A specialized automation skill for deploying the 5-package FlagOS multi-chip AI software stack inside running GPU containers with full validation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install install-stack-flagos
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install install-stack-flagos using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Install Stack FlagOS?
This skill automates the configuration and installation of a 5-package multi-chip software stack inside a running GPU-enabled Docker container. It facilitates compiling and setting up key AI packages including vLLM, FlagTree, FlagGems, FlagCX, and vllm-plugin-FL. Using automated network mirror detection, precise environment collection, and strict dependency mapping, this skill ensures deep integration with multiple hardware architectures.
Optimized for heterogeneous AI infrastructure, this package is one of the essential Openclaw Skills designed to streamline operations across distinct GPU vendors, ensuring developers can move quickly from raw container initialization to verified inference-ready setups without manually troubleshooting dependency conflicts.
Install Stack FlagOS Use Cases
- Multi-Chip AI Deployment: Setting up GPU containers to run LLM workloads across varying non-NVIDIA/specialized AI hardware.
- Continuous Integration (CI) Automation: Integrating automated environment building pipelines using standardized Openclaw Skills.
- Network-Isolated Setups: Instantiating software stacks safely inside air-gapped or restricted network environments where mirrors are required for PyPI or GitHub.
- Inference Stack Standardization: Setting up reproducible, validation-verified inference stacks for production-scale model serving with vLLM.
How Install Stack FlagOS Works
- Resolve Container & Vendor: Verifies the Docker container is active and invokes script files internally to retrieve OS, architecture, and Python context.
- Network Environment Profiling: Dynamically scans for accessible PyPI and GitHub mirrors to bypass regional network constraints.
- Prerequisite Check: Confirms hardware specifications and validates disk space availability (minimum 10GB free).
- Ordered Dependency Installation: Executes pip operations, git checkouts, and system compilations in a strict, safe order (vLLM -> FlagTree -> FlagGems -> FlagCX -> vllm-plugin-FL).
- Post-Installation Package Verification: Imports and stress-tests all components to confirm precise version matches and successful C++ API bindings.
- Path Persistence & Reporting: Sets persistent container environment variables and outputs a structured execution report detailing step-level status.
Install Stack FlagOS Setup
Ensure you have a running Docker container provisioned with PyTorch and GPU access. Obtain the active container name and target GPU vendor.
To run this execution sequence through the Openclaw interface:
# Standard manual invocation
openclaw run install-stack-flagos --container <CONTAINER_NAME> --vendor <GPU_VENDOR>
The automation automatically injects utility scripts (detect_network.py, collect_env_info.py, select_flagtree_wheel.py, validate_packages.py) into the running container, handles dependencies sequentially, and applies appropriate make flags for FlagCX based on the detected hardware profile.
Install Stack FlagOS Data Schema & Taxonomy
Component & Data Architecture
The skill relies on a robust schema of validation scripts, environment mapping templates, and references:
| Path | Purpose | Output Format |
|---|---|---|
scripts/detect_network.py |
Probes network access to PyPI/GitHub | JSON mirror configurations |
scripts/collect_env_info.py |
Collects Python version, glibc version, free disk | JSON host system parameters |
scripts/select_flagtree_wheel.py |
Maps vendor/python/glibc to the correct precompiled Triton wheel | Specified Wheel URL string |
scripts/validate_packages.py |
Runs import-tests across all installed AI modules | Comprehensive status JSON |
references/vendor-mappings.md |
Contains compilation flags and build matrix | Reference Documentation |
references/network-mirrors.md |
Hosts fallback network mirrors rules | Reference Documentation |
Execution Telemetry Schema
Upon execution, a final structured JSON report is generated:
{
"status": "PASS | PARTIAL | FAIL",
"stage": "install-stack",
"container": "<name>",
"vendor": "<vendor>",
"network": {"github_mirror": true, "pypi_mirror": true},
"python_version": "3.11",
"glibc_version": "2.34",
"packages": {
"vllm": {"status": "PASS", "version": "0.13.0"},
"flagtree": {"status": "PASS", "version": "0.4.1+ascend3.2"},
"flaggems": {"status": "PASS", "version": "..."},
"flagcx": {"status": "PASS", "version": "0.10.0"},
"vllm_plugin_fl": {"status": "PASS", "version": "..."}
},
"flagcx_path": "/tmp/FlagCX",
"gate_passed": true,
"errors": []
}
Install Stack FlagOS Advanced Features
- Network-Aware Routing: Auto-detects fast local mirrors (such as regional PyPI and GitHub mirrors) to bypass latency or network access limits.
- Safe Failure Cascades (Gates): Incorporates strict gate rules where failing core modules (like vLLM) halts execution immediately, while minor issues (like FlagGems) only trigger warnings.
- Dynamic Make Flag Selection: Determines optimized compiler flags dynamically for FlagCX compilation based on hardware vendor mappings.
- Post-Install Verification Framework: Leverages an end-to-end modular testing engine inside the container to verify actual CUDA/ROCm dynamic compilation before closing the session.
SKILL.md
Loading
Related Openclaw Skills
Patent & Software Copyright Generation (AI + Big Data)
An AI agent skill designed to generate CNIPA-compliant patent documents and CPCC software copyright materials directly from codebases and technical documentation.
3DGS Engineering Guide
A comprehensive developer-focused guide and workflow assistant for deploying 3D Gaussian Splatting (3DGS) pipelines from academic research to production.
3DGS Visualizer
A publication-quality visualization tool for 3D Gaussian Splatting (3DGS) researchers to generate static and interactive radar charts, comparison tables, and evolution timelines.
License Mismatch Audit
A benign security-audit and validation skill designed to test if marketplaces and clients accept frontmatter-only license metadata without bundled LICENSE files.
FlagRelease Pipeline Orchestrator
An automated orchestration pipeline for verifying, deploying, and benchmarking Large Language Models (LLMs) across multi-chip GPU backend environments.
model-citizen
An accountability and security framework for AI agents, providing verifiable Ed25519 identities, scoped delegation, and cryptographic receipts.
Featured*
