vlc

VLC — Python Manipulation Guide

This skill teaches setup-gen (configure/launch VLC) and reward-gen (verify playback/config/media) how to work with VLC using Python.

• Libraries: subprocess, requests, xml.etree.ElementTree, librosa, cv2, imagehash, Pillow
• Install: pip3 install requests librosa fastdtw scipy scikit-image opencv-python imagehash Pillow
• Config path (Linux): ~/.config/vlc/vlcrc

---

0. GUI Startup on VM (for setup-gen)

After preparing media/config, setup-gen should open VLC (and target media if required) for the GUI agent.

CRITICAL VM LIMIT: GUI launches must set DISPLAY=:0.

import os
import shlex
import subprocess
import time

def launch_gui(command: str, delay_sec: float = 1.0):
    env = os.environ.copy()
    env["DISPLAY"] = ":0"
    subprocess.Popen(
        shlex.split(command),
        stdout=subprocess.DEVNULL,
        stderr=subprocess.DEVNULL,
        env=env,
    )
    time.sleep(delay_sec)

# Launch VLC with a target file
launch_gui('vlc "/home/user/Desktop/video.mp4"', delay_sec=2.0)

Guidelines:

• Use non-blocking launch (Popen) so script exits cleanly.
• Keep startup deterministic with short sleeps.

---

0.5. Media Asset Library (setup-gen)

Use the real video/audio library at assets/media/vlc/ instead of generating synthetic test patterns.

Available Categories

Audio files can be extracted from videos on the VM using FFmpeg (see Section 1).

Upload Pattern

# Upload video to BOTH VMs
python3 scripts/env_cli.py -c "<workdir>/env_config_initial.json" upload \
    "assets/media/vlc/videos/nature_timelapse_001.mp4" "/home/user/<task_id>.mp4"
python3 scripts/env_cli.py -c "<workdir>/env_config_golden.json"  upload \
    "assets/media/vlc/videos/nature_timelapse_001.mp4" "/home/user/<task_id>.mp4"

# Upload subtitle if needed
python3 scripts/env_cli.py -c "<workdir>/env_config_initial.json" upload \
    "assets/media/vlc/subtitles/english_dialogue.srt" "/home/user/<task_id>.srt"

Then in initial_setup.py:

import shutil
shutil.copy(f'{WORKDIR}/{TASK_ID}.mp4', f'{WORKDIR}/{TASK_ID}_initial_reference.mp4')
launch_gui(f'vlc "{WORKDIR}/{TASK_ID}.mp4"', delay_sec=2.0)

Picking Assets from Manifest

import json, random
manifest = json.load(open("assets/media/manifest.json"))
vlc_videos = [a for a in manifest["assets"] if a["domain"] == "vlc" and a["category"] == "videos"]
# Pick a nature video
nature = [v for v in vlc_videos if "nature" in v.get("tags", [])]
asset = random.choice(nature)

---

1. Creating & Configuring (setup-gen)

vlcrc Configuration File

The vlcrc file uses key=value format. Commented lines start with #. A commented-out key means "use default".

import os, re, shutil, subprocess

VLCRC_PATH = os.path.expanduser("~/.config/vlc/vlcrc")

def read_vlcrc() -> str:
    with open(VLCRC_PATH, "r") as f:
        return f.read()

def set_vlcrc_option(key: str, value: str):
    """Set a vlcrc option. Uncomments the key if commented out."""
    content = read_vlcrc()
    # Pattern: match both commented (#key=...) and uncommented (key=...)
    pattern = re.compile(rf'^(#?\s*){re.escape(key)}=.*$', re.MULTILINE)
    replacement = f'{key}={value}'
    if pattern.search(content):
        content = pattern.sub(replacement, content)
    else:
        content += f'\n{key}={value}\n'
    with open(VLCRC_PATH, "w") as f:
        f.write(content)

