skills

 # Sonilo Skills Agent skills for [Sonilo](https://sonilo.com/?utm_source=github&utm_medium=oss&utm_campaign=skills-repo)'s licensed music, sound-effects, dubbing, and audio-ducking API. These skills follow the [Agent Skills specification](https://agentskills.io/specification) and can be used with any compatible AI coding assistant. Sonilo's flagship capability is **video-to-music**: hand it a finished video and it composes an original soundtrack matched to the cut — the music follows the pacing, motion, and emotion because the model saw them. Every track is licensed and safe for commercial use. ## Installation These skills follow the community [Agent Skills specification](https://agentskills.io/specification), so any compatible tool can install them via `npx skills add`. This repo is also set up as a **Claude Code plugin marketplace**, so Claude Code users can install the same skills with `/plugin` instead, and a **Muse Code plugin** that also connects the hosted Sonilo MCP server — pick whichever fits your workflow. ### Option 1: `npx skills` (any compatible assistant) ```bash npx skills add sonilo-ai/skills ``` Or install a single skill: ```bash npx skills add sonilo-ai/skills --skill video-to-music ``` ### Option 2: Claude Code plugin ``` /plugin marketplace add sonilo-ai/skills /plugin install skills@sonilo-skills ``` This installs the same thirteen skills (`video-to-music`, `text-to-music`, `video-to-sfx`, `text-to-sfx`, `video-to-sound`, `video-analysis`, `audio-ducking`, `auto-dubbing`, `proofread`, `task-recovery`, `account`, `audio-playback`, `setup-api-key`) as a single Claude Code plugin, discovered directly from their existing top-level directories — no separate copy to keep in sync. It's a skills-only plugin (no MCP server, no bundled tools); see [Configuration](#configuration) below for how to connect the Sonilo MCP server itself. ### Option 3: Muse Code plugin ```bash MUSE_EXPERIMENTAL_PLUGINS=1 muse plugins marketplace add sonilo-skills https://github.com/sonilo-ai/skills MUSE_EXPERIMENTAL_PLUGINS=1 muse plugins install skills@sonilo-skills MUSE_EXPERIMENTAL_PLUGINS=1 muse plugins approve skills ``` Or from a clone: `muse plugins install ./skills --scope user`, then the same `approve`. `approve` is what activates the MCP server — Muse leaves a third-party plugin's server off until it is reviewed; the skills are active as soon as the plugin is installed. `muse skills list --source plugin` shows the thirteen skills. Unlike the Claude Code plugin above, this one also brings the tools: it connects Muse to the hosted Sonilo MCP server (`https://api.sonilo.com/mcp`), so there is no key to paste. The first Sonilo tool call asks you to sign in with a **Sonilo Platform** account ([platform.sonilo.com](https://platform.sonilo.com/sign-up), separate from a consumer sonilo.com account). Plugin management is behind `MUSE_EXPERIMENTAL_PLUGINS` as of Muse Code 1.4.0; what it installs runs without the flag. If you would rather sign in ahead of time, add the same server to `~/.config/muse/settings.json` and run `muse mcp login sonilo` (1.4.0 only logs in to servers listed there): ```json { "mcp_servers": { "sonilo": { "transport": "streamable_http", "url": "https://api.sonilo.com/mcp" } } } ``` **Muse app** (muse.ai, not the CLI): there is no plugin install. Ask Muse to *"build a custom integration to Sonilo — its MCP server URL is https://api.sonilo.com/mcp, streamable HTTP, OAuth sign-in"*, and sign in with your Sonilo Platform account when it asks. ## Available Skills | Skill | Description | |-------|-------------| | [video-to-music](./video-to-music) | Score a video with original music matched to its pacing and emotion (`video_to_music`, `video_to_video_music`) | | [text-to-music](./text-to-music) | Generate music from a text prompt alone, no video (`text_to_music`) | | [video-to-sfx](./video-to-sfx) | Generate sound effects matched to a video, with optional timed segments (`video_to_sfx`, `video_to_video_sfx`) | | [text-to-sfx](./text-to-sfx) | Generate one sound effect from a text description alone (`text_to_sfx`) | | [video-to-sound](./video-to-sound) | Generate