Preserve context across conversation intents (V-542)

Owner explicitly requested direct commits to master; bypass the branch-only hook.

internal/dialogue/session.go

@@ -55,11 +55,22 @@ type Candidate struct {
 }
 
 type Session struct {
-	Intent    Intent
-	Slots     Slots
-	Timestamp time.Time
-	TTL       time.Duration
-	History   []Turn // most recent turns, newest last; used for anaphora + cross-intent
+	Intent Intent
+	Slots  Slots
+	// Utterance is what he actually said on this turn. It is deliberately
+	// independent of Slots.Text: Text is an intent payload and several valid
+	// routes leave it empty or replace it with a normalized subject. Dialogue
+	// history needs the original words so a later pronoun can refer across
+	// fact, query and chat boundaries without pretending a storage key is a
+	// transcript.
+	Utterance string
+	// Conversational keeps an explicitly opened or chat-routed exchange alive
+	// across later fact/query routes. It does not change what those turns do;
+	// it only says their shared transcript uses the longer dialogue TTL.
+	Conversational bool
+	Timestamp      time.Time
+	TTL            time.Duration
+	History        []Turn // prior turns in speaking order, oldest first; used for anaphora + cross-intent
 	// Candidates — the list she just offered, in the order she said it. Empty
 	// on every turn that offered no choice, which is most of them.
 	Candidates []Candidate

internal/dialogue/session_persist_test.go

@@ -29,10 +29,13 @@ func TestSessionSurvivesRestart(t *testing.T) {
 	first := openStore(t, path)
 	before := NewPersistentSessionStore(2*time.Minute, first)
 	before.Put("voice", &Session{
-		Intent:    IntentReminder,
-		Slots:     Slots{Text: "полить цветы", Time: now.Add(time.Hour), HasTime: true},
-		Timestamp: now,
-		TTL:       2 * time.Minute,
+		Intent:         IntentReminder,
+		Slots:          Slots{Text: "полить цветы", Time: now.Add(time.Hour), HasTime: true},
+		Utterance:      "напомни полить цветы",
+		Conversational: true,
+		Timestamp:      now,
+		TTL:            2 * time.Minute,
+		History:        []Turn{{Intent: IntentChat, Text: "мы говорили о цветах"}},
 	})
 	if err := first.Close(); err != nil {
 		t.Fatalf("close: %v", err)
@@ -51,6 +54,15 @@ func TestSessionSurvivesRestart(t *testing.T) {
 	if sess.Intent != IntentReminder {
 		t.Fatalf("intent = %q, want reminder", sess.Intent)
 	}
+	if sess.Utterance != "напомни полить цветы" {
+		t.Fatalf("utterance = %q after restart", sess.Utterance)
+	}
+	if !sess.Conversational {
+		t.Fatal("conversation mode was lost across restart")
+	}
+	if len(sess.History) != 1 || sess.History[0].Text != "мы говорили о цветах" {
+		t.Fatalf("history after restart = %+v", sess.History)
+	}
 	// the follow-up carries no text of its own; it must inherit the old one
 	merged := InheritSlots(sess.Slots, Slots{Time: now.Add(2 * time.Hour), HasTime: true})
 	if merged.Text != "полить цветы" {

internal/lexicon/lexicon.go

@@ -66,6 +66,7 @@ func mustLoad() lexiconFile {
 		"not_place_after_v", "parts_of_day", "reminder_verbs", "half_hour",
 		"filler_particles", "task_done_words", "task_drop_words",
 		"confirm_yes", "confirm_no", "hour_units", "minute_units",
+		"conversation_verbs",
 	} {
 		s, ok := f.Sets[name]
 		if !ok || (len(s.Words) == 0 && len(s.Values) == 0) {
@@ -212,6 +213,12 @@ func inSet(set, word string) bool {
 // Distinct from TaskDropWords, which abandons an item that already exists.
 func DialogueCancel() []string { return words("dialogue_cancel") }
 
