Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/Arjun-Chandra-7/Music-SFX itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 25, 2026Music SFX
Music SFX is a local, professional audio-processing sub-agent for the Agentic YouTube Evolution System. It gives autonomous agents and human operators one deterministic engine: a JSON-first CLI, a loopback HTTP API, and a rack-inspired browser studio.
The project is intentionally non-generative in v0.1. It edits authorized source audio with a transparent FFmpeg signal chain, preserves the original, and produces an auditable manifest for every action.
What is included
- Seven curated mastering, voice, social, utility, and creative presets
- Eight bounded controls for tone, dynamics, space, width, blend, and output
- WAV 24-bit, FLAC, MP3 320 kbps, M4A 256 kbps, Ogg, and Opus delivery
- Source/output SHA-256 hashes and ffprobe metadata
- Append-only JSONL events and inspectable per-job manifests
- Rights classification, dry runs, safe output behavior, and explicit overwrite
- Browser waveform, source auditioning, preset browser, live faders, status polling, downloads, and job history
- No Python runtime dependencies beyond the standard library
Quick start
Requirements: Python 3.10+ and FFmpeg (including ffprobe).
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -e .
music-sfx serve
Open http://127.0.0.1:8765. Drop in an authorized audio asset, audition it, choose a preset, adjust the processor, select the rights state and output format, then render. The source is never modified.
Register the autonomous agent skill
From this checkout, run the one-command setup in the Python environment the agent will use:
python3 scripts/setup_agent.py
This installs the editable music-sfx CLI, copies the complete skill contract and its routed references into $CODEX_HOME/skills/music-sfx (or ~/.codex/skills/music-sfx when CODEX_HOME is unset), and verifies the CLI, FFmpeg, ffprobe, and required skill files. Restart Codex afterward so $music-sfx is discovered and can be selected automatically.
To inspect an existing installation without changing it:
python3 scripts/setup_agent.py --check
Set CODEX_HOME before setup or pass --skill-dir /path/to/skills/music-sfx when the agent uses a non-default skill registry. The installer updates only Music SFX files; it does not remove unrelated skills or configuration.
For agent use:
music-sfx inspect samples/narration.wav
music-sfx presets
music-sfx process samples/narration.wav \
--output deliverables/narration-master.wav \
--preset broadcast_voice \
--rights owned \
--actor channel-agent \
--intent "Narration master for episode 4"
Every command prints JSON. A completed render returns exit status 0; blocked and failed renders return 2. Run music-sfx process --help for all options.
Architecture
Channel agent ── CLI / local JSON API ──┐
├─ Policy check → FFmpeg engine → New asset
Human owner ─── Browser studio ─────────┘ │
manifest + append-only events
The same preset definitions, validation, policy checks, and render engine serve all interfaces. The HTTP server binds to 127.0.0.1 by default and has no authentication; keep it local unless you add an authenticated boundary.
Runtime data defaults to .music-sfx/:
.music-sfx/
├── events.jsonl # append-only operational events
├── jobs/ # complete job manifests
├── renders/ # UI and default CLI outputs
└── uploads/ # browser source uploads
See SKILL.md for the autonomous-agent contract, API.md for orchestration, and AUDIO_GUIDE.md before making custom processing decisions.
Development
python3 -m unittest discover -s tests -v
python3 -m music_sfx.cli --help
python3 scripts/setup_agent.py --check
The test suite validates policy boundaries, parameter ranges, filters, manifests, and a real render when FFmpeg is installed.
Safety and scope
Music SFX does not acquire music, infer copyright ownership, publish media, or conceal provenance. Unknown rights are always surfaced as approval_required. Creative output still requires listening review; automated processing cannot detect every clipped transient, phase issue, edit artifact, or content-level problem.
License
MIT. See LICENSE.