scragnog
Community@scragnog
scragnog maintains deep technical skills for HOT-Step CPP, a local AI music generation engine covering MiniMax-Music3 backends, LoRA adapter training, and C++/Node/React development.
Agent Skills by scragnog
Showing 17 vetted skills indexed across 1 GitHub repositories.
mm3-backend
Maps HOT-Step's native MiniMax-Music3 C++/GGML backend modules, endpoints, and parity infrastructure.
server-feature-dev
Guides adding Express routes, services, and SQLite schema changes to the HOT-Step Node server.
lua-plugin-authoring
Write, test, and debug Lua 5.4 sampling plugins for the HOT-Step CPP audio engine.
mm3-lm-adapter-training
Trains MiniMax-Music3 planner-LM LoRA adapters for album-style music cloning.
debugging-runtime
Diagnoses HOT-Step CPP generation failures and engine crashes from session log folders.
mm3-captioning
Formats Structured Captions for MiniMax-Music3 music generation prompts.
flash-attn-training
Ports fused flash-attention training ops to GGML-based model trainers with parity and VRAM gates.
upstream-sync
Merges upstream acestep.cpp changes into the HOT-Step engine fork while preserving integration hooks.
project-map
Maps HOT-Step CPP features to route files, services, UI folders, and engine subsystems.
validating-changes
Validates code changes across TypeScript, C++ engine, UI, and audio tiers with smoke tests.
engine-dev-loop
Rebuild and smoke-test the HOT-Step C++ engine without triggering respawn or file-lock failures.
ui-feature-dev
Guides adding React studios, controls, and API wiring to the HOT-Step CPP UI.
engine-performance
Diagnose and benchmark HOT-Step generation performance across GGML and TensorRT inference backends.
adapter-system
Documents LoRA/LoKr adapter loading, merging, stacking, and per-section masking in the HOT-Step engine.
release-process
Cut and publish HOT-Step CPP releases via git tags triggering multi-platform CI builds.
generation-request-flow
Traces music generation requests through UI, Node queue, and C++ engine to locate dropped parameters.
model-management
Manages GGUF model files, quantization, and the Model Manager download service in HOT-Step CPP.
Frequently Asked Questions About scragnog
FAQPage SchemaWhat tasks can I accomplish with scragnog's HOT-Step CPP skills?▼
The 17 skills cover the full HOT-Step CPP music generation stack: MiniMax-Music3 backend integration and structured captioning, LoRA/LoKr adapter training and runtime stacking, Lua sampler/scheduler plugin authoring, Express/SQLite server features, React studio UI development, TensorRT performance tuning, GGUF model management, and release publishing.
Who are these skills designed for?▼
They target engineers maintaining or extending the HOT-Step CPP fork of acestep.cpp — C++ engine developers, TypeScript/Node server developers, React UI builders, and ML practitioners training MM3 LM adapters or profiling DiT/LM/VAE generation performance.
How do I debug a failed or stalled music generation?▼
Use the debugging-runtime skill to inspect logs/ session folders for engine crashes, 500/503 responses, and hangs, then apply generation-request-flow to trace the request from UI form through the Node job queue, aceClient, engine LM/synth stages, and the SQLite song row, catching silently dropped parameters.
What prerequisites and components does the HOT-Step stack require?▼
The stack spans a C++ engine (engine/src, ggml CUDA flash-attention ops), a Node/TypeScript Express server with better-sqlite3, a React/Vite/Tailwind UI, Lua plugins for solvers and schedulers, GGUF-quantized model checkpoints, and optional TensorRT paths (dit-trt, lm-trt) for accelerated inference.
How are releases and upstream changes managed?▼
The release-process skill cuts releases via v* git tags triggering multi-platform CI builds and drafted GitHub Releases, while upstream-sync safely merges acestep.cpp changes into the fork, repairing integration hooks and diagnosing dead solvers or unresolved hotstep_sampler_linked_ symbols after a sync.