| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "io" |
| 7 | "net/http" |
| 8 | "net/http/httptest" |
| 9 | "sync" |
| 10 | "testing" |
| 11 | |
| 12 | "reasonix/internal/event" |
| 13 | "reasonix/internal/provider" |
| 14 | "reasonix/internal/provider/anthropic" |
| 15 | ) |
| 16 | |
| 17 | func TestCustomAnthropicCompatibilityToolLoop(t *testing.T) { |
| 18 | for _, tc := range []struct { |
| 19 | name, first string |
| 20 | rejectFollowup bool |
| 21 | }{ |
| 22 | {"complete missing thinking", missingReasoningToolSSE, false}, |
| 23 | {"complete unsigned thinking", recoveredReasoningToolSSE, false}, |
| 24 | {"server rejects unsigned history", recoveredReasoningToolSSE, true}, |
| 25 | } { |
| 26 | t.Run(tc.name, func(t *testing.T) { |
| 27 | var mu sync.Mutex |
| 28 | var bodies [][]byte |
| 29 | srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 30 | body, err := io.ReadAll(r.Body) |
| 31 | if err != nil { |
| 32 | t.Error(err) |
| 33 | w.WriteHeader(http.StatusBadRequest) |
| 34 | return |
| 35 | } |
| 36 | mu.Lock() |
| 37 | bodies = append(bodies, body) |
| 38 | n := len(bodies) |
| 39 | mu.Unlock() |
| 40 | if n == 2 && tc.rejectFollowup { |
| 41 | w.WriteHeader(http.StatusBadRequest) |
| 42 | _, _ = io.WriteString(w, `{"error":{"message":"The content[].thinking in the thinking mode must be passed back to the API"}}`) |
| 43 | return |
| 44 | } |
| 45 | if n > 3 { |
| 46 | t.Errorf("unexpected HTTP attempt %d", n) |
| 47 | w.WriteHeader(http.StatusUnauthorized) |
| 48 | return |
| 49 | } |
| 50 | w.Header().Set("Content-Type", "text/event-stream") |
| 51 | response := finalAnswerSSE |
| 52 | if n == 1 { |
| 53 | response = tc.first |
| 54 | } |
| 55 | _, _ = io.WriteString(w, response) |
| 56 | })) |
| 57 | defer srv.Close() |
| 58 | p, err := anthropic.New(provider.Config{Name: "custom-anthropic", BaseURL: srv.URL, Model: "deepseek-v4-flash", APIKey: "fake-key", Extra: map[string]any{"thinking": "adaptive"}}) |
| 59 | if err != nil { |
| 60 | t.Fatal(err) |
| 61 | } |
| 62 | sink := &recordSink{} |
| 63 | a := New(p, echoRegistry(), NewSession(""), Options{MissingReasoningWarnStateDir: t.TempDir()}, sink) |
| 64 | if err := a.Run(withNoClosedLoop(context.Background()), "go"); err != nil { |
| 65 | t.Fatal(err) |
| 66 | } |
| 67 | mu.Lock() |
| 68 | defer mu.Unlock() |
| 69 | expected := 2 |
| 70 | if tc.rejectFollowup { |
| 71 | expected = 3 |
| 72 | } |
| 73 | if len(bodies) != expected || len(sink.kinds(event.ToolResult)) != 1 { |
| 74 | t.Fatalf("HTTP=%d tools=%d", len(bodies), len(sink.kinds(event.ToolResult))) |
| 75 | } |
| 76 | if !bytes.Contains(bodies[1], []byte(`"type":"tool_result"`)) { |
| 77 | t.Fatalf("follow-up omitted completed tool: %s", bodies[1]) |
| 78 | } |
| 79 | if tc.first == recoveredReasoningToolSSE && !bytes.Contains(bodies[1], []byte(`"thinking":"call echo safely"`)) { |
| 80 | t.Fatalf("lost unsigned thinking: %s", bodies[1]) |
| 81 | } |
| 82 | if bytes.Contains(bodies[1], []byte(`"signature"`)) { |
| 83 | t.Fatal("fabricated a signature") |
| 84 | } |
| 85 | if tc.rejectFollowup { |
| 86 | if !bytes.Contains(bodies[2], []byte("completed_tools")) || bytes.Contains(bodies[2], []byte(`"type":"tool_use"`)) { |
| 87 | t.Fatalf("invalid recovery view: %s", bodies[2]) |
| 88 | } |
| 89 | } else if len(sink.kinds(event.Retrying)) != 0 { |
| 90 | t.Fatal("compatible turn regenerated") |
| 91 | } |
| 92 | var originals int |
| 93 | for _, m := range a.Session().Snapshot() { |
| 94 | if m.Role == provider.RoleAssistant && m.ReasoningContent == "call echo safely" { |
| 95 | originals++ |
| 96 | } |
| 97 | } |
| 98 | if tc.first == recoveredReasoningToolSSE && originals != 1 { |
| 99 | t.Fatal("canonical thinking not retained") |
| 100 | } |
| 101 | }) |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | func TestNativeTextConversionDoesNotClaimMissingReasoningIncident(t *testing.T) { |
| 106 | p, err := anthropic.New(provider.Config{Name: "native", Model: "claude-sonnet-4-6", APIKey: "fake-key", Extra: map[string]any{"thinking": "adaptive"}}) |
| 107 | if err != nil { |
| 108 | t.Fatal(err) |
| 109 | } |
| 110 | a := New(p, echoRegistry(), NewSession(""), Options{MissingReasoningWarnStateDir: t.TempDir()}, event.Discard) |
| 111 | m := provider.Message{Role: provider.RoleAssistant, ReasoningContent: "complete unsigned text", ReasoningState: provider.ReasoningComplete} |
| 112 | if missing, retry := a.observeMissingAssistantReasoning(m, true); missing || retry || a.sess.missingReasoning.active { |
| 113 | t.Fatal("compatible text consumed strict recovery budget") |
| 114 | } |
| 115 | } |
| 116 |