feat(captcha): audio captcha transcription solver

internal/captcha/audio.go

@@ -0,0 +1,25 @@
+package captcha
+
+import (
+	"context"
+	"strings"
+
+	"github.com/lukaszraczylo/captcha-solver-mcp/internal/llm"
+)
+
+// AudioResult is the transcription answer.
+type AudioResult struct {
+	Text       string  `json:"text"`
+	Confidence float64 `json:"confidence"`
+}
+
+// SolveAudio transcribes an audio captcha. Confidence is fixed at 1.0 for
+// transcription providers (which do not return token-level confidence here);
+// the raw transcript is trimmed.
+func SolveAudio(ctx context.Context, c llm.Client, audio []byte, filename, language string) (AudioResult, error) {
+	txt, err := c.Transcribe(ctx, audio, filename, llm.Options{Language: language})
+	if err != nil {
+		return AudioResult{}, err
+	}
+	return AudioResult{Text: strings.TrimSpace(txt), Confidence: 1.0}, nil
+}

internal/captcha/audio_test.go

@@ -0,0 +1,29 @@
+package captcha
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/lukaszraczylo/captcha-solver-mcp/internal/llm"
+)
+
+func TestSolveAudio(t *testing.T) {
+	f := &llm.Fake{TranscribeResp: "  forty two  "}
+	res, err := SolveAudio(context.Background(), f, []byte("RIFF"), "a.wav", "en")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Text != "forty two" {
+		t.Fatalf("got %q", res.Text)
+	}
+}
+
+func TestSolveAudioUnsupported(t *testing.T) {
+	f := &llm.Fake{TranscribeErr: llm.ErrUnsupported{Provider: "anthropic"}}
+	_, err := SolveAudio(context.Background(), f, []byte{1}, "a.wav", "")
+	var u llm.ErrUnsupported
+	if !errors.As(err, &u) {
+		t.Fatalf("want ErrUnsupported, got %v", err)
+	}
+}