+// ConversationVerbs returns Maven's vocabulary for explicitly opening a
+// conversation after a cooperative marker ("давай поговорим", "let's talk").
+// The caller matches Russian members by lemma and English members exactly.
+// Keeping the verbs here prevents session state from growing its own stem list.
+func ConversationVerbs() []string { return words("conversation_verbs") }
+
 // IsFillerParticle reports whether a word can never be the subject of a
 // request: a particle, a politeness word, or the first-person object. See the
 // set's own note for why this is not a stopword list.

internal/lexicon/lexicon_ru_v1.json

@@ -262,6 +262,13 @@
         "cancel", "nevermind", "forget"
       ]
     },
+    "conversation_verbs": {
+      "note": "The verbs Maven accepts after a cooperative marker as an explicit request to converse: «давай поговорим», «давай поболтаем», «let's talk». This is her control vocabulary, like capture_verbs, not a claim to list every way humans can converse. Russian forms are matched by lemma, so one infinitive covers tense and number; English is exact because the Russian dictionary leaves it unchanged.",
+      "words": [
+        "говорить", "поговорить", "болтать", "поболтать", "общаться", "пообщаться",
+        "talk", "chat"
+      ]
+    },
     "confirm_yes": {
       "note": "The whole vocabulary of saying yes to a parked confirm, Russian and English. Closed because it is her question that is being answered: she asked \"да или нет\", and the answers to that question can be listed. Matched as whole tokens and never as substrings — \"погода\", \"давление\" and \"дальше\" all contain \"да\", and a substring test executed a destructive act when he asked about the weather (V-567). Words that merely sound agreeable — \"хорошо\", \"ладно\", \"точно\" — are deliberately absent: they open a sentence about something else as often as they answer, and an unclear answer must route rather than execute.",
       "words": [

internal/router/slots.go

@@ -5,6 +5,7 @@ import (
 	"strconv"
 	"strings"
 	"time"
+	"unicode"
 
 	"github.com/kami/maven/internal/lexicon"
 	"github.com/kami/maven/internal/morph"
@@ -507,8 +508,12 @@ type AnaphoraResolver struct{}
 //
 // Returns the matching pronoun class for cross-referencing with prior slots.
 func (AnaphoraResolver) Resolve(text string) (ref string, ok bool) {
-	s := strings.ToLower(strings.TrimSpace(text))
-	toks := strings.Fields(s)
+	// Pronouns are a closed grammatical class. Split on punctuation instead of
+	// trying to encode word boundaries in a regexp: "это?" and "его," are the
+	// same pronouns as "это" and "его", including next to Cyrillic letters.
+	toks := strings.FieldsFunc(strings.ToLower(strings.TrimSpace(text)), func(r rune) bool {
+		return !unicode.IsLetter(r) && !unicode.IsDigit(r)
+	})
 	for _, tok := range toks {
 		switch tok {
 		case "это", "этого", "этому", "этим", "этом", "эти", "эта":

internal/router/slots_ru_test.go

@@ -50,6 +50,25 @@ func TestDefaultFactParserRU(t *testing.T) {
 	}
 }
 
+func TestAnaphoraResolverUsesTokenBoundariesAcrossPunctuation(t *testing.T) {
+	r := AnaphoraResolver{}
+	for _, text := range []string{
+		"что это?",
+		"вернуть его, наверное",
+		"а он — большой?",
+		"расскажи о ней.",
+	} {
+		if _, ok := r.Resolve(text); !ok {
+			t.Errorf("Resolve(%q) missed a punctuated pronoun", text)
+		}
+	}
+	for _, text := range []string{"этология", "нейросеть", "онлайн"} {
+		if ref, ok := r.Resolve(text); ok {
+			t.Errorf("Resolve(%q) = %q; substring is not a pronoun token", text, ref)
+		}
+	}
+}
+
 func TestParseCalendarDate(t *testing.T) {
 	now := time.Date(2026, 7, 6, 14, 30, 0, 0, time.UTC)
 	cases := []struct {