Run any Skill in Manus with one click

innexus

Stars8

Forks0

UpdatedJuly 18, 2026 at 20:39

Domain knowledge for the Innexus harmonic-resynthesis instrument — the analysis→resynthesis signal path, the two analysis front-ends, load-bearing invariants (mono harmonic-sieve, per-partial bandwidth), and where the deep docs live. Use when working on Innexus, harmonic/additive resynthesis, sample or live spectral analysis, or Innexus presets — especially outside plugins/innexus/ where the leaf CLAUDE.md won't auto-load.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

rolandzwaga

rolandzwaga/krate-audio

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

SKILL.md

readonly

name	innexus
description	Domain knowledge for the Innexus harmonic-resynthesis instrument — the analysis→resynthesis signal path, the two analysis front-ends, load-bearing invariants (mono harmonic-sieve, per-partial bandwidth), and where the deep docs live. Use when working on Innexus, harmonic/additive resynthesis, sample or live spectral analysis, or Innexus presets — especially outside plugins/innexus/ where the leaf CLAUDE.md won't auto-load.
allowed-tools	Read, Glob, Grep

Innexus — Harmonic Analysis / Resynthesis Instrument

Innexus (AU aumu) analyzes audio into per-frame harmonic snapshots (HarmonicFrame + ResidualFrame) then additively resynthesizes them with evolution/physics/modulation shaping. When editing under plugins/innexus/, its leaf CLAUDE.md auto-loads with build/test/param facts — this skill carries the domain lore that applies when reasoning about Innexus from elsewhere (DSP library, presets, tooling).

Signal path (per voice)

Two input sources feed one frame contract, selected by inputSource_:

Sample mode (≤0.5): drag-dropped WAV → SampleAnalyzer (background std::thread) → immutable SampleAnalysis, published to the audio thread by atomic pointer swap; voice advances currentFrameIndex.
Sidechain mode (>0.5): aux bus 0 downmixed to mono → LiveAnalysisPipeline on the audio thread, emitting one frame per STFT hop.

Frame selection (voice_.morphedFrame): Recall│Capture│ManualFreeze+morph(lerp)│confidence-gated freeze, then applyModulatorAmplitude → applyHarmonicPhysics → broadcastFrameToVoices (copy to each voice). Per voice: HarmonicOscillatorBank (up to kMaxPartials = 96 Gordon-Smith MCF sines; freq/amp/phase seeded from partials, phase state NOT reset across frames = continuity) + exciter{Impact│Bow│Residual} → resonator{Modal│Waveguide, equal-power xfade} → BodyResonance → PhysicalModelMixer(residual↔physical). Σ voices × 1/√activeCount → global SympatheticResonance → master gain → tanh soft-limiter → out.

Analysis pipeline

Both paths share: PreProcessingPipeline (DC → 30 Hz HPF → transient suppress → noise gate) → YIN pitch → dual Blackman-Harris STFT (short 1024/512 for temporal tracking, long 4096/2048 for low-freq resolution) → PartialTracker (Hungarian match + harmonic sieve) → HarmonicModelBuilder (dual-timescale smooth) → residual bands. Offline adds a two-pass polyphony guard: if F0 is unstable/noisy, pass 2 re-runs with a fixed F0 and voiced=false (sieve OFF) so all peaks survive — the conditional poly fix.

Load-bearing constraints (violating these breaks presets, tests, or sound)

Never globally disable the mono harmonic sieve. The sieve runs only when f0.voiced; poly/pass-2 code turns it off conditionally via a neutral voiced=false F0. The monophonic sieved path is load-bearing.
Per-partial bandwidth (0=sine..1=noise) is the historical noise bug. Inflated bandwidth injects noise on every partial. Physics/modulators touch amplitudes only; pass-2 poly sets bandwidth=0 (unreliable without the sieve). Treat any bandwidth write with suspicion.
No frees on the audio thread. Superseded SampleAnalysis goes to pendingDeletion_, freed only in setActive/dtor. currentAnalysis_ is an atomic acquire/release swap.
UI data bridge: sendDisplayData piggybacks a DisplayData block on the DataExchange handler (~30 Hz throttle) with a Tier-3 sharedDisplay_ polling fallback — don't add new IMessage loops.
Param IDs: section bases — 200s = envelope, 400s = harmonic. kPartialCountId=202 is a StringListParameter ("48"/"64"/"80"/"96"), not a range — never swap a registered VST3 param type.
Preset categories are FIXED (7): Brass and Winds, Drums and Perc, Found Sound, Keys, Pads and Drones, Strings, Voice. Never invent new ones.

Deep tooling (read these for real work)

Plugin specifics: plugins/innexus/CLAUDE.md — skeleton, param-ID scheme, innexus_tests target (unit tests/unit/{processor,vst}/ + tests/integration/, run both), pluginval path, AU config.
Frame contract & shared DSP: dsp/include/krate/dsp/processors/harmonic_types.h (HarmonicFrame, Partial, kMaxPartials), residual_types.h, harmonic_oscillator_bank.h.
Analysis internals: plugins/innexus/src/dsp/ (sample_analyzer, live_analysis_pipeline, pre_processing_pipeline, dual_stft_config.h) and shared partial_tracker.h / harmonic_model_builder.h.

More from this repository

same repository

gradus

rolandzwaga/krate-audio

Domain knowledge for the Gradus step-arpeggiator plugin — the MIDI/event signal path (note input → ArpeggiatorCore → MidiNoteDelay echoes → event out + audition), load-bearing invariants (arp param IDs shared with Ruinae, unified save prefix + intentionally-local load path, fixed event-buffer caps), and where the deep tooling lives. Use when working on Gradus, arpeggiators, the shared arp engine, or Gradus presets — especially outside plugins/gradus/ where the leaf CLAUDE.md won't auto-load.

