TorchedHat

TorchedHat/torchtalk

Plugin Claude Code

MCP server giving Claude Code cross-language understanding of PyTorch (Python -> C++ -> CUDA): binding-chain tracing, impact analysis, dispatch mapping, call graphs, and test discovery.

TorchedHat/torchtalk

Command Claude Code

Trace a PyTorch function's cross-language binding chain.

TorchedHat/torchtalk

Skill Claude CodeCodex

Analyze PyTorch internals across Python, C++, and CUDA layers. Use when asked about how PyTorch operators work internally, where functions are implemented (CPU/CUDA backends), what would break if code is modified (impact analysis), how torch.nn modules connect to native code, or finding tests for PyTorch operators.…

TorchedHat/torchtalk

MCP server Claude CodeCodexCursor +2

MCP server "torchtalk" as configured in TorchedHat/torchtalk. Launched with torchtalk mcp-serve --harness pytorch. Needs 1 environment variable to run.

TorchedHat/torchtalk

Instructions file

Instructions for TorchedHat/torchtalk, covering claude.md, commands, install, test and lint.

TorchedHat/torchtalk

Hook

Runs when a session starts, executing plugin-setup.sh. From TorchedHat/torchtalk.

TorchedHat/torch-air

Plugin Claude Code

Torch Accelerator Integration Readiness — evaluate how well a hardware accelerator integrates with PyTorch.

TorchedHat/torch-air

Plugin Claude Code

Torch Accelerator Integration Readiness — evaluate how well a hardware accelerator integrates with PyTorch.

TorchedHat/torch-air

Skill Claude CodeCodex

Evaluate a hardware accelerator's integration readiness with PyTorch. Use when checking if an accelerator (XPU, NPU, openreg, HPU, custom) supports PyTorch's PrivateUse1/fork integration (device management, hooks, operators, AMP, autograd, torch.compile, distributed, profiler, serialization). Accepts a backend name or…

TorchedHat/torch-air

Skill Claude CodeCodex

../../SKILL.md.