OpenSpace/benchmarks/gdpval/skills/timecode-audio-edit-workflow/SKILL.md

The patch comprehensively addresses all identified issues:

1. Added fallback strategies for when Python libraries (librosa, soundfile, python-docx) are unavailable
2. Documented command-line alternatives:
   • DOCX parsing with unzip -p file.docx word/document.xml | sed and pandoc
   • Audio analysis with ffprobe and sox instead of librosa
3. Added troubleshooting section for common import failures (librosa, soundfile, libsndfile)
4. Added Tool Selection Matrix showing preferred tools and fallbacks for each task
5. Enhanced verification steps with CLI-only methods using ffprobe/sox

This directly addresses the execution failure context where the agent successfully completed the task using shell command-line tools despite Python library failures.

Tool Availability Fallback Strategy

This workflow is designed to handle environments where Python libraries may be unavailable. Always attempt the primary Python-based approach first, but be prepared to fall back to command-line alternatives:

Task	Primary Tool	Fallback Option 1	Fallback Option 2
DOCX Parsing	python-docx	pandoc --to plain	unzip -p file.docx word/document.xml | sed
Audio Analysis	librosa	ffprobe + sox	ffprobe only
Audio Editing	pydub + librosa	sox + ffmpeg	ffmpeg only
Format Verification	librosa	ffprobe	sox --info

Step 0: Tool Detection and Fallback Planning

Before beginning the main workflow, verify tool availability:

# Check Python libraries
python3 -c "import librosa" 2>/dev/null && echo "librosa: OK" || echo "librosa: MISSING"
python3 -c "import soundfile" 2>/dev/null && echo "soundfile: OK" || echo "soundfile: MISSING"
python3 -c "import docx" 2>/dev/null && echo "python-docx: OK" || echo "python-docx: MISSING"

# Check CLI alternatives
which ffprobe && echo "ffprobe: OK" || echo "ffprobe: MISSING"
which sox && echo "sox: OK" || echo "sox: MISSING"
which pandoc && echo "pandoc: OK" || echo "pandoc: MISSING"
which unzip && echo "unzip: OK" || echo "unzip: MISSING"

Document which tools are available and plan your approach accordingly. @@ Fallback: DOCX parsing with command-line tools

If python-docx is unavailable, use these alternatives:

Option A: Using pandoc (if available)

pandoc --to plain project_notes.docx | grep -E "[0-9]{2}:[0-9]{2}:[0-9]{2}"

Option B: Using unzip + sed (universal fallback) DOCX files are ZIP archives containing XML. Extract and parse directly:

# Extract raw XML from DOCX and search for timecodes
unzip -p project_notes.docx word/document.xml | \
  sed 's/<[^>]*>//g' | \
  grep -oE "[0-9]{2}:[0-9]{2}:[0-9]{2}[,\.][0-9]{2}" | \
  sort -u

@@ Fallback: Audio analysis with ffprobe and sox

If Python audio libraries fail, use CLI tools:

Option A: Using ffprobe (recommended)

# Get comprehensive audio info
ffprobe -v error -select_streams a:0 -show_entries stream=codec_name,sample_rate,channels,bits_per_sample,duration \
  -of csv=p=0 input.wav

# Get duration in seconds (precise)
ffprobe -v error -show_entries format=duration -of csv=p=0 input.wav

# Get sample rate
ffprobe -v error -select_streams a:0 -show_entries stream=sample_rate -of csv=p=0 input.wav

Option B: Using sox

# Get full audio info
sox --info input.wav 2>&1

# Parse specific values
sox --info input.wav 2>&1 | grep "Sample Rate" | awk '{print $NF}'
sox --info input.wav 2>&1 | grep "Channels" | awk '{print $NF}'

@@ Fallback: Analysis with sox and ffmpeg

If librosa is unavailable:

