| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "testing" |
| 8 | |
| 9 | "reasonix/internal/agent/testutil" |
| 10 | "reasonix/internal/event" |
| 11 | "reasonix/internal/evidence" |
| 12 | "reasonix/internal/provider" |
| 13 | "reasonix/internal/tool" |
| 14 | |
| 15 | _ "reasonix/internal/tool/builtin" |
| 16 | ) |
| 17 | |
| 18 | func TestTodoProgressGuardPausesSemanticToolDrift(t *testing.T) { |
| 19 | turns := []testutil.Turn{{ToolCalls: []provider.ToolCall{{ |
| 20 | ID: "todo", Name: "todo_write", |
| 21 | Arguments: `{"todos":[{"content":"finish the task","status":"in_progress"}]}`, |
| 22 | }}}} |
| 23 | // The first unique read renews the lease; exact repeats after it do not. |
| 24 | for i := 0; i < maxTodoStallRounds+1; i++ { |
| 25 | turns = append(turns, testutil.Turn{ToolCalls: []provider.ToolCall{{ |
| 26 | ID: fmt.Sprintf("read-%d", i), Name: "inspect", Arguments: `{"path":"same"}`, |
| 27 | }}}) |
| 28 | } |
| 29 | |
| 30 | reg := tool.NewRegistry() |
| 31 | reg.Add(fakeTool{name: "inspect", readOnly: true}) |
| 32 | reg.Add(mustBuiltinTool(t, "todo_write")) |
| 33 | mp := testutil.NewMock("m", turns...) |
| 34 | a := New(mp, reg, NewSession(""), Options{}, event.Discard) |
| 35 | |
| 36 | err := a.Run(context.Background(), "work until the todo is complete") |
| 37 | var pause *todoStallPause |
| 38 | if !errors.As(err, &pause) { |
| 39 | t.Fatalf("Run error = %v, want todoStallPause", err) |
| 40 | } |
| 41 | if mp.CallCount() != maxTodoStallRounds+2 { |
| 42 | t.Fatalf("provider calls = %d, want %d", mp.CallCount(), maxTodoStallRounds+2) |
| 43 | } |
| 44 | if !sessionContains(a, "Host progress check") { |
| 45 | t.Fatal("semantic drift did not receive the adaptive progress nudge") |
| 46 | } |
| 47 | } |
| 48 | |
| 49 | func TestTodoProgressGuardRenewsOnUniqueHostWork(t *testing.T) { |
| 50 | turns := []testutil.Turn{{ToolCalls: []provider.ToolCall{{ |
| 51 | ID: "todo", Name: "todo_write", |
| 52 | Arguments: `{"todos":[{"content":"finish the task","status":"in_progress"}]}`, |
| 53 | }}}} |
| 54 | for i := 0; i < todoProgressNudgeRounds-1; i++ { |
| 55 | turns = append(turns, testutil.Turn{ToolCalls: []provider.ToolCall{{ |
| 56 | ID: fmt.Sprintf("read-a-%d", i), Name: "inspect", Arguments: `{"path":"same"}`, |
| 57 | }}}) |
| 58 | } |
| 59 | turns = append(turns, |
| 60 | testutil.Turn{ToolCalls: []provider.ToolCall{{ID: "write", Name: "write_file", Arguments: `{"path":"result.txt","content":"done"}`}}}, |
| 61 | ) |
| 62 | for i := 0; i < todoProgressNudgeRounds-1; i++ { |
| 63 | turns = append(turns, testutil.Turn{ToolCalls: []provider.ToolCall{{ |
| 64 | ID: fmt.Sprintf("read-b-%d", i), Name: "inspect", Arguments: `{"path":"same"}`, |
| 65 | }}}) |
| 66 | } |
| 67 | turns = append(turns, |
| 68 | testutil.Turn{ToolCalls: []provider.ToolCall{{ |
| 69 | ID: "done", Name: "complete_step", |
| 70 | Arguments: `{"step":"finish the task","result":"done","evidence":[{"kind":"files","summary":"created result","paths":["result.txt"]}]}`, |
| 71 | }}}, |
| 72 | testutil.Turn{Text: "done"}, |