# Examples — all config keys the evaluator checks:
set_vlcrc_option("qt-bgcone", "0")                          # splash screen cone: 0=off, 1=on (default: 1)
set_vlcrc_option("qt-max-volume", "200")                     # max volume: integer (default: 125)
set_vlcrc_option("qt-minimal-view", "1")                     # minimal interface: 0=normal, 1=minimal (default: 0)
set_vlcrc_option("qt-slider-colours", "10;10;10;10;10;10;10;10;10;10;10;10")  # R;G;B groups
set_vlcrc_option("global-key-play-pause", "Space")           # global hotkey: empty=""=disabled, key=enabled
set_vlcrc_option("one-instance-when-started-from-file", "0") # single instance: 0=off, 1=on (default: 1)
set_vlcrc_option("play-and-exit", "0")                       # auto-close after playback: 0=off, 1=on (default: 0)
set_vlcrc_option("input-record-path", "/home/user/Desktop")  # recording output path

Launching VLC

# Play a local file
subprocess.Popen(["vlc", "/home/user/Desktop/video.mp4"])

# Play with specific options
subprocess.Popen([
    "vlc",
    "--no-audio",                    # mute audio
    "--no-video-title-show",         # hide title overlay
    "--start-time=10",               # start at 10 seconds
    "--stop-time=30",                # stop at 30 seconds
    "--play-and-pause",              # pause when reaching stop time
    "/home/user/Desktop/video.mp4"
])

# Play with HTTP interface enabled (for status monitoring)
subprocess.Popen([
    "vlc",
    "--extraintf=http",
    "--http-password=password",
    "--http-port=8080",
    "/home/user/Desktop/video.mp4"
])

# Stream an HLS URL
subprocess.Popen(["vlc", "https://example.com/stream/master.m3u8"])

# Suppress verbose output
env = os.environ.copy()
env["VLC_VERBOSE"] = "-1"
subprocess.Popen(["vlc", "/home/user/Desktop/video.mp4"], env=env)

Killing and Restarting VLC

# Kill VLC (needed before modifying vlcrc)
subprocess.run(["pkill", "-f", "vlc"], capture_output=True)
import time; time.sleep(2)

# Restart VLC to pick up config changes
subprocess.Popen(["vlc", "--extraintf=http", "--http-password=password"])

Creating Media Files for Testing

# Generate silent audio with ffmpeg
subprocess.run([
    "ffmpeg", "-f", "lavfi", "-i", "anullsrc=r=44100:cl=mono",
    "-t", "5", "-q:a", "9", "/home/user/Desktop/silence.mp3"
], check=True)

# Generate test video (color bars)
subprocess.run([
    "ffmpeg", "-f", "lavfi", "-i", "testsrc=duration=5:size=640x480:rate=30",
    "-pix_fmt", "yuv420p", "/home/user/Desktop/test.mp4"
], check=True)

# Extract audio from video
subprocess.run([
    "ffmpeg", "-i", "/home/user/Desktop/video.mp4",
    "-vn", "-acodec", "libmp3lame", "/home/user/Desktop/audio.mp3"
], check=True)

# Capture a frame as screenshot
subprocess.run([
    "ffmpeg", "-i", "/home/user/Desktop/video.mp4",
    "-ss", "00:00:05", "-frames:v", "1", "/home/user/Desktop/frame.png"
], check=True)

# Rotate video
subprocess.run([
    "ffmpeg", "-i", "/home/user/Desktop/input.mp4",
    "-vf", "transpose=1", "/home/user/Desktop/rotated.mp4"
], check=True)

---

2. Reading & Verifying (reward-gen)

Verifying vlcrc Config Options

def get_vlcrc_option(key: str, default: str = None) -> str:
    """Read a vlcrc option value. Returns default if key is commented or missing."""
    content = read_vlcrc()
    for line in content.split("\n"):
        stripped = line.strip()
        if stripped.startswith("#") or not stripped:
            continue
        if f"{key}=" in stripped:
            return stripped.split("=", 1)[-1].strip()
    return default

