Skill Claude CodeCodex
A security-focused code review for Python, C++, Shell, and Markdown files. It checks for common vulnerabilities using guidance such as OWASP Top 10, CWE Top 25, and CERT secure-coding standards.
Skill Claude CodeCodex
A migration tool for simple vector Triton operators, moving them from GPUs to Huawei Ascend NPU hardware. It supports Triton code that can be compiled successfully and cannot automatically handle operators with compilation problems.
Skill Claude CodeCodex
Comprehensive security auditor for AI agent skills, prompts, and instructions. Checks for typosquatting, dangerous permissions, prompt injection, supply chain risks, and data exfiltration patterns — before you use any agent or skill.
Skill Claude CodeCodex
A setup tool for SwanLab, an experiment-tracking platform used to record and monitor machine-learning runs. It manages login credentials and SwanLab connections, including setups inside containers.
Skill Claude CodeCodex
A migration guide for moving TileLang operators from GPUs using CUDA to Huawei Ascend NPU hardware. It examines the GPU implementation and generates a corresponding NPU implementation.
Skill Claude CodeCodex
A code generator that turns an Ascend NPU Triton operator design, or a direct requirement, into Triton kernel code. A kernel is the device-side code that performs the operator's computation, and the output includes basic correctness tests.
Skill Claude CodeCodex
A static code review for Triton operators running on Huawei Ascend NPU hardware, covering both host and device code. It looks for likely bugs, incorrect API use, and performance risks without running the code.
Skill Claude CodeCodex
A design guide for new Triton operators on Huawei Ascend NPU hardware. It turns a requirement into a document describing behavior, interfaces, data tiling, and the planned kernel implementation.
Skill Claude CodeCodex
A guided workflow for developing Triton operators on Ascend NPUs. Triton operators are small GPU- or accelerator-focused programs used to implement fast tensor calculations.
Skill Claude CodeCodex
A documentation generator for Triton operators on Ascend NPUs. It creates interface documentation such as README files, API descriptions, parameter explanations, and support tables.
Skill Claude CodeCodex
An environment setup and validation tool for developing Triton operators on Ascend NPUs. It checks dependencies such as CANN, Python, PyTorch, torch-npu, and Triton for Ascend.
Skill Claude CodeCodex
A performance evaluation tool for Triton operators running on Ascend NPUs. It uses profiling data to show whether an operator is limited by memory movement or by computation.
Skill Claude CodeCodex
A performance optimization guide for Triton operators designed for Ascend NPUs. It addresses issues such as local-buffer overflow, low Cube-unit usage, and inefficient tiling, which divides work into accelerator-sized pieces.
Skill Claude CodeCodex
A precision checker for Triton operators. It compares an operator's results with a PyTorch reference implementation and calculates error measurements.
Skill Claude CodeCodex
An automated workflow for analyzing closed issues in the vLLM-ascend project, a tool for serving language models on Ascend hardware. It turns issue patterns into debugging frequently asked questions.
Skill Claude CodeCodex
A performance optimization guide for vector Triton operators on Ascend NPUs. It covers accelerator-specific techniques such as managing fast local memory, processing multiple tokens, and overlapping data movement with computation.
Skill Claude CodeCodex
A configuration generator for single asynchronous DAPO training with Verl, a toolkit for training language models. It creates training scripts, checks settings before training, and connects monitoring through SwanLab.
Skill Claude CodeCodex
A deployment and configuration tool for Verl distributed training services. It can launch training containers and Ray clusters, which distribute work across machines, for RLHF or DAPO language-model training.
Skill Claude CodeCodex
A deployment workflow for serving large language models on Ascend NPU hardware. NPU means a processor designed to run AI models; the workflow checks the machine and configuration before deployment, then verifies the service.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: