- Home
- Skills
- Workflow Automation
- Audio Toolkit: Track Analysis and FX
More screenshots
Works with the AI Agents you already use
Audio Toolkit: Track Analysis and FX
Measure before you touch: dual-engine tempo and key verdicts, honest LUFS, onset maps, staged pedalboard FX, and feature envelopes for audio-reactive visuals.
$5
Audio Toolkit: Track Analysis and FX
Example session with this skill installed
I found an unreleased WAV in an old project folder — no notes, no session file. Before I cut a lyric video for it, tell me what I'm working with: real tempo, key, how loud it actually is, and where the hits are. Don't change the file.
- Read your context and instructions
- Compiled the audio toolkit
Decode discipline first: the file arrived as a 44.1 kHz WAV, so it was verified and used as the single analysis source; the original was never written to.
Tempo, two engines. essentia: 118.71 BPM, 39 beats, first grid points at 0.49, 0.98, 1.49, 1.97 — a beat every ~0.5 seconds. aubio: 121.85 BPM. Decision-table read: the two agree within ~5 BPM with no 2x ratio, so the verdict is trust it — nominal 120 BPM, and the essentia beat array is the grid to cut against.
Key: A minor, extraction strength 0.732 — above the 0.6 trust line, so the label ships without a caveat.
Onsets: 120 detected across 20 seconds. The strong hits land on the beat grid; the detections between them are offbeat snare cracks and soft pad attacks. For the video: primary cut points are the on-grid hits, offbeat hits marked as secondary accents.
Loudness: integrated −15.59 LUFS, true peak +1.60 dBTP. That true peak is the finding of the whole pass — the file overshoots 0 dBTP, so it will clip when encoded to a lossy format despite looking tame in a waveform view. Recommendation recorded in the report: pull the ceiling down about 2 dB and re-measure before this bounce is used as a video soundtrack or sent anywhere. No FX was staged to mask it; a loudness defect gets fixed at the source, not painted over.
Report delivered in the standard template, confidence rubric 9/10 — provenance docked one point because the decode step was a verification rather than a conversion. Safe to build the lyric video on: 120 BPM grid, A minor, cuts on the on-grid onsets.
Connects securely to your tools. The creator never sees your data.
About this skill
This skill measures a music track before anyone is allowed to have an opinion about it. Every analysis runs on a single decoded WAV copy — never the original — so every timestamp in the session agrees with every other, and any FX experiment happens on staging files with their parameters logged.
Tempo is the headline discipline. Two independent estimators (essentia and aubio) run on every track, and a decision table turns their agreement or disagreement into a verdict: agreement means trust it, a clean 2x split means a half-time or double-time feel to report honestly, and a wild disagreement means re-run on a steady section instead of averaging two numbers into a tempo nothing is playing at. Key comes with its extraction strength attached, and weak readings are labeled as suggestions or dropped entirely on atonal material.
Loudness is measured the way platforms measure it — integrated LUFS plus true peak from ffmpeg's loudnorm — with an explicit warning against presenting raw energy values as LUFS. Onset maps mark the transients worth cutting a video on, YIN pitch contours run behind a confidence floor so the tracker doesn't invent melodies, and pedalboard FX chains follow a fixed fix → shape → space → color order, auditioned on a 30-second excerpt before any full render.
The deliverable is a report in a fixed template, scored on a measurement confidence rubric (tempo, key, loudness, timing, provenance) that must reach 8 of 10 with no zero in tempo or provenance before anyone builds on the numbers — plus, when asked, an envelope JSON of per-window energy bands and a landmarks file that audio-reactive visuals can consume directly.
The analysis is mono-summed: on very wide stereo material the report is flagged as mid-weighted and cannot speak for the stereo image. Tempo and key estimators assume a mostly steady tempo and tonal material — live recordings with real tempo drift get a range and a drift note instead of a single BPM, and atonal or noise-based tracks get spectrum and onsets instead of a key label. FX staging produces audition files and parameter logs; final mixing and delivery finishing are separate crafts this skill deliberately does not claim.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago