More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Sound Check: Give Your Agent Ears for Audio

    1

    Your agent can write code that renders a film, a demo or a game with sound.

    $4.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Sound Check: Give Your Agent Ears for Audio

    Sound Check: Give Your Agent Ears for Audio

    Example session with this skill installed

    My demo cuts to the title card at 1.20 seconds. Check that hit_late.wav lands on it, 40 ms tolerance.

    • Read your context and instructions
    • Compiled the sound check

    Verdict: the sync check fails. The sound starts 80 ms after the cut, outside the 40 ms tolerance.

    MeasurementValue
    Cut1.20 s
    First onset1.2802 s
    Offset+80.2 ms (positive is late)
    Peak-6.0 dBFS, no clipping
    Decay320 ms

    Sounds like (inferred from measurements, not heard): a short transient with a 320 ms decay, warm, tonal near 65 Hz. The loudness figure and the spectrum are estimates.

    Next step: start the sound 80 ms earlier, write it to a new file, and probe again. After the change the same check reports offset +0.2 ms, pass, no findings.

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

    About this skill

    Your agent can write code that renders a film, a demo or a game with sound. It cannot hear what it made. It cannot tell a clipped hit from a clean one, spot a sound landing 80 ms after the cut, or notice a silent gap in the middle.

    This skill gives it evidence instead. Ask "is this audio clipping?" or "does the hit land on the cut at 1.20 s?" and the agent measures the file and answers with numbers: peak and clipping runs, silence before, after and inside the sound, onset times, attack and decay, a tempo estimate with a confidence figure, brightness and a five-band energy split, stereo correlation, and for each cut time the offset in milliseconds with a pass or fail against your tolerance.

    It also writes sounds. Nine presets (click, tick, pop, thump, whoosh, riser, hit, chime, drop) are synthesised from code and are deterministic for a given seed. The loop is make, measure, change one parameter, measure again, until the numbers match the brief.

    How it works. Two bundled Python scripts, standard library only. audio_probe.py reads a WAV (PCM 8, 16, 24 or 32-bit, or 32/64-bit float) and prints text or JSON. sfx_synth.py writes a WAV. The agent reads the numbers and explains them. A bounded cost keeps long files quick: at most 1200 seconds are read, and the spectrum uses at most 64 frames.

    What makes it different

    • Honest about limits: measurements are not hearing, and the skill says so. Loudness is a plain RMS estimate and is never called LUFS. Tempo and spectrum are labelled as estimates, and "not enough rhythm" is a valid answer.
    • Evidence, not opinion: no model, so no "this sounds expensive". It can say a sound is 4 dB louder or 80 ms late.
    • Safe by design: the probe only reads, the synthesiser writes only the one path you give it, and neither installs anything.

    Where it stops. WAV only (convert other formats first). It cannot judge taste, quality or genre, and it finds clipping only where samples sit at full scale.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Productivity.

    WAV only: mp3, ogg, flac and m4a must be converted first, and the skill never installs a converter. Measurements are not hearing: there is no model in the scripts, so no judgement of taste, quality or genre. Loudness is a plain RMS estimate over gated 400 ms blocks, not LUFS. Tempo and spectrum are estimates; tempo can be double or half the felt tempo and returns "not enough rhythm" when there is no steady pulse. Clipping is found only where samples sit at full scale. An onset hidden under the tail of a louder sound may not register. At most 1200 seconds of audio are read and at most 64 spectrum frames are used. The sound-effect presets are small synthesis recipes, not recordings.

    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 buying a skill.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed today

    Software
    Python 3.10+

    WAV files only. No packages, no network access, no API key.

    Permissions

    File Scopes

    Reads
    the
    WAV
    file
    you
    name
    (audio_probe.py
    read-only)
    Writes
    one
    only
    to
    path
    give
    (sfx_synth.py);
    never
    overwrites
    unless
    pass
    --force

    Runs two Python scripts from temporary files. No network access, no environment variables, nothing installed.

    Listedtoday

    What's inside

    Frequently Asked Questions

    No. It measures samples and says what the numbers show. The one-line "sounds like" is inferred from measurements and is labelled that way.

    No. It is a plain RMS of gated 400 ms blocks with no frequency weighting, and the skill never calls it LUFS.

    No. It uses only the Python standard library. If your file is not a WAV, the skill asks you to convert it with a tool you already have.

    No. The probe only reads. The synthesiser writes only the path you give it, and refuses if a file is already there unless you pass --force.

    10 minutes of 44.1 kHz stereo took 5.6 seconds when I rechecked. The cost is bounded by a 1200 second cap.

    The tests pass on Python 3.10, 3.11, 3.12 and 3.13 on Linux, Windows and macOS (3.10 and 3.13 on the last two).

    Any agent that supports SKILL.md and can run a Python script from a temporary file.

    New in Productivity