feat(offscreen): record proposal_made/action_taken/proposal_dismissed

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-06-05 19:48:50 -04:00
parent 5bdb98c1a7
commit 16b44c9e90
2 changed files with 155 additions and 5 deletions
@@ -6,15 +6,19 @@ package offscreen

 import (
 	"context"
+	"crypto/rand"
+	"encoding/hex"
 	"encoding/json"
 	"errors"
 	"fmt"
+	"log"
 	"sync"
 	"time"

 	"keel/internal/ai"
 	"keel/internal/harness"
 	"keel/internal/knowledge"
+	"keel/internal/memory"
 	"keel/internal/mode"
 	"keel/internal/tasks"
 )
@@ -28,6 +32,12 @@ const (
 	statusDone     = "done"
 )

+const (
+	kindProposalMade      = "proposal_made"
+	kindActionTaken       = "action_taken"
+	kindProposalDismissed = "proposal_dismissed"
+)
+
 // Mode is the one-shot off-screen advisor: collect → propose → confirm/dismiss.
 type Mode struct {
 	mu    sync.Mutex
@@ -35,12 +45,15 @@ type Mode struct {
 	ai    *ai.Service
 	tasks tasks.Provider
 	know  knowledge.Source // may be nil; assembleBrief must guard
+	mem   memory.Store     // may be nil; record helpers guard
+	now   func() time.Time // event timestamps; never nil after New

-	status   string
-	gen      int
-	proposal *ai.OffscreenProposal
-	errMsg   string
-	done     bool
+	status     string
+	gen        int
+	proposal   *ai.OffscreenProposal
+	errMsg     string
+	done       bool
+	proposalID string
 }

 var _ mode.Mode = (*Mode)(nil)
@@ -48,10 +61,16 @@ var _ mode.Mode = (*Mode)(nil)
 // New builds an off-screen mode from the shared services. It does not start the
 // brain; call Command(ctx, "start", nil) (or use Factory) to kick off a brief.
 func New(svc harness.Services) *Mode {
+	now := svc.Clock
+	if now == nil {
+		now = time.Now
+	}
 	m := &Mode{
 		ai:     svc.AI,
 		tasks:  svc.Tasks,
 		know:   svc.Knowledge,
+		mem:    svc.Memory,
+		now:    now,
 		status: statusPending,
 	}
 	m.async = harness.NewAsync(&m.mu, svc.Notify)
@@ -78,8 +97,14 @@ func (m *Mode) Command(ctx context.Context, name string, _ json.RawMessage) erro
 		return m.confirm(ctx)
 	case "dismiss":
 		m.mu.Lock()
+		id := m.proposalID
 		m.done = true
 		m.mu.Unlock()
+		m.recordSync(memory.Event{
+			Kind: kindProposalDismissed,
+			At:   m.now(),
+			Data: map[string]any{"proposal_id": id, "mode": "offscreen"},
+		})
 		return nil
 	default:
 		return fmt.Errorf("offscreen: unknown command %q", name)
@@ -112,6 +137,18 @@ func (m *Mode) start() error {
 			}
 			m.status = statusProposed
 			m.proposal = &p
+			id := newProposalID()
+			m.proposalID = id
+			m.recordAsync(memory.Event{
+				Kind: kindProposalMade,
+				At:   m.now(),
+				Data: map[string]any{
+					"proposal_id": id,
+					"mode":        "offscreen",
+					"next_action": p.NextAction,
+					"rationale":   p.Rationale,
+				},
+			})
 		})
 	return nil
 }
@@ -129,9 +166,15 @@ func (m *Mode) confirm(ctx context.Context) error {
 		return err
 	}
 	m.mu.Lock()
+	id := m.proposalID
 	m.status = statusDone
 	m.done = true
 	m.mu.Unlock()
+	m.recordSync(memory.Event{
+		Kind: kindActionTaken,
+		At:   m.now(),
+		Data: map[string]any{"proposal_id": id, "mode": "offscreen", "next_action": p.NextAction},
+	})
 	return nil
 }