# Config verification helpers with correct defaults
CONFIG_DEFAULTS = {
    "qt-bgcone": "1",
    "qt-max-volume": "125",
    "qt-minimal-view": "0",
    "qt-slider-colours": "153;210;153;20;210;20;255;199;15;245;39;29",
    "one-instance-when-started-from-file": "1",
    "play-and-exit": "0",
}

def verify_vlcrc_option(key: str, expected: str) -> bool:
    actual = get_vlcrc_option(key, CONFIG_DEFAULTS.get(key))
    return str(actual) == str(expected)

# Verify global-key-play-pause (special: empty = disabled, non-empty = enabled)
def verify_global_key_enabled(expected_enabled: bool) -> bool:
    val = get_vlcrc_option("global-key-play-pause", "")
    is_enabled = val != ""
    return is_enabled == expected_enabled

# Verify slider colors are "blackish" (all RGB values < 100)
def verify_slider_blackish() -> bool:
    colours = get_vlcrc_option("qt-slider-colours", CONFIG_DEFAULTS["qt-slider-colours"])
    values = [int(x) for x in colours.split(";")]
    return all(v < 100 for v in values)

# Verify recording path
def verify_recording_path(expected_path: str) -> bool:
    return get_vlcrc_option("input-record-path") == expected_path

Verifying VLC Playback Status (HTTP Interface)

import requests
from xml.etree import ElementTree

def get_vlc_status(host: str = "localhost", port: int = 8080, password: str = "password") -> ElementTree.Element:
    """Fetch VLC status XML via HTTP interface."""
    resp = requests.get(f"http://{host}:{port}/requests/status.xml", auth=("", password), timeout=5)
    return ElementTree.fromstring(resp.content)

def verify_vlc_playing(expected_filename: str = None, expected_url: str = None,
                       host: str = "localhost", port: int = 8080) -> bool:
    tree = get_vlc_status(host, port)
    state = tree.find("state").text
    if state != "playing":
        return False
    if expected_filename:
        # Check multiple XML paths for filename
        for xpath in [
            'information/category[@name="meta"]/info[@name="filename"]',
            'information/category[@name="meta"]/info[@name="title"]',
            'information/category[@name="meta"]/info[@name="uri"]',
            'information/category[@name="meta"]/info[@name="name"]',
        ]:
            elem = tree.find(xpath)
            if elem is not None and elem.text:
                if os.path.basename(elem.text) == expected_filename or elem.text.endswith(expected_filename):
                    return True
        return False
    if expected_url:
        from urllib.parse import urlparse
        for xpath in [
            'information/category[@name="meta"]/info[@name="url"]',
            'information/category[@name="meta"]/info[@name="filename"]',
            'information/category[@name="meta"]/info[@name="title"]',
        ]:
            elem = tree.find(xpath)
            if elem is not None and elem.text:
                if expected_url in elem.text:
                    return True
                # HLS streams may show just the filename
                if elem.text == os.path.basename(urlparse(expected_url).path):
                    return True
        return False
    return True  # just verify playing state

Verifying Fullscreen Mode

def verify_fullscreen(window_size: dict, screen_size: dict) -> bool:
    """Compare window size to screen size."""
    if not window_size or not screen_size:
        return False
    return (window_size["width"] == screen_size["width"] and
            window_size["height"] == screen_size["height"])

Comparing Audio Files (MFCC + DTW)

import librosa
import numpy as np
from fastdtw import fastdtw
from scipy.spatial.distance import cosine

def compare_audios(audio1: str, audio2: str) -> float:
    """Compare two audio files using MFCC features + DTW. Returns 0.0-1.0."""
    try:
        y1, sr1 = librosa.load(audio1)
        y2, sr2 = librosa.load(audio2)
    except Exception:
        return 0.0
    if y1.shape[0] == 0 or y2.shape[0] == 0:
        return 0.0
    mfcc1 = librosa.util.normalize(librosa.feature.mfcc(y=y1, sr=sr1), axis=1)
    mfcc2 = librosa.util.normalize(librosa.feature.mfcc(y=y2, sr=sr2), axis=1)
    distance, path = fastdtw(mfcc1.T, mfcc2.T, dist=lambda x, y: cosine(x, y))
    normalized = distance / len(path) if path else float("inf")
    return float(np.exp(-normalized))

