prov, registry, NewSession("You are a concise tool-using assistant."), Options{ |
| 97 | MaxSteps: 4, |
| 98 | MissingReasoningWarnStateDir: t.TempDir(), |
| 99 | }, sink) |
| 100 | ctx, cancel := context.WithTimeout(context.Background(), 120*time.Second) |
| 101 | defer cancel() |
| 102 | if err := a.Run(ctx, fmt.Sprintf("Live recovery probe %d: call echo exactly once, then report that it completed.", attempt)); err != nil { |
| 103 | t.Fatalf("live agent run: %v", err) |
| 104 | } |
| 105 | |
| 106 | result := liveRecoveryResult{executions: int(executions.Load())} |
| 107 | for _, msg := range a.Session().Snapshot() { |
| 108 | if msg.Role == provider.RoleAssistant && len(msg.ToolCalls) > 0 { |
| 109 | result.toolTurns++ |
| 110 | } |
| 111 | } |
| 112 | for _, notice := range sink.kinds(event.Notice) { |
| 113 | text := strings.ToLower(notice.Text + " " + notice.Detail) |
| 114 | if strings.Contains(text, "reasoning_content") || strings.Contains(text, "replayable thinking") { |
| 115 | result.warnings++ |
| 116 | } |
| 117 | } |
| 118 | result.retryAttempts = sink.recoveryCount(event.ProtocolRecoveryMissingReasoningRetryAttempted) |
| 119 | result.recovered = sink.recoveryCount(event.ProtocolRecoveryMissingReasoningRetryRecovered) |
| 120 | result.replaced = sink.recoveryCount(event.ProtocolRecoveryMissingReasoningRetryReplaced) |
| 121 | result.fallbacks = sink.recoveryCount(event.ProtocolRecoveryMissingReasoningFallback) |
| 122 | result.requests = int(proxy.requests.Load()) |
| 123 | result.strippedFields = int(proxy.strippedFields.Load()) |
| 124 | proxy.mu.Lock() |
| 125 | result.identicalRetry = len(proxy.firstBody) > 0 && bytes.Equal(proxy.firstBody, proxy.retryBody) |
| 126 | proxy.mu.Unlock() |
| 127 | return result |
| 128 | } |
| 129 | |
| 130 | type liveRecoveryEchoTool struct{ executions *atomic.Int32 } |
| 131 | |
| 132 | func (t liveRecoveryEchoTool) Name() string { return "echo" } |
| 133 | func (t liveRecoveryEchoTool) Description() string { return "Return a fixed live-test marker." } |
| 134 | func (t liveRecoveryEchoTool) Schema() json.RawMessage { |
| 135 | return json.RawMessage(`{"type":"object","properties":{},"additionalProperties":false}`) |
| 136 | } |
| 137 | func (t liveRecoveryEchoTool) ReadOnly() bool { return true } |
| 138 | func (t liveRecoveryEchoTool) Execute(context.Context, json.RawMessage) (string, error) { |
| 139 | t.executions.Add(1) |
| 140 | return "live recovery marker", nil |
| 141 | } |
| 142 | |
| 143 | type liveReasoningStripProxy struct { |
| 144 | stripResponses int32 |
| 145 | requests atomic.Int32 |
| 146 | toolResponses atomic.Int32 |
| 147 | strippedFields atomic.Int32 |
| 148 | mu sync.Mutex |
| 149 | firstRequestNo int32 |
| 150 | firstBody []byte |
| 151 | retryBody []byte |
| 152 | } |
| 153 | |
| 154 | func (p *liveReasoningStripProxy) ServeHTTP(w http.ResponseWriter, r *http.Request) { |
| 155 | body, err := io.ReadAll(r.Body) |
| 156 | if err != nil { |
| 157 | http.Error(w, "read request", http.StatusBadRequest) |
