| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "strings" |
| 7 | "sync/atomic" |
| 8 | "testing" |
| 9 | |
| 10 | "reasonix/internal/event" |
| 11 | "reasonix/internal/provider" |
| 12 | "reasonix/internal/tool" |
| 13 | ) |
| 14 | |
| 15 | type unavailableTool struct { |
| 16 | fakeTool |
| 17 | calls *int32 |
| 18 | } |
| 19 | |
| 20 | func (u unavailableTool) ProviderVisible(context.Context) bool { return false } |
| 21 | |
| 22 | func (u unavailableTool) Execute(ctx context.Context, args json.RawMessage) (string, error) { |
| 23 | if u.calls != nil { |
| 24 | atomic.AddInt32(u.calls, 1) |
| 25 | } |
| 26 | return u.fakeTool.Execute(ctx, args) |
| 27 | } |
| 28 | |
| 29 | func TestContextualToolHostGateRunsBeforePermissionAndExecute(t *testing.T) { |
| 30 | var executions int32 |
| 31 | reg := tool.NewRegistry() |
| 32 | reg.Add(unavailableTool{fakeTool: fakeTool{name: "phase_tool", readOnly: false}, calls: &executions}) |
| 33 | gate := &recordingPermissionGate{allow: true} |
| 34 | a := New(nil, reg, NewSession("sys"), Options{Gate: gate}, event.Discard) |
| 35 | |
| 36 | out := a.executeOne(context.Background(), &a.turn, provider.ToolCall{ID: "phase", Name: "phase_tool", Arguments: `{}`}) |
| 37 | if !out.blocked || !strings.Contains(out.output, "unavailable") { |
| 38 | t.Fatalf("contextual tool outcome = %+v", out) |
| 39 | } |
| 40 | if executions != 0 || len(gate.calls) != 0 { |
| 41 | t.Fatalf("unavailable tool crossed host gate: executions=%d permission=%+v", executions, gate.calls) |
| 42 | } |
| 43 | } |
| 44 | |
| 45 | func TestMixedContextualBatchExecutesAvailableCallsOnce(t *testing.T) { |
| 46 | var executions int32 |
| 47 | reg := tool.NewRegistry() |
| 48 | reg.Add(unavailableTool{fakeTool: fakeTool{name: "phase_tool", readOnly: true}, calls: &executions}) |
| 49 | reg.Add(fakeTool{name: "read_file", readOnly: true, calls: &executions}) |
| 50 | prov := &scriptedProvider{name: "p", turns: [][]provider.Chunk{ |
| 51 | {toolCallChunk("phase", "phase_tool", `{}`), toolCallChunk("read", "read_file", `{}`), {Type: provider.ChunkDone}}, |
| 52 | {{Type: provider.ChunkText, Text: "answer after the available read"}, {Type: provider.ChunkDone}}, |
| 53 | }} |
| 54 | a := New(prov, reg, NewSession("sys"), Options{}, event.Discard) |
| 55 | if err := a.Run(context.Background(), "inspect the file"); err != nil { |
| 56 | t.Fatalf("mixed contextual batch failed: %v", err) |
| 57 | } |
| 58 | if got := atomic.LoadInt32(&executions); got != 1 { |
| 59 | t.Fatalf("available tool executions = %d, want exactly one", got) |
| 60 | } |
| 61 | if got := lastToolResult(a.Session(), "phase_tool"); !strings.Contains(got, "unavailable") { |
| 62 | t.Fatalf("contextual result = %q", got) |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | func TestRepeatedMixedContextualBatchKeepsIndependentCallsRunning(t *testing.T) { |
| 67 | var executions int32 |
| 68 | reg := tool.NewRegistry() |
| 69 | reg.Add(unavailableTool{fakeTool: fakeTool{name: "phase_tool", readOnly: true}, calls: &executions}) |
| 70 | reg.Add(fakeTool{name: "read_file", readOnly: true, calls: &executions}) |
| 71 | prov := &scriptedProvider{name: "p", turns: [][]provider.Chunk{ |
| 72 | {toolCallChunk("phase-1", "phase_tool", `{}`), toolCallChunk("read-1", "read_file", `{}`), {Type: provider.ChunkDone}}, |
| 73 | {toolCallChunk("phase-2", "phase_tool", `{}`), toolCallChunk("read-2", "read_file", `{}`), {Type: provider.ChunkDone}}, |
| 74 | {{Type: provider.ChunkText, Text: "answer after repeated local failures"}, {Type: provider.ChunkDone}}, |
| 75 | }} |
| 76 | a := New(prov, reg, NewSession("sys"), Options{}, event.Discard) |
| 77 | if err := a.Run(context.Background(), "inspect the file"); err != nil { |
| 78 | t.Fatalf("repeated contextual batch failed: %v", err) |
| 79 | } |
| 80 | if got := atomic.LoadInt32(&executions); got != 2 { |
| 81 | t.Fatalf("available tool executions = %d, want both independent calls", got) |
| 82 | } |
| 83 | if got := lastToolResult(a.Session(), "phase_tool"); !strings.Contains(got, "unavailable") { |
| 84 | t.Fatalf("second unavailable call result = %q", got) |
| 85 | } |
| 86 | if prov.call != 3 { |
| 87 | t.Fatalf("provider calls = %d, want a normal final round", prov.call) |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | func TestRepeatedPureContextualCallWithAnswerRemainsRecoverable(t *testing.T) { |
| 92 | reg := tool.NewRegistry() |
| 93 | reg.Add(unavailableTool{fakeTool: fakeTool{name: "phase_tool", readOnly: true}}) |
| 94 | prov := &scriptedProvider{name: "p", turns: [][]provider.Chunk{ |
| 95 | {toolCallChunk("phase-1", "phase_tool", `{}`), {Type: provider.ChunkDone}}, |
| 96 | {{Type: provider.ChunkText, Text: "占位 Lorem 占位 ipsum — the request is fully handled."}, toolCallChunk("phase-2", "phase_tool", `{}`), {Type: provider.ChunkDone}}, |
| 97 | {{Type: provider.ChunkText, Text: "final answer"}, {Type: provider.ChunkDone}}, |
| 98 | }} |
| 99 | a := New(prov, reg, NewSession("sys"), Options{}, event.Discard) |
| 100 | if err := a.Run(context.Background(), "answer normally"); err != nil { |
| 101 | t.Fatalf("repeated contextual call failed: %v", err) |
| 102 | } |
| 103 | if prov.call != 3 { |
| 104 | t.Fatalf("provider calls = %d, want a normal final round", prov.call) |
| 105 | } |
| 106 | if got := lastToolResult(a.Session(), "phase_tool"); !strings.Contains(got, "unavailable") { |
| 107 | t.Fatalf("second unavailable call result = %q", got) |
| 108 | } |
| 109 | } |
| 110 |