How to use the Audio Jingle Generator agent skill

What the audio-jingle skill makes the agent do

An agent skill is a set of instructions and files that an AI agent loads so it can perform a specific task. This one covers audio. It works in three modes: music (jingles and beds), speech (voiceover) and sfx (sound effects). The active project's audioKind setting decides which mode runs.

Before generating anything, the agent writes a short plan covering genre, tempo, script or sound texture, depending on the mode. It then sends the prompt to a configured audio generation service through a dispatcher command. The result is saved into the project folder so it can be previewed with playback controls.

• music: routes to Suno V5 (default), Udio or Lyria 2
• speech: routes to MiniMax TTS (default), Fish or ElevenLabs V3
• sfx: routes to ElevenLabs SFX (default) or AudioCraft

How do you install the audio-jingle skill?

There are two ways to add the skill. Download the ZIP from the skill's page on this site and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder for that project only.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The ZIP contains two files, SKILL.md and example.html. SKILL.md holds the instructions the agent follows, and example.html is the preview page.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with a natural request?

You do not need a special command. The skill responds to trigger words in your request, including music, jingle, bed, voiceover, tts, and sound effect. Chinese triggers 音乐, 配音 and 音效 are also listed.

Describe the sound you want in plain language and say what it is for. Give the length and mood where you can. The skill reads duration from the project metadata, so set audioDuration in seconds when you want a specific length.

• 'Make a 30-second upbeat jingle for our product launch video.'
• 'Write a voiceover for this final script in a warm, calm voice.' (speech mode)
• 'Create a short sound effect: a soft whoosh with a sharp attack.' (sfx mode)

What does a generation look like in practice?

The skill's own example prompt is a 30-second upbeat indie-pop jingle for a coffee shop launch, with warm electric piano lead, brushed drums, gentle bass, and a single sun-soaked choir on the chorus, with no vocals and a loop-friendly tail. Requests this specific give the agent more to work with than a general mood word.

The agent dispatches the job with a command of this form. It fills in the bracketed values from your project and request. The command prints one line of JSON with the filename of the result.

What do you need before it will run?

The skill cannot generate audio on its own. It needs the open-design media dispatcher, which the environment reaches through the OD_NODE_BIN and OD_BIN variables pointing to the daemon tooling. It also needs an account and API key for at least one audio model provider (Suno, Udio, Lyria, MiniMax, Fish, ElevenLabs or AudioCraft), configured in your host application.

The skill's files do not say how API keys are stored, so check your host application's documentation for that.

• Open-design media dispatcher with OD_NODE_BIN and OD_BIN set
• Provider account and API key for each model you choose
• Host application with the provider configured

What safety and licensing points should you know?

The skill runs scripts and makes network calls to the audio providers you configure. Review the files before installing, as you would with any skill that runs code.

The skill does not check rights or model terms. Those depend on the provider you use, so do not treat output as cleared for final broadcast or commercial use without checking the provider's terms yourself.

The skill is not designed for voice cloning or imitating a real person. Its files describe no safeguards for this and do not address consent.

• Runs scripts and uses the network
• No rights or licence checks on generated audio
• Not for cloning or imitating real voices

What tips help you get better results?

For speech, the script is sent literally, so proof it before generating. Even one stray comma can change the cadence. For MiniMax TTS, the voice must be a real voice_id, such as male-qn-qingse, not a description. If you only have a written voice brief, leave out the voice setting so the default applies, or choose an id.

For music, match length to structure: under 30 seconds is a single section, 30 to 90 seconds is intro plus body, and 90 seconds or more is a full arc. For sound effects, one well-described layer works better than a long list of adjectives.

If you want a variation, ask for one specific change, such as moving the tempo from 92 to 108 BPM, rather than asking for something 'different'.