Input Audio (ShellAgent Plugin)

Input Audio (ShellAgent Plugin) is the audio entry point for ShellAgent workflows: it takes an audio file - either chosen from a dropdown of what's in ComfyUI's input folder, or one the app uploads - and loads it into ComfyUI's AUDIO format. Once it's in that format, it plugs straight into audio tooling like the pack's Save Audio node or ComfyUI's core audio nodes.

If you've ever had to wrangle audio into ComfyUI, you know the stock story is "there's a Load Audio node, but it lives in the Audio stuff and its file list is filtered." This node is essentially the same loader, renamed as a ShellAgent input and given the input_name/description schema hooks so an app can feed it programmatically.

How it works

On run it resolves the selected file from the input directory and loads it with torchaudio, returning a dict of {waveform, sample_rate} on the AUDIO output - the same shape the core Save Audio node expects. The dropdown (default_value) is populated from audio and video files in your input folder (filtered by content type), so what you see is what the app can hand over. It also has a SHA-256-based IS_CHANGED check, meaning ComfyUI only re-runs downstream nodes when the actual file bytes change, not every queue.

The AUDIO output is a single tensor, not a batch. If your workflow needs multiple audio files, you'll chain multiple instances or handle batching downstream.

Inputs and outputs

• input_name - the variable name the app sees. Default input_audio.
• default_value - the file to load, from the dropdown (drag to upload audio).
• description - optional help text for the app's UI.

Output: AUDIO on the audio socket. Wire it to Save Audio nodes (including this pack's deprecated FLAC savers), to audio-conditioned workflows, or to the Video Combine Encrypt node's audio input for muxing into video.

Installing it

Same one-pack install as everything in this family:

cd ComfyUI/custom_nodes
git clone https://github.com/myshell-ai/ComfyUI-ShellAgent-Plugin

restart, or grab ComfyUI-ShellAgent-Plugin from ComfyUI Manager. No model files. Note this node pulls in torchaudio indirectly via ComfyUI's core audio support - if loading errors out, check your ComfyUI install has audio extras rather than the pack itself.

Common issues

• "Invalid audio file" error. The file isn't in your input directory, or the format isn't one torchaudio can read. Re-upload it.
• The dropdown only shows some files. The list is filtered by audio/video content types, so a file with a misleading extension may not appear.
• Downstream re-runs every time. If something downstream keeps re-triggering, check the file hasn't changed; the node's IS_CHANGED is content-hash based, which is stricter than a filename check.

It's a modest node, but it's the missing piece if your pipeline starts or ends in sound. For the save side, the pack has the FLAC savers - though their display name now says DEPRECATED, so check those before building on them.