Comparing Video Files (Perceptual Hashing)

import cv2, imagehash
from PIL import Image

def compare_videos(video1: str, video2: str, max_frames: int = 100, threshold: int = 5) -> float:
    """Compare two videos frame-by-frame using perceptual hashing. Returns 0.0-1.0."""
    cap1, cap2 = cv2.VideoCapture(video1), cv2.VideoCapture(video2)
    total, matches = 0, 0
    for _ in range(max_frames):
        ret1, f1 = cap1.read()
        ret2, f2 = cap2.read()
        if not ret1 or not ret2:
            if ret1 == ret2:  # both ended
                break
            cap1.release(); cap2.release()
            return matches / max(total, 1)
        h1 = imagehash.phash(Image.fromarray(cv2.cvtColor(f1, cv2.COLOR_BGR2RGB)))
        h2 = imagehash.phash(Image.fromarray(cv2.cvtColor(f2, cv2.COLOR_BGR2RGB)))
        total += 1
        if h1 - h2 <= threshold:
            matches += 1
    cap1.release(); cap2.release()
    return matches / max(total, 1)

Comparing Images (SSIM)

from skimage.metrics import structural_similarity as ssim

def compare_images(img1_path: str, img2_path: str, base_score: float = None) -> float:
    """Compare two images using SSIM. Returns 0.0-1.0."""
    img1 = Image.open(img1_path).convert("L")
    img2 = Image.open(img2_path).convert("L")
    # Resize both to the smaller width and height independently
    new_size = (min(img1.size[0], img2.size[0]), min(img1.size[1], img2.size[1]))
    img1 = img1.resize(new_size, Image.Resampling.LANCZOS)
    img2 = img2.resize(new_size, Image.Resampling.LANCZOS)
    score = ssim(np.array(img1), np.array(img2))
    if base_score is not None:
        return max(0, (score - base_score) / (1 - base_score)) if score >= base_score + 0.01 else 0
    return score

---

3. Playlist Operations (setup-gen & reward-gen)

Reading Playlist via HTTP Interface

def get_vlc_playlist(host: str = "localhost", port: int = 8080, password: str = "password") -> list:
    """Fetch VLC playlist as list of dicts with id, name, uri, duration."""
    resp = requests.get(f"http://{host}:{port}/requests/playlist.xml", auth=("", password), timeout=5)
    tree = ElementTree.fromstring(resp.content)
    items = []
    for leaf in tree.iter("leaf"):
        items.append({
            "id": leaf.get("id"),
            "name": leaf.get("name", ""),
            "uri": leaf.get("uri", ""),
            "duration": int(leaf.get("duration", 0)),
        })
    return items

def verify_playlist_count(expected_count: int, host: str = "localhost", port: int = 8080) -> bool:
    """Verify the number of items in the playlist."""
    items = get_vlc_playlist(host, port)
    return len(items) == expected_count

def verify_playlist_order(expected_names: list, host: str = "localhost", port: int = 8080) -> bool:
    """Verify playlist item order by filename substrings."""
    items = get_vlc_playlist(host, port)
    if len(items) != len(expected_names):
        return False
    for item, expected in zip(items, expected_names):
        if expected.lower() not in item["name"].lower() and expected.lower() not in item["uri"].lower():
            return False
    return True

def verify_playlist_contains(filename: str, host: str = "localhost", port: int = 8080) -> bool:
    """Check if a specific file is in the playlist."""
    items = get_vlc_playlist(host, port)
    return any(filename.lower() in item["name"].lower() or filename.lower() in item["uri"].lower()
               for item in items)

Adding Items to Playlist via HTTP

def vlc_playlist_add(uri: str, host: str = "localhost", port: int = 8080, password: str = "password"):
    """Add a media file to VLC playlist via HTTP interface."""
    from urllib.parse import quote
    requests.get(
        f"http://{host}:{port}/requests/status.xml?command=in_enqueue&input={quote(uri)}",
        auth=("", password), timeout=5
    )

def vlc_playlist_clear(host: str = "localhost", port: int = 8080, password: str = "password"):
    """Clear the VLC playlist."""
    requests.get(
        f"http://{host}:{port}/requests/status.xml?command=pl_empty",
        auth=("", password), timeout=5
    )

