| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "encoding/json" |
| 5 | "strings" |
| 6 | "testing" |
| 7 | |
| 8 | "reasonix/internal/evidence" |
| 9 | "reasonix/internal/tool" |
| 10 | ) |
| 11 | |
| 12 | func TestDeliveryReviewGateExplainsOpaqueMutationRecovery(t *testing.T) { |
| 13 | ledger := evidence.NewLedger() |
| 14 | ledger.Record(evidence.Receipt{ |
| 15 | ToolName: "bash", |
| 16 | Success: true, |
| 17 | Mutation: true, |
| 18 | Command: "opaque-writer", |
| 19 | }) |
| 20 | |
| 21 | reg := tool.NewRegistry() |
| 22 | reg.Add(fakeTool{name: "review", readOnly: true}) |
| 23 | reg.Add(fakeTool{name: "security_review", readOnly: true}) |
| 24 | a := &Agent{deliveryProfile: true, evidence: ledger, tools: reg} |
| 25 | |
| 26 | got := a.deliveryReviewGateFailure() |
| 27 | for _, want := range []string{"high-risk", "git status --short", "git diff", "mutation did not report file paths"} { |
| 28 | if !strings.Contains(got, want) { |
| 29 | t.Fatalf("review gate = %q, want %q", got, want) |
| 30 | } |
| 31 | } |
| 32 | if strings.HasSuffix(got, "covering: ") { |
| 33 | t.Fatalf("review gate must not end with empty coverage: %q", got) |
| 34 | } |
| 35 | |
| 36 | ledger.Record(evidence.Receipt{ToolName: "review_report", Success: true, Args: json.RawMessage(`{ |
| 37 | "kind":"review", |
| 38 | "verdict":"pass", |
| 39 | "reviewed_paths":["internal/agent/agent.go"], |
| 40 | "findings":[] |
| 41 | }`)}) |
| 42 | got = a.deliveryReviewGateFailure() |
| 43 | if !strings.Contains(got, "security_review") || !strings.Contains(got, "mutation did not report file paths") { |
| 44 | t.Fatalf("security review gate = %q, want opaque-mutation recovery guidance", got) |
| 45 | } |
| 46 | |
| 47 | ledger.Record(evidence.Receipt{ToolName: "review_report", Success: true, Args: json.RawMessage(`{ |
| 48 | "kind":"security", |
| 49 | "verdict":"pass", |
| 50 | "reviewed_paths":["internal/agent/agent.go"], |
| 51 | "findings":[] |
| 52 | }`)}) |
| 53 | if got := a.deliveryReviewGateFailure(); got != "" { |
| 54 | t.Fatalf("review gate = %q after both reports, want ready", got) |
| 55 | } |
| 56 | } |
| 57 | |
| 58 | func TestNonDeliveryProfileNeverRequiresStructuredReview(t *testing.T) { |
| 59 | ledger := evidence.NewLedger() |
| 60 | ledger.Record(evidence.ReceiptFromToolCall("edit_file", json.RawMessage(`{"path":"internal/permission/gate.go"}`), true, false)) |
| 61 | |
| 62 | reg := tool.NewRegistry() |
| 63 | reg.Add(fakeTool{name: "review", readOnly: true}) |
| 64 | reg.Add(fakeTool{name: "security_review", readOnly: true}) |
| 65 | a := &Agent{deliveryProfile: false, evidence: ledger, tools: reg} |
| 66 | |
| 67 | if got := a.deliveryReviewGateFailure(); got != "" { |
| 68 | t.Fatalf("non-Delivery review gate = %q, want disabled", got) |
| 69 | } |
| 70 | } |
| 71 | |
| 72 | func TestDeliveryReviewGateHighRiskStillRequiresSecurityReview(t *testing.T) { |
| 73 | ledger := evidence.NewLedger() |
| 74 | ledger.Record(evidence.ReceiptFromToolCall("edit_file", json.RawMessage(`{"path":"internal/permission/gate.go"}`), true, false)) |
| 75 | |
| 76 | reg := tool.NewRegistry() |
| 77 | reg.Add(fakeTool{name: "review", readOnly: true}) |
| 78 | reg.Add(fakeTool{name: "security_review", readOnly: true}) |
| 79 | a := &Agent{deliveryProfile: true, evidence: ledger, tools: reg} |
| 80 | |
| 81 | if got := a.deliveryReviewGateFailure(); !strings.Contains(got, "high-risk") { |
| 82 | t.Fatalf("review gate = %q, want high-risk review demand", got) |
| 83 | } |
| 84 | |