2026-07-228

release

rolandzwaga/krate-audio

Guardrailed version bump + release commit for a single Krate Audio plugin. Edits ONLY version.json, adds the matching CHANGELOG entry, then commits — never pushes. Builds and runs that plugin's tests + pluginval only when the tree is not already verified. Use when the user asks to "release", "bump version", "cut a release", or "ship X.Y.Z" for a plugin.

2026-07-208

testing-guide

rolandzwaga/krate-audio

VST3 plugin testing patterns and practices — the broad, general-purpose testing skill. Use when writing tests, debugging test failures, adding new test cases, when tests don't appear after changes, running the test suites, or wiring integration tests. Covers build-before-test discipline, Catch2 patterns and test doubles, integration testing (processor-level wiring, parameter application, host environment), VST3 validation, and anti-patterns. For deep DSP-verification (FFT aliasing measurement, THD/SNR, click/artifact detection, spectral golden references) use the companion testing-dsp-analysis skill.

2026-07-198

code-review

rolandzwaga/krate-audio

DSP and VST3 specialized code review agent. Use to review code changes for real-time safety violations, thread safety issues, numerical stability problems, layer dependency violations, parameter handling bugs, performance concerns, and cross-platform compatibility. Accepts file paths, git diff ranges, or reviews staged/unstaged changes.

2026-07-128

membrum

rolandzwaga/krate-audio

Domain knowledge for the Membrum drum-synth plugin — per-voice signal path, preset/kit invariants, ring/range constraints, and where the deep tooling lives. Use when working on Membrum, drum synthesis, kits, pads, bodies/exciters, or Membrum preset retuning, especially outside the plugins/membrum/ tree where the leaf CLAUDE.md won't auto-load.

2026-07-128

testing-dsp-analysis

rolandzwaga/krate-audio

Deep DSP-verification techniques — FFT-based aliasing measurement, THD/SNR and harmonic analysis, click/LPC/spectral-anomaly artifact detection, frequency estimation, and deterministic test-signal generation. Use ONLY when verifying the numerical/spectral correctness of a DSP algorithm (waveshaper aliasing, oscillator anti-aliasing, saturation THD, click/discontinuity hunting, golden spectral references). For general test writing, Catch2 patterns, build-before-test, integration, and VST3 validation, use the broader testing-guide skill instead.

2026-07-128

name	innexus
description	Domain knowledge for the Innexus harmonic-resynthesis instrument — the analysis→resynthesis signal path, the two analysis front-ends, load-bearing invariants (mono harmonic-sieve, per-partial bandwidth), and where the deep docs live. Use when working on Innexus, harmonic/additive resynthesis, sample or live spectral analysis, or Innexus presets — especially outside plugins/innexus/ where the leaf CLAUDE.md won't auto-load.
allowed-tools	Read, Glob, Grep

Innexus — Harmonic Analysis / Resynthesis Instrument

Signal path (per voice)

Two input sources feed one frame contract, selected by inputSource_:

Sample mode (≤0.5): drag-dropped WAV → SampleAnalyzer (background std::thread) → immutable SampleAnalysis, published to the audio thread by atomic pointer swap; voice advances currentFrameIndex.
Sidechain mode (>0.5): aux bus 0 downmixed to mono → LiveAnalysisPipeline on the audio thread, emitting one frame per STFT hop.

Analysis pipeline

Load-bearing constraints (violating these breaks presets, tests, or sound)

Never globally disable the mono harmonic sieve. The sieve runs only when f0.voiced; poly/pass-2 code turns it off conditionally via a neutral voiced=false F0. The monophonic sieved path is load-bearing.
Per-partial bandwidth (0=sine..1=noise) is the historical noise bug. Inflated bandwidth injects noise on every partial. Physics/modulators touch amplitudes only; pass-2 poly sets bandwidth=0 (unreliable without the sieve). Treat any bandwidth write with suspicion.
No frees on the audio thread. Superseded SampleAnalysis goes to pendingDeletion_, freed only in setActive/dtor. currentAnalysis_ is an atomic acquire/release swap.
UI data bridge: sendDisplayData piggybacks a DisplayData block on the DataExchange handler (~30 Hz throttle) with a Tier-3 sharedDisplay_ polling fallback — don't add new IMessage loops.
Param IDs: section bases — 200s = envelope, 400s = harmonic. kPartialCountId=202 is a StringListParameter ("48"/"64"/"80"/"96"), not a range — never swap a registered VST3 param type.
Preset categories are FIXED (7): Brass and Winds, Drums and Perc, Found Sound, Keys, Pads and Drones, Strings, Voice. Never invent new ones.

Deep tooling (read these for real work)

Plugin specifics: plugins/innexus/CLAUDE.md — skeleton, param-ID scheme, innexus_tests target (unit tests/unit/{processor,vst}/ + tests/integration/, run both), pluginval path, AU config.
Frame contract & shared DSP: dsp/include/krate/dsp/processors/harmonic_types.h (HarmonicFrame, Partial, kMaxPartials), residual_types.h, harmonic_oscillator_bank.h.
Analysis internals: plugins/innexus/src/dsp/ (sample_analyzer, live_analysis_pipeline, pre_processing_pipeline, dual_stft_config.h) and shared partial_tracker.h / harmonic_model_builder.h.