| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "strings" |
| 5 | "testing" |
| 6 | |
| 7 | "reasonix/internal/evidence" |
| 8 | "reasonix/internal/instruction" |
| 9 | "reasonix/internal/provider" |
| 10 | ) |
| 11 | |
| 12 | func readinessLedger(receipts ...evidence.Receipt) *evidence.Ledger { |
| 13 | l := evidence.NewLedger() |
| 14 | for _, r := range receipts { |
| 15 | l.Record(r) |
| 16 | } |
| 17 | return l |
| 18 | } |
| 19 | |
| 20 | func TestFinalReadinessFailureBranches(t *testing.T) { |
| 21 | check := instruction.VerifyCheck{Command: "go test ./...", SourcePath: "AGENTS.md", Line: 3} |
| 22 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 23 | readOnly := evidence.Receipt{ToolName: "read_file", Success: true, Read: true, Paths: []string{"a.go"}} |
| 24 | checkAfter := evidence.Receipt{ToolName: "bash", Success: true, Command: "go test ./..."} |
| 25 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 26 | completeAfter := evidence.Receipt{ToolName: "complete_step", Success: true, Step: "edit"} |
| 27 | doneTodo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "completed"}}} |
| 28 | |
| 29 | cases := []struct { |
| 30 | name string |
| 31 | checks []instruction.VerifyCheck |
| 32 | evidence *evidence.Ledger |
| 33 | wantEmpty bool |
| 34 | wantContain string |
| 35 | }{ |
| 36 | {"nil evidence never gates", []instruction.VerifyCheck{check}, nil, true, ""}, |
| 37 | {"no writer never gates", []instruction.VerifyCheck{check}, readinessLedger(checkAfter), true, ""}, |
| 38 | {"todo-only turn may end with incomplete list", nil, readinessLedger(todo), true, ""}, |
| 39 | {"read-only context plus todo may end with incomplete list", nil, readinessLedger(readOnly, todo), true, ""}, |
| 40 | {"completed todo without writer satisfies", nil, readinessLedger(doneTodo), true, ""}, |
| 41 | {"writer without checks or todo never gates", nil, readinessLedger(writer), true, ""}, |
| 42 | {"missing project check after writer is reported", []instruction.VerifyCheck{check}, readinessLedger(checkAfter, writer), false, "go test ./..."}, |
| 43 | {"project check run after writer satisfies", []instruction.VerifyCheck{check}, readinessLedger(writer, checkAfter), true, ""}, |
| 44 | {"todo writer without complete_step is reported", nil, readinessLedger(writer, todo), false, "incomplete items"}, |
| 45 | {"complete_step without final todo update is reported", nil, readinessLedger(writer, todo, completeAfter), false, "latest successful todo_write"}, |
| 46 | {"todo writer with complete_step and completed todo satisfies", nil, readinessLedger(writer, todo, completeAfter, doneTodo), true, ""}, |
| 47 | } |
| 48 | for _, tc := range cases { |
| 49 | t.Run(tc.name, func(t *testing.T) { |
| 50 | a := &Agent{evidence: tc.evidence, projectChecks: tc.checks} |
| 51 | got := a.ReadinessResult() |
| 52 | if tc.wantEmpty { |
| 53 | if !got.Ready { |
| 54 | t.Fatalf("ReadinessResult() = %+v, want ready (no gate)", got) |
| 55 | } |
| 56 | return |
| 57 | } |
| 58 | if got.Ready { |
| 59 | t.Fatalf("ReadinessResult() = ready, want a failure mentioning %q", tc.wantContain) |
| 60 | } |
| 61 | if !strings.Contains(got.Reason, tc.wantContain) { |
| 62 | t.Fatalf("ReadinessResult().Reason = %q, want it to mention %q", got.Reason, tc.wantContain) |
| 63 | } |
| 64 | if len(got.Missing) == 0 { |
| 65 | t.Fatalf("ReadinessResult().Missing empty for %q", tc.wantContain) |
| 66 | } |
| 67 | }) |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | func TestFinalReadinessAllowsIncompleteTodosInPlanMode(t *testing.T) { |
| 72 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "draft implementation plan", Status: "pending"}}} |
| 73 | a := &Agent{evidence: readinessLedger(todo)} |
| 74 | a.SetPlanMode(true) |
| 75 | |
| 76 | if got := a.ReadinessResult(); !got.Ready { |
| 77 | t.Fatalf("ReadinessResult() = %+v, want ready in plan mode", got) |
| 78 | } |
| 79 | if got := a.finalReadinessCheckFor(); got.applies { |
| 80 | t.Fatalf("finalReadinessCheckFor() applies in plan mode: %+v", got) |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | func TestFinalReadinessCheckAuditsIncompleteTodos(t *testing.T) { |
| 85 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 86 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 87 | a := &Agent{evidence: readinessLedger(writer, todo)} |
| 88 | |
| 89 | got := a.finalReadinessCheckFor() |
| 90 | if !got.applies { |
| 91 | t.Fatalf("finalReadinessCheckFor() applies = false, want true") |
| 92 | } |
