clawdbot-workspace/scripts/captcha_agent_browser.sh

#!/bin/bash
# Audio Captcha Solver - Agent Browser Edition
# Uses agent-browser for network interception instead of BlackHole
#
# Usage:
#   ./captcha_agent_browser.sh <url> [mode] [target]
#   ./captcha_agent_browser.sh "https://site.com/login" transcribe
#   ./captcha_agent_browser.sh "https://site.com/login" identify "stream"
#   ./captcha_agent_browser.sh "https://site.com/login" describe

set -e

URL="${1:-}"
MODE="${2:-transcribe}"
TARGET="${3:-}"

if [ -z "$URL" ]; then
    echo "Usage: $0 <url> [mode] [target]"
    echo ""
    echo "Modes:"
    echo "  transcribe  - Speech-to-text (default)"
    echo "  identify    - Which sound is X? (requires target)"
    echo "  describe    - List all sounds heard"
    echo ""
    echo "Example:"
    echo "  $0 'https://example.com/login' identify 'stream'"
    exit 1
fi

OUTPUT_DIR="/tmp/captcha-audio"
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
AUDIO_FILE="$OUTPUT_DIR/captcha_$TIMESTAMP.mp3"

mkdir -p "$OUTPUT_DIR"

echo "🌐 Audio Captcha Solver (Agent Browser)"
echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
echo ""

# Open URL in agent-browser
echo "📍 Opening: $URL"
agent-browser open "$URL" --headed

echo ""
echo "👆 Find and click the AUDIO CAPTCHA button on the page"
echo "   (I'm watching network requests for audio files...)"
echo ""

# Wait a moment for page load
sleep 2

# Poll for audio requests
echo "👂 Listening for audio file requests..."
MAX_ATTEMPTS=60
ATTEMPT=0
AUDIO_URL=""

while [ $ATTEMPT -lt $MAX_ATTEMPTS ]; do
    # Get network requests and look for audio
    REQUESTS=$(agent-browser network requests --json 2>/dev/null || echo "[]")
    
    # Look for audio URLs in the requests
    AUDIO_URL=$(echo "$REQUESTS" | grep -oE 'https?://[^"]+\.(mp3|wav|ogg|m4a|webm)[^"]*' | head -1 || true)
    
    # Also check for audio content types or captcha audio patterns
    if [ -z "$AUDIO_URL" ]; then
        AUDIO_URL=$(echo "$REQUESTS" | grep -oE 'https?://[^"]*audio[^"]*' | head -1 || true)
    fi
    if [ -z "$AUDIO_URL" ]; then
        AUDIO_URL=$(echo "$REQUESTS" | grep -oE 'https?://[^"]*captcha[^"]*\.(mp3|wav|ogg)[^"]*' | head -1 || true)
    fi
    if [ -z "$AUDIO_URL" ]; then
        AUDIO_URL=$(echo "$REQUESTS" | grep -oE 'https?://[^"]*recaptcha[^"]*audio[^"]*' | head -1 || true)
    fi
    
    if [ -n "$AUDIO_URL" ]; then
        echo "🎵 Found audio URL!"
        break
    fi
    
    sleep 1
    ATTEMPT=$((ATTEMPT + 1))
    
    # Show progress every 10 seconds
    if [ $((ATTEMPT % 10)) -eq 0 ]; then
        echo "   Still listening... ($ATTEMPT seconds)"
    fi
done

if [ -z "$AUDIO_URL" ]; then
    echo "❌ No audio file detected after ${MAX_ATTEMPTS} seconds"
    echo ""
    echo "Debugging: Here are recent network requests:"
    agent-browser network requests 2>/dev/null | head -20
    exit 1
fi

echo "📥 Downloading: $AUDIO_URL"
curl -sL "$AUDIO_URL" -o "$AUDIO_FILE"

if [ ! -s "$AUDIO_FILE" ]; then
    echo "❌ Failed to download audio file"
    exit 1
fi

echo "✅ Saved to: $AUDIO_FILE"
echo ""

# Analyze based on mode
case "$MODE" in
    transcribe)
        echo "🧠 Transcribing with Whisper..."
        whisper "$AUDIO_FILE" \
            --model small \
            --language en \
            --output_format txt \
            --output_dir "$OUTPUT_DIR" \
            2>/dev/null
        
        TXT_FILE="${AUDIO_FILE%.mp3}.txt"
        if [ -f "$TXT_FILE" ]; then
            RAW_TEXT=$(cat "$TXT_FILE")
            EXTRACTED=$(echo "$RAW_TEXT" | grep -oE '[A-Za-z0-9]' | tr -d '\n' | tr '[:lower:]' '[:upper:]')
            
            echo ""
            echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
            echo "📝 Raw text: $RAW_TEXT"
            echo "🔤 Extracted: $EXTRACTED"
            echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
            
            echo "$EXTRACTED" | pbcopy
            echo "📋 Copied to clipboard!"
        else
            echo "❌ Transcription failed"
        fi
        ;;
        
    identify)
        if [ -z "$TARGET" ]; then
            echo "❌ identify mode requires a target sound"
            echo "   Example: $0 '$URL' identify 'stream'"
            exit 1
        fi
        
        echo "🧠 Asking Gemini: which sound is '$TARGET'?"
        
        PROMPT="Listen to this audio captcha. It contains multiple sounds. Which sound is a \"$TARGET\"? Reply with ONLY the number (1, 2, 3, etc.) of the matching sound. Just the number, nothing else."
        
        RESPONSE=$(gemini -p "$PROMPT" -f "$AUDIO_FILE" 2>/dev/null)
        ANSWER=$(echo "$RESPONSE" | grep -oE '[0-9]+' | head -1)
        
        echo ""
        echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
        echo "🎯 Target: $TARGET"
        echo "✅ Answer: ${ANSWER:-$RESPONSE}"
        echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
        
        echo "${ANSWER:-$RESPONSE}" | pbcopy
        echo "📋 Copied to clipboard!"
        ;;
        
    describe)
        echo "🧠 Asking Gemini to describe all sounds..."
        
        PROMPT="Listen to this audio and describe each distinct sound you hear. Format as: 1: [description], 2: [description], etc."
        
        RESPONSE=$(gemini -p "$PROMPT" -f "$AUDIO_FILE" 2>/dev/null)
        
        echo ""
        echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
        echo "🔊 Sounds detected:"
        echo "$RESPONSE"
        echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
        ;;
        
    *)
        echo "❌ Unknown mode: $MODE"
        echo "   Use: transcribe, identify, or describe"
        exit 1
        ;;
esac

echo ""
echo "Done! Browser is still open if you need to enter the answer."