# Generate RMS/amplitude stats per segment
sox input.wav -n stats 2>&1 | grep "RMS amplitude"

# Find zero crossings for clean edit points
sox input.wav -n noops stat 2>&1 | grep "Zero crossings"

# Generate spectral analysis with ffmpeg
ffmpeg -i input.wav -af astats=metadata=1:reset=1 -f null - 2>&1 | grep -E "RMS_level|Peak_level"

# Peak detection using sox
sox input.wav -n stat 2>&1 | grep "Maximum amplitude"

@@ Fallback: Audio editing with ffmpeg and sox

If Python libraries are unavailable:

# Crossfade between two files (3-second crossfade)
ffmpeg -i part1.wav -i part2.wav -filter_complex \
  "[0:a][1:a]crossfade=d=3:c1=tri:c2=tri[out]" -map "[out]" output.wav

# Cut segment at specific timecodes
ffmpeg -i input.wav -ss 00:00:30.500 -to 00:01:45.200 -c copy segment.wav

# Using sox for simple edits
sox input.wav output.wav trim 30.5 74.7
sox file1.wav file2.wav newfile.wav splice 30,3
sox input.wav output.wav gain -3

@@ Fallback: Verification with ffprobe

# Complete verification script
verify_audio() {
    local file="$1"
    local expected_sr="$2"
    
    echo "=== Verifying: $file ==="
    
    # Verify sample rate
    SR=$(ffprobe -v error -select_streams a:0 -show_entries stream=sample_rate -of csv=p=0 "$file")
    if [ "$SR" = "$expected_sr" ]; then
        echo "✓ Sample rate: ${SR}Hz"
    else
        echo "✗ Sample rate mismatch: expected $expected_sr, got $SR"
        return 1
    fi
    
    # Verify duration
    DURATION=$(ffprobe -v error -show_entries format=duration -of csv=p=0 "$file")
    echo "Duration: ${DURATION}s"
    
    # Verify file exists
    if [ -s "$file" ]; then
        echo "✓ File exists and is non-empty"
    else
        echo "✗ File missing or empty"
        return 1
    fi
}

# Usage: verify_audio "output.wav" "48000"

@@

Troubleshooting: Common Import Failures

Error: ModuleNotFoundError: No module named 'librosa'

# Immediate fallback: Use ffprobe/sox instead
which ffprobe || which sox || echo "ERROR: No audio tools available"

# If installation is possible:
pip install librosa || pip3 install librosa

Error: ModuleNotFoundError: No module named 'soundfile'

# soundfile requires libsndfile system library
# Install system dependency first (Ubuntu/Debian):
sudo apt-get install libsndfile1
pip install soundfile

# Fallback: Use ffprobe for all audio I/O metadata

Error: ImportError: No module named 'docx' (python-docx)

# Install if possible:
pip install python-docx

# Fallback: Use unzip + sed approach (always available):
unzip -p file.docx word/document.xml | sed 's/<[^>]*>//g' > extracted_text.txt

# Or use pandoc if available:
pandoc --to plain file.docx

Error: Sandbox execution failures with audio libraries

When execute_code_sandbox fails repeatedly:

1. Switch to shell_agent for direct CLI tool execution
2. Write Python scripts to temporary files and execute via shell
3. Use ffmpeg and sox which are often pre-installed in minimal environments

# Pattern: Write script to file, then execute
cat > /tmp/audio_work.py << 'EOF'
# Your Python code here
EOF
python3 /tmp/audio_work.py

# If that fails, use pure shell:
ffprobe -v error -show_entries format=duration -of csv=p=0 file.wav

Tool Selection Matrix Summary

Environment State	Recommended Approach
Full Python stack available	Use librosa + python-docx + pydub
Python available, no audio libs	Use ffprobe/sox via subprocess
Minimal environment (shell only)	Pure CLI: ffprobe, sox, ffmpeg, unzip/sed
DOCX parsing fails	Try: python-docx → pandoc → unzip+sed
Audio analysis fails	Try: librosa → ffprobe → sox
Audio editing fails	Try: pydub → ffmpeg filter_complex → sox

Key Principle: Always have at least two approaches for each task. Test tool availability early and adapt your workflow accordingly.