More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Ears — Let AI Agents Actually Listen to Audio

    2

    Give any AI agent a pair of ears: record the audio playing on the computer, render it as a spectrogram, and let the agent itself judge what it hears — instead of guessing from a title.

    Free

    Security scanned
    Ears — Let AI Agents Actually Listen to Audio

    Ears — Let AI Agents Actually Listen to Audio

    Example session with this skill installed

    Have a listen to what's playing on my PC — do you actually like it, and does it go anywhere, or is it flat the whole way?

    • Read your context and instructions
    • Compiled the ears

    Summary

    Recorded a 12-second loopback clip, rendered its spectrogram, and judged it from that
    evidence — no guessing from a title. Verdict: it starts almost silent and then opens up
    into something fuller and brighter around the 4.5-second mark. The payoff is worth the wait;
    before that it's just holding back.

    WhatMeasuredWhat it means
    Clip length12.0 sanalyzed segment
    Loudness (overall)−21.4 dBFS RMSmoderate, not loud
    Loudness trend (first → last third)0.052 → 0.099 (×1.89)clearly builds toward the end
    Brightness / spectral centroid574 → 1606 Hz (×2.8)opens up from muffled to bright
    Shapebuilds and fills out; dark→bright; has peaks越到后面越有劲、越满;音色从偏闷慢慢打开、变亮;中间有起落

    What it actually sounds like (agent's read):

    The first four or five seconds are a warm-up — low, muffled, holding back. Then it lifts:
    the top end opens, the whole thing gets fuller, and it stays there. I like the payoff more
    than the build itself. And that's the point — if I'd only heard the first five seconds I'd
    have called this "flat", which is exactly the wrong call a title-based guess would make.

    Next steps

    • Ask the agent for a second, longer clip to confirm the build holds (or ask for a different track).
    • Try it on an instrumental — there are no lyrics, so it judges purely from the sound image.
    • Optional: the agent can also fetch the current track name and title-verified lyrics, but it
      never invents lyrics if none are found.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Hear the audio, not the title.

    AI agents can't listen — they only see text. Ask one "how does this song sound?" and it invents an answer from the song name. Ears fixes that: it records whatever is playing on the computer, renders it into a spectrogram (a picture of the sound), and optionally fetches the current track name and lyrics. The agent then looks at the real audio and forms its own opinion.

    What it does

    Records system audio (loopback) into a WAV clip. Renders the clip to a spectrogram PNG + an honest "shape" summary (build-up, drops, brightness). Optional: reads the currently playing track via SMTC, and looks up title-verified lyrics. Why it's different

    It never judges for you. The tool only records / renders / reports; the agent does the listening and the opinion. It never invents lyrics — no lyrics means it says nothing about lyrics. Works with any multimodal agent that can run a Python script (Claude Code, Cursor, Copilot, and more). Good for: music companions, "rate what's playing" bots, audio-reactive agents, accessibility tooling.

    ⚠️ Tested on Windows (WASAPI loopback + SMTC). macOS/Linux may work after audio-device tweaks but are untested.

    Developed and tested on Windows (WASAPI + SMTC). Requires OS-level microphone permissions. Does not perform speech-to-text transcription. Silent playback results in empty spectrograms.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 13 days ago

    • Free with an Agensi account

    Requires

    163
    Python 3.9+
    SoundCard
    librosa
    Matplotlib
    soundfile
    Windows 10
    Windows 11

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files

    Allowed Hosts

    music.163.com

    File Scopes

    captures/**

    Records the computer's own audio playback (loopback) to a local WAV file; writes the spectrogram PNG into captures/. The only network call is an optional lyric lookup (title-verified). No audio is ever uploaded anywhere.

    Listed13 days ago

    What's inside

    Frequently Asked Questions

    给它一段正在播放的声音,它把声音画成频谱图,让你的 agent 自己"看"这段声音、自己判断好不好听——而不是看着歌名瞎猜。

    任何能跑 Python 脚本、并且能看图的(多模态)agent,比如 Claude Code、Cursor、Copilot 等。核心是标准 SKILL.md。

    不会。录音、画图全在你本机完成;唯一联网是可选的"按歌名查歌词",不会上传音频。

    因为脚本只能量数字(响度、音高),分不清"宁静"和"压抑"。判断留给你自己的 AI——它看图才看得懂评价

    New in Agents & Orchestration