feat(wing): LLM classifier with tolerant, fail-closed reply parsing (#205 S1)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-20 05:57:51 +02:00
parent 5bbf47064d
commit 85d8ecb5c7
2 changed files with 166 additions and 0 deletions
@@ -0,0 +1,67 @@
+package wing_test
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/wing"
+)
+
+func completes(reply string, err error) wing.CompleteFunc {
+	return func(_ context.Context, _, _ string) (string, error) { return reply, err }
+}
+
+func TestLLMClassifier_ParsesTheReply(t *testing.T) {
+	tests := []struct {
+		name    string
+		reply   string
+		want    wing.Suggestion
+		wantErr bool
+	}{
+		{"bare json", `{"wing":"tapir","confidence":0.9,"reason":"video summariser"}`, wing.Suggestion{Wing: "tapir", Confidence: 0.9, Reason: "video summariser"}, false},
+		{"json inside prose and fences", "Sure!\n```json\n{\"wing\":\"homelab\",\"confidence\":0.7,\"reason\":\"k3s\"}\n```\nDone.", wing.Suggestion{Wing: "homelab", Confidence: 0.7, Reason: "k3s"}, false},
+		{"none means abstain", `{"wing":"none","confidence":0.2,"reason":"generic"}`, wing.Suggestion{Wing: "", Confidence: 0.2, Reason: "generic"}, false},
+		{"confidence above 1 is clamped", `{"wing":"tapir","confidence":7,"reason":"x"}`, wing.Suggestion{Wing: "tapir", Confidence: 1, Reason: "x"}, false},
+		{"negative confidence is clamped", `{"wing":"tapir","confidence":-1,"reason":"x"}`, wing.Suggestion{Wing: "tapir", Confidence: 0, Reason: "x"}, false},
+		{"empty reply (a reasoning model that ran out of tokens)", ``, wing.Suggestion{}, true},
+		{"no json at all", `I think it is tapir`, wing.Suggestion{}, true},
+		{"malformed json", `{"wing":"tapir","confidence":`, wing.Suggestion{}, true},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			c := wing.NewLLMClassifier(completes(tt.reply, nil))
+			got, err := c.Classify(context.Background(), "note", []string{"tapir", "homelab"})
+			if tt.wantErr {
+				require.Error(t, err)
+				return
+			}
+			require.NoError(t, err)
+			assert.Equal(t, tt.want, got)
+		})
+	}
+}
+
+func TestLLMClassifier_PropagatesTheModelError(t *testing.T) {
+	boom := errors.New("model down")
+	_, err := wing.NewLLMClassifier(completes("", boom)).Classify(context.Background(), "n", []string{"tapir"})
+	assert.ErrorIs(t, err, boom)
+}
+
+func TestLLMClassifier_PromptListsCandidatesAndBoundsTheNote(t *testing.T) {
+	var gotUser string
+	c := wing.NewLLMClassifier(func(_ context.Context, _, user string) (string, error) {
+		gotUser = user
+		return `{"wing":"tapir","confidence":0.9,"reason":"x"}`, nil
+	})
+	long := strings.Repeat("word ", 5000)
+	_, err := c.Classify(context.Background(), long, []string{"tapir", "homelab"})
+	require.NoError(t, err)
+	assert.Contains(t, gotUser, "tapir")
+	assert.Contains(t, gotUser, "homelab")
+	assert.Less(t, len(gotUser), 12000, "a whole transcript must not be sent as the classification input")
+}