@@ -156,3 +199,37 @@ func Factory(svc harness.Services) mode.Mode {
 	_ = m.start()
 	return m
 }
+
+// recordAsync persists an event off the mode's lock, best-effort. Used on the
+// propose path, which runs under the mode mutex (async apply).
+func (m *Mode) recordAsync(ev memory.Event) {
+	if m.mem == nil {
+		return
+	}
+	go func() {
+		if err := m.mem.Record(context.Background(), ev); err != nil {
+			log.Printf("offscreen: memory record %s: %v", ev.Kind, err)
+		}
+	}()
+}
+
+// recordSync persists an event inline, best-effort. Used on confirm/dismiss,
+// which already do synchronous work.
+func (m *Mode) recordSync(ev memory.Event) {
+	if m.mem == nil {
+		return
+	}
+	if err := m.mem.Record(context.Background(), ev); err != nil {
+		log.Printf("offscreen: memory record %s: %v", ev.Kind, err)
+	}
+}
+
+// newProposalID returns a short random hex id correlating a proposal with its
+// outcome event.
+func newProposalID() string {
+	var b [8]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		return "proposal"
+	}
+	return hex.EncodeToString(b[:])
+}
@@ -8,6 +8,7 @@ import (

 	"keel/internal/ai"
 	"keel/internal/harness"
+	"keel/internal/memory"
 	"keel/internal/tasks"
 )

@@ -138,3 +139,75 @@ func TestUnknownCommandErrors(t *testing.T) {
 		t.Fatal("unknown command should error")
 	}
 }
+
+// newTestModeMem is newTestMode with an injected in-memory Store.
+func newTestModeMem(t *testing.T, ft *fakeTasks, backend ai.Backend, mem *memory.Fake) *Mode {
+	t.Helper()
+	svc := harness.Services{
+		AI:     ai.NewService(backend),
+		Tasks:  ft,
+		Memory: mem,
+		Clock:  func() time.Time { return time.Unix(1000, 0) },
+		Notify: func() {},
+	}
+	return New(svc)
+}
+
+// waitEvent polls the fake until an event of kind appears, or fails after 2s.
+func waitEvent(t *testing.T, f *memory.Fake, kind string) memory.Event {
+	t.Helper()
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		for _, e := range f.Events() {
+			if e.Kind == kind {
+				return e
+			}
+		}
+		time.Sleep(5 * time.Millisecond)
+	}
+	t.Fatalf("event %q never recorded (events: %+v)", kind, f.Events())
+	return memory.Event{}
+}
+
+func TestProposeRecordsProposalMade(t *testing.T) {
+	mem := memory.NewFake()
+	m := newTestModeMem(t, &fakeTasks{}, okBackend(), mem)
+	if err := m.Command(context.Background(), "start", nil); err != nil {
+		t.Fatalf("start: %v", err)
+	}
+	waitStatus(t, m, statusProposed)
+
+	ev := waitEvent(t, mem, "proposal_made")
+	if ev.Data["next_action"] != "call mum" || ev.Data["proposal_id"] == "" {
+		t.Fatalf("proposal_made data = %+v", ev.Data)
+	}
+}
+
+func TestConfirmRecordsActionTakenWithSameID(t *testing.T) {
+	mem := memory.NewFake()
+	m := newTestModeMem(t, &fakeTasks{}, okBackend(), mem)
+	_ = m.Command(context.Background(), "start", nil)
+	waitStatus(t, m, statusProposed)
+	made := waitEvent(t, mem, "proposal_made")
+
+	if err := m.Command(context.Background(), "confirm", nil); err != nil {
+		t.Fatalf("confirm: %v", err)
+	}
+	taken := waitEvent(t, mem, "action_taken")
+	if taken.Data["proposal_id"] != made.Data["proposal_id"] {
+		t.Fatalf("action_taken id %v != proposal_made id %v", taken.Data["proposal_id"], made.Data["proposal_id"])
+	}
+}
+
+func TestDismissRecordsProposalDismissed(t *testing.T) {
+	mem := memory.NewFake()
+	m := newTestModeMem(t, &fakeTasks{}, okBackend(), mem)
+	_ = m.Command(context.Background(), "start", nil)
+	waitStatus(t, m, statusProposed)
+	_ = waitEvent(t, mem, "proposal_made")
+
+	if err := m.Command(context.Background(), "dismiss", nil); err != nil {
+		t.Fatalf("dismiss: %v", err)
+	}
+	_ = waitEvent(t, mem, "proposal_dismissed")
+}