| 158 | return |
| 159 | } |
| 160 | requestNo := p.requests.Add(1) |
| 161 | p.mu.Lock() |
| 162 | if p.firstRequestNo != 0 && requestNo == p.firstRequestNo+1 { |
| 163 | p.retryBody = append([]byte(nil), body...) |
| 164 | } |
| 165 | p.mu.Unlock() |
| 166 | |
| 167 | upstream, err := http.NewRequestWithContext(r.Context(), http.MethodPost, |
| 168 | "https://api.deepseek.com/chat/completions", bytes.NewReader(body)) |
| 169 | if err != nil { |
| 170 | http.Error(w, "create upstream request", http.StatusInternalServerError) |
| 171 | return |
| 172 | } |
| 173 | upstream.Header.Set("Authorization", r.Header.Get("Authorization")) |
| 174 | upstream.Header.Set("Content-Type", "application/json") |
| 175 | resp, err := (&http.Client{Timeout: 90 * time.Second}).Do(upstream) |
| 176 | if err != nil { |
| 177 | http.Error(w, "upstream request failed", http.StatusBadGateway) |
| 178 | return |
| 179 | } |
| 180 | defer resp.Body.Close() |
| 181 | responseBody, err := io.ReadAll(resp.Body) |
| 182 | if err != nil { |
| 183 | http.Error(w, "read upstream response", http.StatusBadGateway) |
| 184 | return |
| 185 | } |
| 186 | if resp.StatusCode == http.StatusOK && bytes.Contains(responseBody, []byte(`"tool_calls"`)) { |
| 187 | toolResponse := p.toolResponses.Add(1) |
| 188 | if toolResponse <= p.stripResponses { |
| 189 | if toolResponse == 1 { |
| 190 | p.mu.Lock() |
| 191 | p.firstRequestNo = requestNo |
| 192 | p.firstBody = append([]byte(nil), body...) |
| 193 | p.mu.Unlock() |
| 194 | } |
| 195 | responseBody = p.stripReasoning(responseBody) |
| 196 | } |
| 197 | } |
| 198 | w.Header().Set("Content-Type", resp.Header.Get("Content-Type")) |
| 199 | w.WriteHeader(resp.StatusCode) |
| 200 | _, _ = w.Write(responseBody) |
| 201 | } |
| 202 | |
| 203 | func (p *liveReasoningStripProxy) stripReasoning(body []byte) []byte { |
| 204 | lines := bytes.Split(body, []byte("\n")) |
| 205 | for i, line := range lines { |
| 206 | if !bytes.HasPrefix(line, []byte("data: ")) { |
| 207 | continue |
| 208 | } |
| 209 | data := bytes.TrimSpace(bytes.TrimPrefix(line, []byte("data: "))) |
| 210 | if bytes.Equal(data, []byte("[DONE]")) { |
| 211 | continue |
| 212 | } |
| 213 | var payload map[string]any |
| 214 | if json.Unmarshal(data, &payload) != nil { |
| 215 | continue |
| 216 | } |
| 217 | choices, _ := payload["choices"].([]any) |
| 218 | changed := false |
| 219 | for _, rawChoice := range choices { |
| 220 | choice, _ := rawChoice.(map[string]any) |
| 221 | delta, _ := choice["delta"].(map[string]any) |
| 222 | for _, field := range []string{"reasoning_content", "reasoning"} { |
| 223 | if _, ok := delta[field]; ok { |
| 224 | delete(delta, field) |
| 225 | p.strippedFields.Add(1) |
| 226 | changed = true |
| 227 | } |
| 228 | } |
| 229 | } |
| 230 | if changed { |
| 231 | encoded, err := json.Marshal(payload) |
| 232 | if err == nil { |
| 233 | lines[i] = append([]byte("data: "), encoded...) |
| 234 | } |
| 235 | } |
| 236 | } |
| 237 | return bytes.Join(lines, []byte("\n")) |
| 238 | } |
| 239 |