| 93 | if got.incompleteTodos != 1 { |
| 94 | t.Fatalf("incompleteTodos = %d, want 1", got.incompleteTodos) |
| 95 | } |
| 96 | if !strings.Contains(got.reason, "latest successful todo_write") { |
| 97 | t.Fatalf("reason = %q, want incomplete todo message", got.reason) |
| 98 | } |
| 99 | audit := got.audit(evidence.ReadinessBlocked, false) |
| 100 | if audit.IncompleteTodos != 1 { |
| 101 | t.Fatalf("audit.IncompleteTodos = %d, want 1", audit.IncompleteTodos) |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | func TestFinalReadinessAllowsFinalAfterLoopGuardedToolBlocker(t *testing.T) { |
| 106 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 107 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 108 | ledger := readinessLedger(writer, todo) |
| 109 | a := &Agent{evidence: ledger} |
| 110 | a.armLoopGuardPass(ledger.Len()) |
| 111 | |
| 112 | got := a.finalReadinessCheckFor() |
| 113 | if !got.applies { |
| 114 | t.Fatalf("finalReadinessCheckFor() applies = false, want true audit after loop guard") |
| 115 | } |
| 116 | if got.reason != "" { |
| 117 | t.Fatalf("finalReadinessCheckFor() reason = %q, want loop guard to allow final blocker report", got.reason) |
| 118 | } |
| 119 | } |
| 120 | |
| 121 | // TestFinalReadinessLoopGuardPassSurvivesBookkeeping proves the exact actions |
| 122 | // the loop guard recommends — ask, todo_write, complete_step — do not revoke |
| 123 | // the pass: the model must be able to record the blocker and then report it. |
| 124 | func TestFinalReadinessLoopGuardPassSurvivesBookkeeping(t *testing.T) { |
| 125 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 126 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 127 | ledger := readinessLedger(writer, todo) |
| 128 | a := &Agent{evidence: ledger} |
| 129 | a.armLoopGuardPass(ledger.Len()) |
| 130 | |
| 131 | ledger.Record(evidence.Receipt{ToolName: "ask", Success: true}) |
| 132 | ledger.Record(evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}}) |
| 133 | ledger.Record(evidence.Receipt{ToolName: "complete_step", Success: true, Step: "edit"}) |
| 134 | |
| 135 | if got := a.finalReadinessCheckFor(); got.reason != "" { |
| 136 | t.Fatalf("finalReadinessCheckFor() reason = %q, want bookkeeping after the guard to keep the pass", got.reason) |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | // TestFinalReadinessLoopGuardPassRevokedByRealProgress proves a successful |
| 141 | // write or command receipt after the guard revokes the pass: receipts are |
| 142 | // obtainable again, so readiness resumes enforcing them. |
| 143 | func TestFinalReadinessLoopGuardPassRevokedByRealProgress(t *testing.T) { |
| 144 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 145 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 146 | ledger := readinessLedger(writer, todo) |
| 147 | a := &Agent{evidence: ledger} |
| 148 | a.armLoopGuardPass(ledger.Len()) |
| 149 | |
| 150 | ledger.Record(evidence.Receipt{ToolName: "bash", Success: true, Command: "go test ./..."}) |
| 151 | |
| 152 | if got := a.finalReadinessCheckFor(); got.reason == "" { |
| 153 | t.Fatal("finalReadinessCheckFor() reason empty, want real progress after the guard to revoke the pass") |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | // TestFinalReadinessIgnoresLoopGuardQuotedInToolOutput proves the pass is host |
| 158 | // state, not message text: a tool result that merely quotes "[loop guard]" |
| 159 | // (a grep over this repo, a pasted transcript) must not unlock readiness. |
| 160 | func TestFinalReadinessIgnoresLoopGuardQuotedInToolOutput(t *testing.T) { |
| 161 | todo := evidence.Receipt{ToolName: "todo_write", Success: true, Todos: []evidence.TodoItem{{Content: "edit", Status: "in_progress"}}} |
| 162 | writer := evidence.Receipt{ToolName: "write_file", Success: true, Write: true, Paths: []string{"a.go"}} |
| 163 | sess := NewSession("") |
| 164 | sess.Add(provider.Message{Role: provider.RoleUser, Content: "edit"}) |
| 165 | sess.Add(provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: "b1", Name: "bash"}}}) |
| 166 | sess.Add(provider.Message{Role: provider.RoleTool, ToolCallID: "b1", Name: "bash", Content: "agent.go:2082: \"[loop guard] %s has now %s %d times\""}) |
| 167 | a := &Agent{evidence: readinessLedger(writer, todo), session: sess} |
| 168 | |
| 169 | if got := a.finalReadinessCheckFor(); got.reason == "" { |
| 170 | t.Fatal("finalReadinessCheckFor() reason empty, want quoted loop-guard text to be ignored") |
| 171 | } |
| 172 | } |
| 173 |