| 85 | ledger.Record(evidence.Receipt{ToolName: "review_report", Success: true, Args: json.RawMessage(`{ |
| 86 | "kind":"review", |
| 87 | "verdict":"pass", |
| 88 | "reviewed_paths":["internal/permission/gate.go"], |
| 89 | "findings":[] |
| 90 | }`)}) |
| 91 | if got := a.deliveryReviewGateFailure(); !strings.Contains(got, "security_review") { |
| 92 | t.Fatalf("security review gate = %q, want security_review demand", got) |
| 93 | } |
| 94 | |
| 95 | ledger.Record(evidence.Receipt{ToolName: "review_report", Success: true, Args: json.RawMessage(`{ |
| 96 | "kind":"security", |
| 97 | "verdict":"pass", |
| 98 | "reviewed_paths":["internal/permission/gate.go"], |
| 99 | "findings":[] |
| 100 | }`)}) |
| 101 | if got := a.deliveryReviewGateFailure(); got != "" { |
| 102 | t.Fatalf("review gate = %q after both reports, want ready", got) |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | func TestDeliveryReviewGateMediumAcceptsHostProvenVerificationAndCoverage(t *testing.T) { |
| 107 | ledger := evidence.NewLedger() |
| 108 | ledger.Record(evidence.ReceiptFromToolCall("edit_file", json.RawMessage(`{"path":"internal/agent/parser.go"}`), true, false)) |
| 109 | ledger.Record(evidence.ReceiptFromToolCall("bash", json.RawMessage(`{"command":"go test ./..."}`), true, true)) |
| 110 | ledger.Record(evidence.ReceiptFromToolCall("read_file", json.RawMessage(`{"path":"internal/agent/parser.go"}`), true, true)) |
| 111 | ledger.Record(evidence.Receipt{ToolName: "complete_step", Success: true, Args: json.RawMessage(`{ |
| 112 | "step":"fix parser", |
| 113 | "evidence":[{"kind":"verification","command":"go test ./..."}] |
| 114 | }`)}) |
| 115 | |
| 116 | reg := tool.NewRegistry() |
| 117 | reg.Add(fakeTool{name: "review", readOnly: true}) |
| 118 | a := &Agent{deliveryProfile: true, evidence: ledger, tools: reg} |
| 119 | if got := a.deliveryReviewGateFailure(); got != "" { |
| 120 | t.Fatalf("medium-risk host proof was rejected: %q", got) |
| 121 | } |
| 122 | |
| 123 | missingVerification := evidence.NewLedger() |
| 124 | missingVerification.Record(evidence.ReceiptFromToolCall("edit_file", json.RawMessage(`{"path":"internal/agent/parser.go"}`), true, false)) |
| 125 | missingVerification.Record(evidence.ReceiptFromToolCall("read_file", json.RawMessage(`{"path":"internal/agent/parser.go"}`), true, true)) |
| 126 | a.evidence = missingVerification |
| 127 | if got := a.deliveryReviewGateFailure(); !strings.Contains(got, "host-proven verification") { |
| 128 | t.Fatalf("medium-risk review without verification = %q, want host-proof guidance", got) |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | func TestDeliveryReviewGateDefersToParentInSubagents(t *testing.T) { |
| 133 | ledger := evidence.NewLedger() |
| 134 | ledger.Record(evidence.ReceiptFromToolCall("edit_file", json.RawMessage(`{"path":"internal/permission/gate.go"}`), true, false)) |
| 135 | |
| 136 | reg := tool.NewRegistry() |
| 137 | reg.Add(fakeTool{name: "review", readOnly: true}) |
| 138 | reg.Add(fakeTool{name: "security_review", readOnly: true}) |
| 139 | a := &Agent{deliveryProfile: true, evidence: ledger, tools: reg, subagentDepth: 1} |
| 140 | |
| 141 | // Inside a sub-agent the structured-review contract belongs to the parent, |
| 142 | // which receives the child's mutation receipts via mergeChildEvidence. The |
| 143 | // child must not wedge against a review_report demand it may be unable to |
| 144 | // satisfy. |
| 145 | if got := a.deliveryReviewGateFailure(); got != "" { |
| 146 | t.Fatalf("subagent review gate = %q, want deferred to parent", got) |
| 147 | } |
| 148 | } |
| 149 |