| 73 | ) |
| 74 | |
| 75 | reg := tool.NewRegistry() |
| 76 | reg.Add(fakeTool{name: "inspect", readOnly: true}) |
| 77 | reg.Add(fakeTool{name: "write_file", readOnly: false}) |
| 78 | reg.Add(mustBuiltinTool(t, "todo_write")) |
| 79 | reg.Add(mustBuiltinTool(t, "complete_step")) |
| 80 | a := New(testutil.NewMock("m", turns...), reg, NewSession(""), Options{}, event.Discard) |
| 81 | if err := a.Run(context.Background(), "finish the todo"); err != nil { |
| 82 | t.Fatalf("Run: %v", err) |
| 83 | } |
| 84 | if sessionContains(a, "Host progress check") { |
| 85 | t.Fatal("unique host work should renew the progress lease before the nudge threshold") |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | func TestCanonicalTodoProgressIgnoresTitleAndPendingListChurn(t *testing.T) { |
| 90 | a := &Agent{todoState: []evidence.TodoItem{ |
| 91 | {Content: "finish the task", Status: "in_progress"}, |
| 92 | {Content: "write tests", Status: "pending"}, |
| 93 | }} |
| 94 | before, tracking := a.canonicalTodoProgress() |
| 95 | if !tracking { |
| 96 | t.Fatal("incomplete todo list should be tracked") |
| 97 | } |
| 98 | a.setTodoState([]evidence.TodoItem{ |
| 99 | {Content: "finish the task carefully", Status: "in_progress"}, |
| 100 | {Content: "write tests", Status: "pending"}, |
| 101 | {Content: "update docs", Status: "pending"}, |
| 102 | }) |
| 103 | after, tracking := a.canonicalTodoProgress() |
| 104 | if !tracking || after != before { |
| 105 | t.Fatalf("title/pending churn changed progress from %d to %d", before, after) |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | func TestMaxStepsGraceSummaryBypassesIncompleteTodoReadiness(t *testing.T) { |
| 110 | reg := tool.NewRegistry() |
| 111 | reg.Add(mustBuiltinTool(t, "todo_write")) |
| 112 | reg.Add(fakeTool{name: "write_file", readOnly: false}) |
| 113 | mp := testutil.NewMock("m", |
| 114 | testutil.Turn{ToolCalls: []provider.ToolCall{ |
| 115 | {ID: "todo", Name: "todo_write", Arguments: `{"todos":[{"content":"unfinished","status":"in_progress"}]}`}, |
| 116 | {ID: "write", Name: "write_file", Arguments: `{"path":"unfinished.txt"}`}, |
| 117 | }}, |
| 118 | testutil.Turn{Text: "Progress saved; the todo remains unfinished."}, |
| 119 | ) |
| 120 | a := New(mp, reg, NewSession(""), Options{MaxSteps: 1, DeliveryProfile: true}, event.Discard) |
| 121 | |
| 122 | err := a.Run(context.Background(), "start a long task") |
| 123 | var pause *maxStepsPause |
| 124 | if !errors.As(err, &pause) { |
| 125 | t.Fatalf("Run error = %v, want maxStepsPause instead of final-readiness retries", err) |
| 126 | } |
| 127 | if mp.CallCount() != 2 { |
| 128 | t.Fatalf("provider calls = %d, want tool round plus one summary round", mp.CallCount()) |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | func mustBuiltinTool(t *testing.T, name string) tool.Tool { |
| 133 | t.Helper() |
| 134 | builtin, ok := tool.LookupBuiltin(name) |
| 135 | if !ok { |
| 136 | t.Fatalf("builtin %q is not registered", name) |
| 137 | } |
| 138 | return builtin |
| 139 | } |
| 140 |