ytslice 1.1.0

Batch-extract time-bounded audio/video segments from long YouTube videos via JSON config.

ytslice

Batch-extract time-bounded audio/video segments from long YouTube videos via a JSON config — cached, fail-soft, no manual yt-dlp + ffmpeg.

---

Prerequisites

Dependency	Why	Install
Python ≥ 3.10	runtime	apt install -y python3 python3-pip
ffmpeg (+ ffprobe)	cuts segments, parses durations	apt install -y ffmpeg
Deno on PATH	yt-dlp's n-parameter challenge solver (see below)	curl -fsSL https://deno.land/install.sh | sh then export PATH="$HOME/.deno/bin:$PATH"

Why Deno

YouTube's player obfuscates stream URLs behind a JavaScript challenge (the n parameter). yt-dlp resolves it via the ejs:github remote-components plugin, which requires a JS runtime — deno is the default. Without deno on PATH, downloads of newer YouTube videos fail mid-stream with HTTP 403 Forbidden and the in-process fallback emits a warning. Install deno via the one-liner above, then deno --version should print a version before you proceed.

---

Install

pip install .
# or, for development:
pip install -e ".[dev]"

The ytslice CLI is installed via [project.scripts].

---

Quick start

config.json:

{
  "output_dir": "./output",
  "videos": [
    {
      "url": "https://www.youtube.com/watch?v=REPLACE_ME",
      "quality": "720",
      "segments": [
        { "name": "intro",     "start": "0:00:00", "end": "0:00:30", "mode": "audio" },
        { "name": "highlight", "start": "0:01:00", "end": "0:02:15", "mode": "both"  }
      ]
    }
  ]
}

Run:

ytslice --config config.json

Expected outputs (one segment per file, plus both produces a pair):

output/
└── <video_id>/
    ├── intro-<video_id>.mp3
    ├── highlight-<video_id>.mp3
    └── highlight-<video_id>.mp4

---

Config schema

Top level:

Field	Type	Required	Notes
output_dir	string	no	default ./output; outputs go to <output_dir>/<video_id>/
videos	array	yes	non-empty list of video objects

Per video:

Field	Type	Required	Notes
url	string	yes	any URL yt-dlp accepts
quality	string	no	one of `best
segments	array	yes	non-empty list of segment objects

Per segment:

Field	Type	Required	Notes
name	string	yes	becomes the on-disk basename after sanitization
start	string	yes	H:MM:SS or MM:SS, minutes/seconds 0–59
end	string	yes	same format; must be > start
mode	string	yes	one of audio (MP3), video (MP4 with audio), both (MP3 + MP4)

The quality enum is exactly what ytslice/config.py validates against. See config.example.json for a runnable example.

---

CLI flags

Flag	Description
--config <path>	path to JSON config (required)
--output-dir <path>	override output_dir from config
--keep-cache	keep source in cache after a successful run (disables auto-evict)
--no-cache	bypass cache: do not look up, do not store; downloads to a tempdir and removes it after the run
--clear-cache	empty the cache root before the run
--dry-run	validate config and exit without downloading or cutting
-h, --help	show help

--no-cache and --keep-cache are mutually exclusive.

--verbose was removed in 1.1.x — use PYTHONLOGLEVEL=DEBUG if you need more detail.

Environment variables

Variable	Effect
YTSLICE_NO_PROGRESS=1	force-disable the rich progress UI (non-TTY opt-out)
YTSLICE_CACHE_DIR=<path>	override cache root (default ~/.cache/ytslice/)
YTSLICE_E2E=1	enable network-using tests under pytest

The progress UI also auto-disables when stderr is not a TTY.

---

Cache lifecycle

Mode	Behavior
default	per-video auto-evict on success — the source file is removed from ~/.cache/ytslice/ once all that video's segments succeed
--keep-cache	source files persist (useful when iterating on segment timestamps)
--no-cache	source is downloaded to a tempdir, used once, and removed after the run
--clear-cache	the entire cache root is purged at the start of the run (sentinel-guarded)

A failed segment keeps the source so a rerun does not re-download. Cache root is ~/.cache/ytslice/ (override via YTSLICE_CACHE_DIR); the directory is marked with a .ytslice-cache sentinel that --clear-cache refuses to purge directories without.

---

Output filename rules

The full V1 rule, in order:

<sanitize_filename(seg.name)>-<video_id><collision_suffix?><extension>

• sanitize_filename strips path separators, control chars, leading/trailing dots+spaces; collapses whitespace; truncates at 200 chars; falls back to unnamed on empty input.
• <video_id> is the YouTube ID resolved by yt-dlp (e.g. dQw4w9WgXcQ).
• <collision_suffix?> is empty when the target path is free; otherwise _1, _2, … (1-indexed, no zero-pad), allocated to the lowest free index. For both-mode segments, the same _N is applied to both the .mp3 and the .mp4 so the pair never splits across different suffixes.
• <extension> is .mp3 / .mp4 per mode.

Rerun caveat: re-running the same config without clearing ./output/<video_id>/ will produce _1, _2, … duplicates because pre-existing files are treated as collisions (by design — no silent overwrites). Either rm -rf ./output/<video_id>/ between runs, or rely on the _N accumulation if you want a versioned history.

---

Clean-Debian walkthrough

End-to-end in debian:stable-slim (container runs as root — no sudo; on a non-container Debian with a non-root user, prefix the apt commands with sudo yourself):

docker run --rm -it debian:stable-slim bash

# system deps
apt update
apt install -y python3 python3-pip python3-venv ffmpeg curl git

# deno (for yt-dlp's n-challenge solver)
curl -fsSL https://deno.land/install.sh | sh
export PATH="$HOME/.deno/bin:$PATH"
deno --version

# ytslice
git clone https://gitlab.com/MisterJB/ytslice.git
cd ytslice
python3 -m venv .venv
. .venv/bin/activate
pip install .

ytslice --help

# run the example (replace REPLACE_ME with a real YouTube ID first)
cp config.example.json config.json
sed -i 's/REPLACE_ME/dQw4w9WgXcQ/' config.json
sed -i 's/ANOTHER/dQw4w9WgXcQ/' config.json
ytslice --config config.json

# verify outputs
ls -la output/dQw4w9WgXcQ/
ffprobe -v error -show_entries format=duration output/dQw4w9WgXcQ/*.mp3 | head

# verify cache auto-evicted on success
ls -la ~/.cache/ytslice/
# Expected: directory exists, contains the .ytslice-cache sentinel,
# but no <video_id>.mp4 — auto-evict-on-success worked.

---

Testing

pytest                  # offline suite (default; mocks downloader + cutter)
YTSLICE_E2E=1 pytest    # additionally runs network-using tests

Done-criteria coverage lives in tests/test_done_criteria.py, mapping every PROJECT.md success metric to a runnable assertion.

---

License

GPL-3.0-or-later. See the LICENSE file for the full text.

---

Changelog (1.1.x)

• Added --keep-cache, --no-cache, --clear-cache, --dry-run.
• Added per-video quality field (best | 1080 | 720 | 480 | 360, default 1080).
• Added rich progress UI (auto-disables in non-TTY; opt out via YTSLICE_NO_PROGRESS=1).
• Filename collisions now resolved as _1, _2, … (pair-consistent for both mode).
• LICENSE: GPL-3.0-or-later (PEP 639 SPDX).
• Removed: --verbose (use PYTHONLOGLEVEL=DEBUG).