oh-my-opencode 风格的多代理编排系统。提供 7 个专业代理协同工作：Sisyphus（主编排）、Oracle（架构顾问）、Librarian（文档研究）、Explore（代码搜索）、Frontend Engineer（UI/UX）、Document Writer（文档撰写）、Multimodal Looker（多模态分析）。当需要复杂开发任务、多角度问题解决、或特定专业领域帮助时使用。
KINTSUGI processing principles: Never sacrifice quality for speed, always use GPU when available. Trigger: performance optimization, CPU/GPU choice, fast mode, quality vs speed.
Parsing binary files with mixed endianness and variable-length records in Python.
Zero-copy deserialization, async I/O patterns, lifetime management, memory-efficient parsing, and safe handling of unsafe code for SSTable parsing. Use when working with performance optimization, memory efficiency, async/await, borrowing/lifetimes, zero-copy patterns, or memory usage under 128MB target.
Your approach to handling ffi bindings. Use this skill when working on files where ffi bindings comes into play.
Core PyTorch fundamentals including tensor operations, autograd, nn.Module architecture, and training loop orchestration. Covers optimizations like pin_memory and lazy module initialization. (pytorch, tensor, autograd, nn.Module, optimizer, training loop, state_dict, pin_memory, lazylinear, requires_grad)
Expert in RTOS, bare-metal programming, and embedded Linux. Specializes in Rust for Embedded and high-reliability firmware.
Fixes function call errors with wrong number or type of arguments.
Your approach to handling global free threading. Use this skill when working on files where global free threading comes into play.
Write high-performance, computational Rust code that parallelizes well and allocates memory efficiently. Use when writing performance-critical Rust code, optimizing existing Rust code for speed or memory, implementing parallel/concurrent algorithms, working with SIMD, or when the user mentions performance, optimization, allocation, or parallelism in a Rust context. Covers build configuration, profiling, memory allocation patterns, parallelism with rayon/crossbeam, and micro-optimization techniques.
Use when building Rust applications requiring memory safety, systems programming, or zero-cost abstractions. Invoke for ownership patterns, lifetimes, traits, async/await with tokio. Keywords: Rust, ownership, borrowing, async, memory safety.
Rust specialist with expertise in async programming, ownership patterns, FFI, and WebAssembly development
PyTorch CUDA environment and performance guidance, with emphasis on CUDA 13 toolkit/driver requirements, PyTorch wheel compatibility, and runtime checks. Use when configuring PyTorch on NVIDIA GPUs, debugging CUDA setup, or migrating to CUDA 13; triggers: pytorch cuda, cuda 13, driver version, nvcc, torch.version.cuda, tf32, streams.
Scaffold a new model provider for obot-tools
Refactor PyTorch code to improve maintainability, readability, and adherence to best practices. Identifies and fixes DRY violations, long functions, deep nesting, SRP violations, and opportunities for modular components. Applies PyTorch 2.x patterns including torch.compile optimization, Automatic Mixed Precision (AMP), optimized DataLoader configuration, modular nn.Module design, gradient checkpointing, CUDA memory management, PyTorch Lightning integration, custom Dataset classes, model factory patterns, weight initialization, and reproducibility patterns.
Generic extension methods for strings, collections, DateTime, decimals, and entities.
Development skill for building a Raspberry Pi Pico-based RC submarine controller with ballast control, depth hold, pitch stabilization, and failsafe protection. Use when working on embedded C firmware for RP2040, implementing PID controllers, writing sensor drivers (pressure, IMU), PWM capture/output, dual-core safety-critical systems, or any task related to the RC submarine project. Also triggers for Power of 10 coding standards, Pico SDK usage, and underwater vehicle control systems.
Implement or modify a Rust function exposed to Python via PyO3 in eo-processor. Use when adding new compute kernels, fixing numerical/shape bugs in the Rust core, or wiring Rust functions into the Python module safely and consistently.
Activates during long sessions or before context compaction. Manages attention markers and ensures critical information survives memory compression.