---

4. Subtitle Operations (setup-gen & reward-gen)

Creating Subtitle Files

def create_srt_file(output_path: str, entries: list):
    """Create an SRT subtitle file.
    entries: list of (index, start_time, end_time, text) tuples
    start_time/end_time format: "HH:MM:SS,mmm"
    """
    with open(output_path, "w", encoding="utf-8") as f:
        for idx, start, end, text in entries:
            f.write(f"{idx}\n{start} --> {end}\n{text}\n\n")

# Example
create_srt_file("/home/user/Desktop/subs.srt", [
    (1, "00:00:01,000", "00:00:04,000", "Hello, welcome!"),
    (2, "00:00:05,000", "00:00:08,000", "This is a subtitle test."),
    (3, "00:00:09,000", "00:00:12,000", "Goodbye!"),
])

VLC Subtitle Configuration (vlcrc)

# Subtitle-related vlcrc options
set_vlcrc_option("sub-auto-detect-file", "1")           # auto-detect subtitle files: 0=off, 1=on
set_vlcrc_option("sub-autodetect-path", ".")             # subtitle search paths
set_vlcrc_option("sub-language", "en")                   # preferred subtitle language
set_vlcrc_option("freetype-fontsize", "24")              # subtitle font size
set_vlcrc_option("freetype-color", "16777215")           # subtitle color (decimal RGB, 16777215 = white)
set_vlcrc_option("freetype-rel-fontsize", "16")          # relative font size: 10=smaller, 20=small, 16=normal
set_vlcrc_option("sub-margin", "50")                     # subtitle margin from bottom (pixels)

Verifying Subtitle State via HTTP

def verify_subtitle_track(expected_track_id: int = None,
                          host: str = "localhost", port: int = 8080) -> bool:
    """Verify current subtitle track. track_id=-1 means subtitles disabled."""
    tree = get_vlc_status(host, port)
    for cat in tree.findall('information/category'):
        if "subtitle" in cat.get("name", "").lower() or cat.get("name") == "Subtitle":
            return True  # subtitle track exists
    # Check via streams info
    for info in tree.findall('.//info[@name="Type"]'):
        if info.text and "subtitle" in info.text.lower():
            return True
    return expected_track_id == -1  # if expecting disabled, return True when not found

def verify_srt_content(srt_path: str, expected_entries: int = None,
                       contains_text: str = None) -> bool:
    """Verify SRT file content."""
    if not os.path.isfile(srt_path):
        return False
    with open(srt_path, "r", encoding="utf-8") as f:
        content = f.read()
    blocks = [b.strip() for b in content.strip().split("\n\n") if b.strip()]
    if expected_entries is not None and len(blocks) != expected_entries:
        return False
    if contains_text is not None and contains_text not in content:
        return False
    return True

---

5. Audio Metadata Verification

import json, subprocess

def get_audio_info(audio_path: str) -> dict:
    """Extract audio metadata using FFprobe."""
    result = subprocess.run([
        "ffprobe", "-v", "quiet", "-print_format", "json",
        "-show_format", "-show_streams", audio_path
    ], capture_output=True, text=True)
    return json.loads(result.stdout) if result.stdout else {}

def verify_audio_sample_rate(audio_path: str, expected_rate: int) -> bool:
    """Verify audio sample rate (e.g., 44100, 48000)."""
    info = get_audio_info(audio_path)
    for stream in info.get("streams", []):
        if stream.get("codec_type") == "audio":
            return int(stream.get("sample_rate", 0)) == expected_rate
    return False

def verify_audio_channels(audio_path: str, expected_channels: int) -> bool:
    """Verify audio channel count (1=mono, 2=stereo)."""
    info = get_audio_info(audio_path)
    for stream in info.get("streams", []):
        if stream.get("codec_type") == "audio":
            return stream.get("channels") == expected_channels
    return False

