Plugin Claude Code
Agent-native GPU training control plane — bundles the gpuctl MCP server (submit/monitor/compare/recommend) with run-experiment, monitor-experiment, and analyze-results skills.
Skill Claude CodeCodex
Compare GPU training runs/checkpoints and pick the best one to keep or deploy. Use when the user asks which run/checkpoint won, to compare experiments, pick the best model, or decide what to evaluate further.
Skill Claude CodeCodex
Watch a running GPU training run for progress and anomalies, and react when it fails. Use when the user asks to monitor/watch/keep an eye on a run, or when a run may have hit OOM, loss-NaN, overfitting, or a crash.
Skill Claude CodeCodex
Submit and start a GPU training run on GPUPlane. Use when the user asks to run/train/launch an experiment, kick off a training job, or start a model run on the GPU cluster.
Instructions file CodexOpenCode
AGENTS.md instructions for EricYuan2007/GPUPlane, covering agents.md — gpuplane agent guide, what this project is (and isn't), the experiment loop, conventions and metrics.
Instructions file
Claude Code instructions for EricYuan2007/GPUPlane, covering what this project is (and isn't), the experiment loop, conventions, metrics and jobs.
MCP server Claude CodeCodexCursor
MCP server "gpuplane" as configured in EricYuan2007/GPUPlane. Runs locally from the gpuplane Python package.