Operate an RTX 3090 Thunderbolt eGPU and the local models served on it: status, safe drain and power-off, power-up, and model load checks.
1 on GitHubby mairpApache-2.0updated 8 days agodevops
/
Plugin gpu-ops from marketplace mairp in mairp/claude-plugins. Commands read from the repo's manifest — 8 days ago.
What it installs
Skills (2)
gpu-ops
Operate and inspect the RTX 3090 eGPU (Thunderbolt-4 enclosure) on the mairp host via the proven scripts in /root/gpu_rtx_3090. Use when asked to "check GPU / eGPU status", "how's the GPU", "free VRAM", "drain the GPU", "safe shutdown / power the GPU down", "detach the eGPU", "power the GPU back up"
local-model-ops
Hardware-aware guidance for running local Qwen 3.8, Muse, and Nemotron models on the RTX 3090 eGPU (llama-swap on this host) - which model to pick, when to think, when to escalate to gpt-5/claude, and what the serving limits are. Use when asked "which local model should I use", "run this on qwen", "
Fast, free, first-pass security smell test. Point a tiny local LLM (antares-1b, a ~2GB Granite-4.0 fine-tune) at a folder to get cheap vulnerability leads across a codebase in seconds — before a deeper audit. A triage aid, NOT a substitute for Semgrep, CodeQL, or Bandit.
Derive a provenance-bound launch contract and an unattended Specstride pipeline from GitHub Spec Kit artifacts. Sources are pinned by SHA-256; editing a spec after derivation blocks the launch until the contract is regenerated.
Development governance for coding agents: takes a Spec Kit feature to done, phase by phase, with the tests run by the engine and an independent reviewer that must approve each phase before it is signed and committed.
Run a Spec Kit command (analyze, clarify, plan, ...) over a range of specs on a chosen harness, with the model and reasoning effort pinned per run, sequentially or several features in parallel.
Transform Claude into an Azure expert. This plugin integrates the Azure MCP server and specialized Azure skills to move beyond generic advice. It enables Claude to perform real-world tasks: listing resources, validating deployments, diagnosing infrastructure issues, and optimizing costs across 50+ Azure services.
Work with incident.io from Claude: respond to and investigate incidents, check on-call schedules and escalations, and author the runbooks, skills and architecture docs that incident.io investigations draw on. Bundles the official incident.io MCP server (OAuth on first use).
Production LLM inference including vLLM, TensorRT-LLM, llama.cpp, and SGLang. Use when deploying models for production inference.
98 skills
Orchestra-Research/AI-Research-SKILLs 14,502
Maintain gpu-ops?
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.