byre/internal/agents/inbox_internal_test.go

package agents

import (
	"context"
	"encoding/json"
	"fmt"
	"path/filepath"
	"strings"
	"testing"
	"time"

	"g.fc.run/misc/byre/internal/api"
	"g.fc.run/misc/byre/internal/store"
)

func TestValidateAsk(t *testing.T) {
	m := &Manager{}
	l := &live{id: "s1"}
	yes := true

	ask := func(context string) askArgs {
		return askArgs{Context: context, Question: "q", WhyBlocked: "w", Blocking: &yes}
	}
	withAttachment := ask("- (*) a\n- ( ) b")
	withAttachment.Attachments = []api.Attachment{{Kind: "video"}}

	cases := []struct {
		name string
		args askArgs
		want []string
	}{
		{"pick one", ask("Background.\n\n- (*) a: does a\n- ( ) b: does b"), nil},
		{"pick any", ask("- [ ] a\n- [x] b\n- [X] c"), nil},
		{"pick any, none recommended", ask("- [ ] a\n- [ ] b"), nil},
		{"empty", askArgs{}, []string{"context is empty", "question is empty", "why_blocked is empty", "at least two choices", "blocking is required"}},
		{"plain list", ask("- a\n- b"), []string{"at least two choices"}},
		{"one choice", ask("- (*) a"), []string{"at least two choices"}},
		{"in a code fence", ask("```\n- (*) a\n- ( ) b\n```"), []string{"at least two choices"}},
		{"no recommendation", ask("- ( ) a\n- ( ) b"), []string{"exactly one"}},
		{"two recommendations", ask("- (*) a\n- (*) b"), []string{"exactly one"}},
		{"mixed", ask("- (*) a\n- [ ] b"), []string{"mix"}},
		{"duplicate", ask("- (*) a\n- ( ) a"), []string{`"a" appears twice`}},
		{"bad attachment", withAttachment, []string{`attachments[0].kind "video"`}},
	}

	for _, c := range cases {
		t.Run(c.name, func(t *testing.T) {
			err := m.validateAsk(context.Background(), l, c.args)
			if len(c.want) == 0 {
				if err != nil {
					t.Fatalf("unexpected error %v", err)
				}
				return
			}
			if err == nil {
				t.Fatalf("want rejection mentioning %v", c.want)
			}
			if !strings.Contains(err.Error(), "nothing was filed") {
				t.Fatalf("rejection doesn't say nothing was filed: %v", err)
			}
			for _, w := range c.want {
				if !strings.Contains(err.Error(), w) {
					t.Fatalf("rejection %q doesn't mention %q", err, w)
				}
			}
		})
	}
}

func TestCancelQuestion(t *testing.T) {
	ctx := context.Background()
	st, err := store.Open(filepath.Join(t.TempDir(), "byre.db"), func(api.Event) {})
	if err != nil {
		t.Fatalf("open store: %v", err)
	}
	t.Cleanup(func() { st.Close() })

	now := time.Now().UTC()
	_, err = st.EnsureUser(ctx, "alice", "Alice")
	if err != nil {
		t.Fatalf("ensure user: %v", err)
	}
	err = st.CreateProject(ctx, api.Project{Id: "p1", Name: "demo", RepoPath: "/nonexistent", DefaultBranch: "main", Remote: "none", CreatedBy: "alice", CreatedAt: now})
	if err != nil {
		t.Fatalf("create project: %v", err)
	}
	err = st.CreateTask(ctx, api.Task{Id: "t1", ProjectId: "p1", Title: "Fix the thing", Branch: "b", WorktreePath: "/nonexistent", BaseRef: "x", Status: api.TaskStatusActive, CreatedBy: "alice", CreatedAt: now}, "alice")
	if err != nil {
		t.Fatalf("create task: %v", err)
	}
	err = st.CreateSession(ctx, api.AgentSession{Id: "s1", TaskId: "t1", Vendor: api.Claude, VendorSessionId: "s1", State: api.AgentSessionStateIdle, Role: api.Implementer, CreatedBy: "alice", CreatedAt: now}, "p1")
	if err != nil {
		t.Fatalf("create session: %v", err)
	}

	m := New(st, DefaultLimits, t.TempDir(), t.TempDir())
	l := &live{id: "s1", dir: t.TempDir(), vendor: api.Claude, taskID: "t1", projectID: "p1"}
	other := &live{id: "s2", dir: t.TempDir(), vendor: api.Claude, taskID: "t1", projectID: "p1"}

	yes := true
	raw, err := json.Marshal(askArgs{Context: "- (*) red\n- ( ) blue", Question: "Red or blue?", WhyBlocked: "w", Blocking: &yes})
	if err != nil {
		t.Fatalf("marshal: %v", err)
	}
	out, err := m.ask(ctx, l, raw)
	if err != nil {
		t.Fatalf("ask: %v", err)
	}
	items, err := st.ListInbox(ctx, store.InboxFilter{SessionID: "s1", State: api.Open})
	if err != nil || len(items) != 1 {
		t.Fatalf("inbox after ask: %+v %v", items, err)
	}
	id := items[0].Id
	payload := items[0].Payload
	if fmt.Sprint(payload["options"]) != "[red blue]" || payload["multiple"] != false || fmt.Sprint(payload["recommended"]) != "[red]" {
		t.Fatalf("choices not parsed from context: %+v", payload)
	}
	if !strings.Contains(out, "question_id: "+id) || !strings.Contains(out, "cancel_question") {
		t.Fatalf("ask result doesn't give the id and cancel_question: %s", out)
	}

