agentsclimarketplace

Audio signal flow explainer

Skill kunitoki/sonic-skills/skills/audio-signal-flow-explainer

Modular Markdown-based audio skills for AI agents and developers, covering signal processing, synthesis, effects, analysis, and spatial audio.

Install
npx -y skills add kunitoki/sonic-skills --skill audio-signal-flow-explainer

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 14 stars14 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Traces and documents signal paths, bus/aux architecture, and sidechain connections in audio systems. Surfaces hidden coupling between audio components. Use when the user asks to trace a signal chain, explain audio routing, understand an audio graph, locate sidechain sources, or describe how audio flows through a plugin or engine. Trigger on phrases like "trace this signal chain", "explain this audio routing", "what calls what in this audio graph", "where does the sidechain come from", "how does audio flow through this plugin".

SKILL.md

3.4 KB, as published. Nobody here has run it

Audio Signal Flow Explainer

Core principle: Audio signal flow is a directed graph. Tracing it reveals hidden coupling, latency paths, and places where the signal can be corrupted.

Step 1 — Map the graph

  • Find all sources (oscillators, audio inputs, file readers, generators)
  • Find all processors (filters, effects, gain stages, dynamics)
  • Find all routers (bus sends, parallel splits, sidechain taps, mixer channels)
  • Find all sinks (audio outputs, recorders, meters, analyzers)
  • Note parallel paths — any branch that rejoins a main path
  • Note feedback loops — any path where output feeds back into an earlier node
  • Note sidechain connections — signals that control a processor without passing audio through it

Step 2 — Identify each node type

Node typeWhat to note
Source (oscillator, audio input, file reader)Sample rate, channel count, sync source
Processor (filter, effect, gain)In/out channel count, latency added, stateful or stateless
Router (bus send, parallel split, sidechain tap)Where signal copies go, gain staging at split
Sink (audio output, recorder, meter)Expected format, callback timing, buffer size

Step 3 — Surface issues

  • Latency compensation mismatches on parallel paths
  • Channel count mismatches between connected nodes
  • Sidechain input not wired — processor receives silence as control signal
  • Uninitialized state in feedback loops (DC buildup, NaN propagation)
  • DC leaking into output (missing highpass or DC-blocking filter)
  • Sample rate mismatch between source and downstream processor
  • Gain staging that causes clipping before a later limiter or saturation stage
  • Missing denormal protection in recursive filter paths

Step 4 — Write the signal flow description

Use this format:

## Signal Flow: [component/system]

### Graph
Source → [gain: 1.0] → [Biquad LPF] → [Comp sidechain tap] → [Output]
                                            ↓
                                       [Compressor] → [Output]

### Latency path
Source → LPF (0 samples) → Compressor (64 samples lookahead) → Output
Total: 64 samples

### Issues found
- Parallel path to Output has no latency compensation for the 64-sample compressor delay

Quick reference

PatternWhat to check
Parallel split rejoiningLatency of each branch must match before summing
Sidechain compressorVerify sidechain input is connected and level-matched
Send/return busCheck send gain, return gain, and whether send is pre- or post-fader
Feedback loopConfirm a delay of at least 1 sample exists to break algebraic loop
Multi-rate graphConfirm resampling nodes at every rate boundary

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.