music **and** SFX for a video in one balanced, single-charge call (`video_to_sound`, `video_to_video_sound`) | | [video-analysis](./video-analysis) | Analyze a video and get a creative brief — a music section plan plus ready-to-use generation prompts, and a sound-design brief, or just one via `mode` — instead of a track (`analyze_video`) | | [audio-ducking](./audio-ducking) | Duck a music bed under a voice track (or a video's voice track) automatically (`audio_ducking`) | | [auto-dubbing](./auto-dubbing) | Dub a video into other languages with re-voiced speech (`dubbing`) | | [proofread](./proofread) | Transcribe a video and translate the transcript into editable `.srt` files — the review step before dubbing (`proofread`) | | [task-recovery](./task-recovery) | Recover the result of a timed-out generation using its task id (`get_sfx_task`, or `get_generation_task` on the hosted server) | | [account](./account) | Check available services, limits, free-trial allowance, and usage history (`get_account_services`, `get_usage`) | | [audio-playback](./audio-playback) | Play a local audio file through the system's speakers (`play_audio`) | | [setup-api-key](./setup-api-key) | Guide through obtaining a Sonilo API key and connecting the Sonilo MCP server | ### Renamed in v2 Each skill is now named for the tool it documents. If you installed an earlier version, reinstall under the new name — the old ones no longer exist: | Was | Now | |---|---| | `music` | `text-to-music` (text prompt only) and `video-to-music` (scoring a video) | | `sound-effects` | `text-to-sfx` (text prompt only) and `video-to-sfx` (matched to a video) | | `dubbing` | `auto-dubbing` | ### Prompting guidance Sonilo needs no prompt — the video is the context. When you want control, a structured audio brief adds your intent (genre, the moment that must hit, the sounds that must not appear) on top of what the model reads from the cut. That guidance is folded into the skills themselves (ported from [sonilo-prompt-assist](https://github.com/sonilo-ai/sonilo-prompt-assist), now deprecated in favor of this repo): - [references/preflight.md](./references/preflight.md) — pre-flight before any paid call: inspect the video, duration caps, credits, verification - [references/music-prompting.md](./references/music-prompting.md) — style-prompt craft for `video_to_music` - [references/sfx-prompting.md](./references/sfx-prompting.md) — action map → segments craft for `video_to_sfx` - [references/api-claims.md](./references/api-claims.md) — verified API behavior the guidance relies on ## Configuration Each skill's Quick Start shows every way to call that capability — pick whichever fits how you're working: - **MCP, local, signed in** — sign in once with either CLI and the [sonilo-mcp](https://github.com/sonilo-ai/sonilo-mcp) server (0.16.0+) reads the same credential, so the host config holds no secret: ```bash npm install -g sonilo-cli && sonilo login # or: pip install sonilo-cli claude mcp add sonilo -- uvx sonilo-mcp # Claude Code codex mcp add sonilo -- uvx sonilo-mcp # Codex ``` Approving in the browser mints a 90-day key named `cli: <hostname>` in `~/.config/sonilo/credentials.json`; `sonilo whoami` shows what is active and `sonilo logout` revokes it. - **MCP, local, with a key** — for CI, containers, or if you prefer holding one: ```bash claude mcp add sonilo --env SONILO_API_KEY=sk-... -- uvx sonilo-mcp ``` `SONILO_API_KEY` takes precedence over a sign-in wherever both exist. Get a key from the [Sonilo dashboard](https://platform.sonilo.com/dashboard/api-keys), or use the `setup-api-key` skill for guided setup. See the [sonilo-mcp README](https://github.com/sonilo-ai/sonilo-mcp) for Claude Desktop / Codex setup and the full environment variable reference (`SONILO_API_URL`, `SONILO_MCP_BASE_PATH`, `SONILO_MCP_ALLOW_ANY_PATH`, `TIME_OUT_SECONDS`). - **MCP, remote (no API key)** — the [sonilo-claude-plugin](https://github.com/sonilo-ai/sonilo-claude-plugin) for Claude Code, and this repo's Muse Code plugin ([Option 3](#option-3-muse-code-plugin)), connect to a hosted, OAuth-authenticated MCP server (`https://api.sonilo.com/mcp`) instead of running anything locally. In Claude Code: ``` /plugin marketplace add sonilo-ai/sonilo-claude-plugin /plugin install sonilo@sonilo ``` You sign in with your Sonilo Platform account on first use — no key to copy or configure. It carries the same tool set as the local server — every tool these skills document (music and SFX from text/video, video-to-video, video-to-sound, dubbing, ducking, account/usage). - **Python / JavaScript SDK, or the `sonilo` CLI** — see [Client Libraries](#client-libraries) below if you're integrating outside an MCP host entirely. ## Client Libraries Sonilo also ships official client libraries, independent of MCP: | Library | Install | Repo | |---|---|---| | Python SDK | `pip install sonilo` | [sonilo-ai/sonilo-python](https://github.com/sonilo-ai/sonilo-python) | | JS/TS SDK | `npm install sonilo` | [sonilo-ai/sonilo-js](https://github.com/sonilo-ai/sonilo-js) | | CLI | `pip install sonilo-cli` or `npm install -g sonilo-cli` | same repos, `sonilo-cli` package | | Video helpers (local ffmpeg mixing) | `pip install sonilo-video-kit` or `npm install sonilo-video-kit` | same repos, `sonilo-video-kit` package | Both SDKs read `SONILO_API_KEY` from the environment by default; both CLIs additionally fall back to a `sonilo login` credential when it is not set. Each skill's Quick Start includes Python, JavaScript, and (where a command exists) CLI examples alongside the MCP tool call and raw cURL — use whichever matches your integration; they all hit the same underlying API. Calling the HTTP API directly? [references/api_reference.md](./references/api_reference.md) is a one-page REST reference: base URL, auth, every endpoint's parameters and limits, a cURL example for each, and the async task flow. Every generation tool has a 1:1 SDK resource and CLI command (`audio_ducking`, the last holdout, gained both in Python SDK 0.13 / JS SDK 0.14, with the CLI command in PyPI `sonilo-cli` 0.12 / npm `sonilo-cli` 0.13). The one deliberate exception is `play_audio`, which just plays a local file — MCP-only by nature. Where a skill's SDK/CLI coverage differs from its MCP tool in any other way, the skill says so explicitly rather than implying full parity. ## Billing Tools marked with a cost warning make an API call that may incur charges. Self-serve accounts start with a few free runs per service (no card required) — call `get_account_services` to check `trial[service].remaining` before a paid call. `dubbing`'s one free run is a **15-second preview** (first single-language call, no scripts); everything after that bills per language. See the [account](./account) skill. ## Commercial use Audio generated through the Sonilo API can be used commercially — in your own product and by your end users, including ads, client work and monetized channels. The API is billed pay-as-you-go from a prepaid balance on platform.sonilo.com, not through a sonilo.com subscription plan, so no Pro or Premium plan is needed; Enterprise API customers follow their enterprise agreement. For film, TV or broadcast use, or for releasing tracks to Spotify, Apple Music or other streaming services, contact sales first. Full answer: [Can my app's users use audio generated through the Sonilo API commercially?](https://sonilo.com/help/licensing/can-my-apps-users-use-sonilo-api-audio-commercially) ## Evaluations [`evals/`](./evals) holds five trigger evals — does the right skill fire for the right query — covering the routing decisions the v2 split put at risk, including the combined music+SFX request that must reach `video-to-sound` rather than billing twice for the two separate skills. ```bash claude plugin eval skills@sonilo-skills ``` `claude plugin eval` is still early-access gated as of Claude Code 2.1.232, so these cases are schema-valid but not yet runnable by every CLI install. They are routing-only and cannot call Sonilo generation tools. See [evals/README.md](./evals/README.md) for the current status. Functional evals — does the call produce correct audio — are still open; contributions welcome. ## License MIT