	cancel := func(l *live, id string) error {
		raw, err := json.Marshal(cancelArgs{QuestionID: id})
		if err != nil {
			t.Fatalf("marshal: %v", err)
		}
		_, err = m.cancelQuestion(ctx, l, raw)
		return err
	}

	if cancel(l, "nope") == nil {
		t.Fatalf("cancelled a question that doesn't exist")
	}
	if cancel(other, id) == nil {
		t.Fatalf("cancelled another session's question")
	}
	err = cancel(l, id)
	if err != nil {
		t.Fatalf("cancel: %v", err)
	}
	item, err := st.GetInboxItem(ctx, id)
	if err != nil || item.State != api.Dismissed || item.ResolvedBy != nil || item.ResolvedAt == nil {
		t.Fatalf("after cancel: %+v %v", item, err)
	}
	waiting, err := st.AwaitingHuman(ctx, "s1")
	if err != nil || waiting != 0 {
		t.Fatalf("still awaiting human after cancel: %d %v", waiting, err)
	}
	if cancel(l, id) == nil {
		t.Fatalf("cancelled a question twice")
	}
}

func TestAsksQuestion(t *testing.T) {
	cases := []struct {
		name string
		text string
		want bool
	}{
		{"prose question (seq 61)", "Which of these do you want to start with?", true},
		{"question mid-message", "I can do either.\n\nWhich do you want first? I'd start with the spinner.", true},
		{"bold question", "**Should I deploy now?**", true},
		{"question mark only inside quotes (seq 108)", "- **A:** block if the final message contains a \"?\".\n- **C:** block if the turn's report lists open questions.\n\nMy recommendation is A and C together.", false},
		{"quoted question", "Without it, a question like \"Which of these do you want to start with?\" still reaches you as plain text.", false},
		{"inline code", "The view calls `GET /api/inbox?task=x` on load.", false},
		{"fenced code", "Run:\n```\nif ok? then\n```\nDone.", false},
		{"url query", "See https://example.com/?a=1 for details.", false},
		{"blockquote", "> did it work?\n\nIt did.", false},
		{"no question", "I committed everything and the tests pass.", false},
	}
	for _, c := range cases {
		t.Run(c.name, func(t *testing.T) {
			if got := asksQuestion(c.text); got != c.want {
				t.Fatalf("asksQuestion(%q) = %v, want %v", c.text, got, c.want)
			}
		})
	}
}

func TestBounceProseQuestion(t *testing.T) {
	m := &Manager{}
	stop := func(l *live, text string) map[string]any {
		t.Helper()
		input, err := json.Marshal(map[string]any{"hook_event_name": "Stop", "last_assistant_message": text})
		if err != nil {
			t.Fatal(err)
		}
		out, err := m.bounceProseQuestion(l)(context.Background(), input)
		if err != nil {
			t.Fatal(err)
		}
		return out
	}

	l := &live{}
	if out := stop(l, "Which of these do you want to start with?"); out["decision"] != "block" || out["reason"] != proseQuestionNudge {
		t.Fatalf("prose question not bounced: %v", out)
	}
	if out := stop(l, "Which of these do you want to start with?"); len(out) != 0 {
		t.Fatalf("bounced twice in one turn: %v", out)
	}

	asked := &live{asked: true}
	if out := stop(asked, "Should I go ahead?"); len(out) != 0 {
		t.Fatalf("bounced a turn that called ask_human: %v", out)
	}

	if out := stop(&live{}, "I committed everything and the tests pass."); len(out) != 0 {
		t.Fatalf("bounced a message with no question: %v", out)
	}
}

func TestDeliveryMessage(t *testing.T) {
	str := func(s string) *string { return &s }
	opts := func(s ...string) *[]string { return &s }
	question := api.InboxItem{Id: "q1", ThreadId: "q1", Kind: api.Question, Payload: map[string]any{"question": "Which color?", "multiple": false}}
	checklist := api.InboxItem{Id: "q2", ThreadId: "q2", Kind: api.Question, Payload: map[string]any{"question": "Which chores?", "multiple": true}}
	failure := api.InboxItem{Id: "e1", ThreadId: "e1", Kind: api.SessionError, Payload: map[string]any{"error": "process died"}}

	msg := deliveryMessage([]store.Pending{
		{Item: question, Response: api.InboxResponse{User: "alice", Type: api.Option, Options: opts("blue")}},
		{Item: checklist, Response: api.InboxResponse{User: "alice", Type: api.Option, Options: opts("tests", "docs"), Text: str("also run go vet")}},
		{Item: question, Response: api.InboxResponse{User: "bob", Type: api.Other, Text: str("what is it for?"),
			Attachments: []api.Attachment{{Kind: api.AttachmentKindUrl, Url: str("https://example.com")}}}},
		{Item: failure, Response: api.InboxResponse{User: "alice", Type: api.Other, Text: str("stop and write tests")}},
	})

	for _, want := range []string{
		"Your question q1: Which color?",
		"alice chose: blue",
		"Your question q2: Which chores?\nalice checked:\n- tests\n- docs\nalice added: also run go vet\n",
		"bob answered (none of your options): what is it for?",
		`calling ask_human with thread_id "q1"`,
		`"url":"https://example.com"`,
		"Your error e1: process died",
		"alice replied: stop and write tests",
	} {
		if !strings.Contains(msg, want) {
			t.Fatalf("message missing %q:\n%s", want, msg)
		}
	}
}