def verify_audio_bitrate(audio_path: str, min_bitrate: int = None, max_bitrate: int = None) -> bool:
    """Verify audio bitrate (bits/sec). Use format-level bitrate as fallback."""
    info = get_audio_info(audio_path)
    bitrate = None
    for stream in info.get("streams", []):
        if stream.get("codec_type") == "audio" and stream.get("bit_rate"):
            bitrate = int(stream["bit_rate"])
            break
    if bitrate is None:
        bitrate = int(info.get("format", {}).get("bit_rate", 0))
    if bitrate == 0:
        return False
    if min_bitrate is not None and bitrate < min_bitrate:
        return False
    if max_bitrate is not None and bitrate > max_bitrate:
        return False
    return True

def verify_audio_duration(audio_path: str, expected_duration: float, tolerance: float = 1.0) -> bool:
    """Verify audio file duration in seconds."""
    info = get_audio_info(audio_path)
    duration = float(info.get("format", {}).get("duration", 0))
    return abs(duration - expected_duration) <= tolerance

def verify_audio_codec(audio_path: str, expected_codec: str) -> bool:
    """Verify audio codec (e.g., 'mp3', 'aac', 'vorbis', 'pcm_s16le')."""
    info = get_audio_info(audio_path)
    for stream in info.get("streams", []):
        if stream.get("codec_type") == "audio":
            return expected_codec.lower() in stream.get("codec_name", "").lower()
    return False

---

6. Gradual Scoring Pattern (reward-gen)

def compute_reward(task_type: str = "config", config_checks: list = None,
                   media_src: str = None, media_golden: str = None,
                   playback_checks: list = None) -> float:
    """Compute a 0.0-1.0 reward score for VLC tasks.

    task_type: "config" (vlcrc settings), "media" (audio/video manipulation),
               "playback" (player state), "mixed" (combination)
    config_checks: list of (check_fn, weight) tuples for vlcrc verification
    playback_checks: list of (check_fn, weight) tuples for HTTP status checks
    """
    score = 0.0
    total_weight = 0.0

    # Dimension 1: Config verification (weight: 0.4 for config tasks)
    if config_checks:
        w = 0.4 if task_type == "config" else 0.2
        total_weight += w
        passed = sum(1 for fn, _ in config_checks if fn())
        score += w * (passed / len(config_checks))

    # Dimension 2: Media file comparison (weight: 0.4 for media tasks)
    if media_src and media_golden:
        w = 0.4 if task_type == "media" else 0.25
        total_weight += w
        sub_score = 0.0

        # Check file existence
        if os.path.isfile(media_src):
            sub_score += 0.2

            # Duration match
            src_info = get_audio_info(media_src) if media_src.endswith((".mp3", ".wav", ".ogg")) else {}
            gld_info = get_audio_info(media_golden) if media_golden.endswith((".mp3", ".wav", ".ogg")) else {}
            src_dur = float(src_info.get("format", {}).get("duration", 0))
            gld_dur = float(gld_info.get("format", {}).get("duration", 0))
            if gld_dur > 0 and abs(src_dur - gld_dur) < 2.0:
                sub_score += 0.3

            # Content similarity (audio or video)
            try:
                if media_src.endswith((".mp3", ".wav", ".ogg", ".flac")):
                    similarity = compare_audios(media_src, media_golden)
                else:
                    similarity = compare_videos(media_src, media_golden)
                sub_score += 0.5 * similarity
            except Exception:
                pass

        score += w * sub_score

    # Dimension 3: Playback state (weight: 0.3 for playback tasks)
    if playback_checks:
        w = 0.3 if task_type == "playback" else 0.15
        total_weight += w
        passed = sum(1 for fn, _ in playback_checks if fn())
        score += w * (passed / len(playback_checks))

    # Dimension 4: VLC process is running (weight: 0.1 — sanity check)
    w = 0.1
    total_weight += w
    result = subprocess.run(["pgrep", "-f", "vlc"], capture_output=True)
    if result.returncode == 0:
        score += w

    return score / total_weight if total_weight > 0 else 0.0

---

7. Bitter Lessons

1. VLC must be killed before modifying vlcrc. VLC overwrites vlcrc on exit. If you edit vlcrc while VLC is running, your changes are lost when VLC closes. Always pkill -f vlc, wait, then edit.

