opensite-ai/opensite-skills / linux-ai-dev-stack
linux-ai-dev-stack
FlaggedInstall and wire the AI toolchain on a Linux dev machine: local LLM inference (Ollama, llama.cpp, vLLM, SGLang, TabbyAPI) with VRAM-based model selection, AI coding agents and CLIs (Claude Code, Codex, aider, OpenCode) with MCP servers, ML frameworks (PyTorch/JAX, Jupyter, HuggingFace) and GPU containers, and cloud/remote GPU workflows (RunPod, Vast.ai, Lambda, remote dev over SSH and Tailscale). Use when someone wants to run local models, set up or connect AI coding agents, configure MCP servers, route agents to local or cloud models, install an ML/GPU Python stack, or offload work to a rented GPU. It runs on the machine and assumes working GPU drivers.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.9 KB |
| agents/openai.yaml | config | 420 B |
| examples/24gb-nvidia-local-plus-cloud.md | markdown | 5.0 KB |
| references/activation.md | markdown | 2.9 KB |
| references/ai-coding-agents.md | markdown | 7.1 KB |
| references/local-llm-inference.md | markdown | 11.9 KB |
| references/mcp-servers.md | markdown | 4.5 KB |
| references/ml-frameworks.md | markdown | 5.0 KB |
| references/model-wiring.md | markdown | 4.6 KB |
| references/remote-gpu.md | markdown | 4.1 KB |
| templates/AGENTS.md | markdown | 1.8 KB |
| templates/gpu-compose.yaml | config | 1.4 KB |
| templates/litellm-config.yaml | config | 2.3 KB |
references/activation.md
View on GitHub# Activation Guide — linux-ai-dev-stack
## Best-fit tasks
Use this skill when an agent is on a Linux machine (with working GPU drivers) and
the user wants to:
- Run local LLMs — install Ollama / llama.cpp / vLLM / SGLang / TabbyAPI / LM Studio,
pick a model that fits VRAM, expose it safely on the LAN, or tune throughput.
- Install or connect AI coding agents — Claude Code, Codex CLI, aider, OpenCode,
Gemini CLI — including auth, config files, sandboxing, and git worktrees.
- Configure MCP servers (filesystem, git, github, context7, playwright, memory,
fetch, sequential-thinking) with correct scopes and least-privilege security.
- Wire agents to both cloud and local models, including a LiteLLM proxy and
cost/model routing (triage → build → review).
- Install an ML/GPU Python stack — uv/pixi env, PyTorch/JAX/TensorFlow with
CUDA/ROCm/XPU, Jupyter, the HuggingFace stack, and GPU containers.
- Offload work to a rented GPU (RunPod, Vast.ai, Lambda, Modal) with remote dev,
code/data sync, and private networking.
## Explicit invocation
- "Use `$linux-ai-dev-stack` to set up a local model server and point Claude Code at it."
- "Use `$linux-ai-dev-stack` to install PyTorch with CUDA and run it in a GPU container."
- "Use `$linux-ai-dev-stack` to add the essential MCP servers to my project safely."
- "Use `$linux-ai-dev-stack` to rent a RunPod GPU and iterate on it over Tailscale."
## When NOT to use (hand off to a sibling)
- **No GPU drivers / `nvidia-smi` fails** → use `linux-dev-workstation` first to
install drivers and the compute stack. This skill assumes they already work.
- **Base dev environment not set up** (shell, editors, language toolchains,
containers, security) → use `linux-dev-workstation`. GPU container passthrough
(NVIDIA Container Toolkit / rootless / CDI) is shared with that skill.
- **Choosing a distro or checking hardware compatibility** → use
`linux-distro-selector`.
- **Installing Linux on the machine** (bootable media, partitioning, first boot)
→ use `linux-install-planner`.
## Cross-agent notes
- **Modes / permissions differ per agent.** Claude Code uses OS-level sandboxed
Bash + allow/ask/deny rules; Codex uses landlock/seccomp with an approval ×
sandbox matrix. Respect the host agent's confirmation flow — never auto-run
destructive commands (cache deletion, unit replacement, firewall/port changes).
- **This skill runs on the machine.** It is a configurator, not a planner: it
executes and verifies commands rather than emitting a runbook for a human. The
two planner siblings (`linux-distro-selector`, `linux-install-planner`) do the
opposite.
- **Currency:** always verify engine/model/package versions with a web search at
runtime; embedded numbers are labeled "as of 2026-07 — verify."
- **Portability:** `AGENTS.md` is the cross-tool convention; symlink `CLAUDE.md` →
`AGENTS.md` so every agent reads one source of truth.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 469k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |