| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "encoding/json" |
| 5 | "path/filepath" |
| 6 | "testing" |
| 7 | |
| 8 | "reasonix/internal/checkpoint" |
| 9 | "reasonix/internal/event" |
| 10 | "reasonix/internal/evidence" |
| 11 | "reasonix/internal/provider" |
| 12 | "reasonix/internal/sessiontemp" |
| 13 | "reasonix/internal/tool" |
| 14 | ) |
| 15 | |
| 16 | func TestScratchScriptExecutionKeepsCheckpointCoverageFailClosed(t *testing.T) { |
| 17 | workspace := t.TempDir() |
| 18 | scratch := t.TempDir() |
| 19 | script := filepath.Join(scratch, "probe.py") |
| 20 | store := checkpoint.New("", workspace) |
| 21 | store.Begin(1, "probe", 0) |
| 22 | observer := checkpoint.NewMutationObserver(checkpoint.ObserverOptions{Store: store}) |
| 23 | a := New(nil, tool.NewRegistry(), NewSession(""), Options{ |
| 24 | WriteWorkspaceRoot: workspace, |
| 25 | MutationObserver: observer, |
| 26 | }, event.Discard) |
| 27 | args, err := json.Marshal(map[string]string{"command": "python " + script}) |
| 28 | if err != nil { |
| 29 | t.Fatal(err) |
| 30 | } |
| 31 | plan := &toolCallPlan{ |
| 32 | call: provider.ToolCall{Name: "bash", Arguments: string(args)}, |
| 33 | tool: fakeTool{name: "bash", readOnly: false}, |
| 34 | evidenceName: "bash", |
| 35 | evidenceArgs: args, |
| 36 | } |
| 37 | a.observeBeforeMutation(t.Context(), plan) |
| 38 | meta := store.List() |
| 39 | if len(meta) != 1 || len(meta[0].CoverageGaps) != 1 || meta[0].CoverageGaps[0].Reason != checkpoint.GapBashSideEffect { |
| 40 | t.Fatalf("coverage gaps = %+v, want bash_side_effect", meta) |
| 41 | } |
| 42 | } |
| 43 | |
| 44 | func TestRecordToolReceiptsKeepsScratchScriptExecutionFailClosed(t *testing.T) { |
| 45 | workspace := t.TempDir() |
| 46 | manager := sessiontemp.NewWithRoot(t.TempDir()) |
| 47 | manager.Retain() |
| 48 | defer manager.Release() |
| 49 | lease, err := manager.Acquire() |
| 50 | if err != nil { |
| 51 | t.Fatal(err) |
| 52 | } |
| 53 | script := filepath.Join(lease.Dir(), "probe.py") |
| 54 | defer lease.Release() |
| 55 | |
| 56 | a := New(nil, tool.NewRegistry(), NewSession(""), Options{ |
| 57 | WriteWorkspaceRoot: workspace, |
| 58 | SessionTemp: manager, |
| 59 | }, event.Discard) |
| 60 | for _, background := range []bool{false, true} { |
| 61 | args, err := json.Marshal(map[string]any{ |
| 62 | "command": "python " + script, |
| 63 | "run_in_background": background, |
| 64 | }) |
| 65 | if err != nil { |
| 66 | t.Fatal(err) |
| 67 | } |
| 68 | plan := &toolCallPlan{ |
| 69 | call: provider.ToolCall{Name: "bash", Arguments: string(args)}, |
| 70 | tool: fakeTool{name: "bash", readOnly: false}, |
| 71 | evidenceName: "bash", |
| 72 | evidenceArgs: args, |
| 73 | effects: evidence.ClassifyToolCall("bash", args, false), |
| 74 | } |
| 75 | a.recordToolReceipts(plan, "ok", nil, nil) |
| 76 | receipts := a.task.ledger.Receipts() |
| 77 | got := receipts[len(receipts)-1] |
| 78 | if got.DeliveryScope == evidence.WriteScopeScratch { |
| 79 | t.Fatalf("background=%v receipt = %+v, script location cannot prove its side effects", background, got) |
| 80 | } |
| 81 | } |
| 82 | } |
| 83 |