2. Commented-out vlcrc keys use default values. A line #qt-bgcone=1 means "use default (1)". To change a setting, you must uncomment it (remove #) AND set the value. Just appending a new line won't work if the commented line is parsed first.

3. vlcrc key matching must check for exact key=. Searching for "play-and-exit" in a line also matches "play-and-exit-title". Always match f"{key}=" to avoid false positives.

4. global-key-play-pause uses empty string for "disabled". An empty value (global-key-play-pause=) means disabled. Any non-empty value (e.g., Space) means enabled. This differs from other boolean 0/1 keys.

5. qt-slider-colours is semicolon-separated RGB triplets. Format: R;G;B;R;G;B;R;G;B;R;G;B (4 color groups of 3 values = 12 values). Parse with split(";") then group by 3.

6. VLC HTTP interface status XML paths vary by media type. Filename may appear under info[@name="filename"], info[@name="title"], or info[@name="uri"] depending on whether it's local file, network stream, or HLS. Always check multiple paths.

7. HLS streams show filename, not full URL. When playing https://cdn.example.com/stream/master.m3u8, VLC status may show just master.m3u8 as the filename. Compare basename of the expected URL as fallback.

8. Config changes need VLC restart to take effect. After modifying vlcrc, kill and restart VLC. The evaluator often uses a postconfig step that restarts VLC to verify persistence.

9. Audio comparison uses MFCC + DTW, not waveform. Direct waveform comparison fails due to encoding differences. The evaluator extracts MFCC features and uses Dynamic Time Warping. Similarity score uses exponential decay: exp(-normalized_distance).

10. Video comparison uses perceptual hashing, not pixel diff. The evaluator computes imagehash.phash() per frame and tolerates hash distance up to a threshold. This handles minor encoding artifacts but requires both videos to have matching frame counts.

11. Image SSIM comparison resizes to smaller image. When comparing images of different sizes, both are resized to min(size1, size2). The reference_base_result option provides incremental scoring above a baseline.

12. VLC HTTP auth uses empty username. The HTTP interface uses basic auth with username "" (empty string) and a password. Default password is "password". Forgetting this causes 401 errors.

13. VLC_VERBOSE=-1 suppresses console spam. Without this environment variable, VLC prints hundreds of debug lines. Set it in the subprocess environment to keep logs clean.

14. vlcrc uses = without spaces. The format is key=value with no spaces around =. Adding spaces breaks parsing. When writing, always use f"{key}={value}" directly.

15. Playlist XML uses <leaf> elements, not <item>. VLC's playlist XML nests <leaf> inside <node> elements. Each <leaf> has id, name, uri, duration attributes. Don't search for <item> tags.

16. Playlist duration is in seconds (integer), not milliseconds. Unlike some media APIs, VLC playlist XML reports duration in whole seconds. -1 means unknown/live stream.

17. Subtitle auto-detection requires matching filenames. VLC looks for .srt/.sub files with the same base name as the video (e.g., video.mp4 → video.srt). Setting sub-auto-detect-file=1 alone won't work if filenames differ.

18. Audio bitrate from FFprobe may be at stream or format level. Some codecs (e.g., VBR MP3) don't report per-stream bitrate. Fall back to format.bit_rate when stream.bit_rate is missing.

19. SRT files must use CRLF or LF line endings and --> separator. Malformed SRT with wrong arrow (->, —>) or missing blank lines between entries will not load in VLC.

20. For vision-based reward, read BEFORE from _initial_reference. The agent and golden_patch both overwrite the canonical artifact. initial_setup.py saves a reference copy at {TASK_ID}_initial_reference.<ext> for reward.py to use as the BEFORE image in call_vision_judge() / call_video_vision_judge().