AWS Hpc Slurm Ssm Connector
Connect Claude Science to a private-subnet AWS HPC cluster (AWS ParallelCluster, AWS PCS, or any Slurm head/login node) reachable ONLY through AWS Systems Manager (SSM), by reducing the SSM channel to an ordinary SSH connection the native compute provider can open. Works for AWS ParallelCluster and AWS PCS (same design; verified on ParallelCluster) as long as SSM access is enabled to the login/head node. Use whenever the user wants to reach an HPC head/login node "over SSM", "through Session Manager", "in a private subnet", or via an ssh_config ProxyCommand; mentions AWS ParallelCluster, AWS P
Install
npx skills add bioteam/claude-science-hpc-integrations --skill aws-hpc-slurm-ssm-connectorgit clone --depth 1 https://github.com/bioteam/claude-science-hpc-integrations.git /tmp/claude-science-hpc-integrations
mkdir -p ~/.claude/skills
cp -r /tmp/claude-science-hpc-integrations/skills/aws-hpc-slurm-ssm-connector ~/.claude/skills/aws-hpc-slurm-ssm-connectorPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Claude Science → private-subnet AWS HPC over SSM This skill is a field-verified playbook for wiring Claude Science to an HPC cluster — an **AWS ParallelCluster** or **AWS PCS** (Parallel Computing Service) running **Slurm**, or any Slurm head/submit node — whose only ingress is **AWS Systems Manager (SSM)**. There is no public IP and no bastion; you reach the box today only with `aws ssm start-session`. It was built and confirmed end-to-end against a live AWS ParallelCluster (Slurm 23.11.10), so the failure modes and fixes here are the ones that actually bite. **ParallelCluster or PCS — the method is the same.** Both present a Slurm controller on a login/head node inside a private subnet; this connector only cares that (a) the node runs `slurmctld` and (b) SSM access is enabled to it (the SSM Agent is running and the instance role allows Session Manager). Everything below applies to either environment unchanged — PCS-managed clusters have SSM on their login nodes by default. ## The one idea Claude Science's native compute provider speaks **SSH**, not SSM. So the whole integration is: **reduce the SSM channel to an ordinary SSH connection** using an `aws ssm start-session` `ProxyCommand` in `ssh_config`, register the head node as an SSH compute provider, and submit Slurm jobs through it.
Details
- Skill name
aws-hpc-slurm-ssm-connector- Repository
- bioteam/claude-science-hpc-integrations
- Path
skills/aws-hpc-slurm-ssm-connector/SKILL.md- License
- MIT
- Repo updated
- 1 July 2026
FAQ
What does the AWS Hpc Slurm Ssm Connector skill do?
Connect Claude Science to a private-subnet AWS HPC cluster (AWS ParallelCluster, AWS PCS, or any Slurm head/login node) reachable ONLY through AWS Systems Manager (SSM), by reducing the SSM channel to an ordinary SSH connection the native compute provider can open. Works for AWS ParallelCluster and AWS PCS (same design; verified on ParallelCluster) as long as SSM access is enabled to the login/head node. Use whenever the user wants to reach an HPC head/login node "over SSM", "through Session Manager", "in a private subnet", or via an ssh_config ProxyCommand; mentions AWS ParallelCluster, AWS P
How do I install the AWS Hpc Slurm Ssm Connector skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add bioteam/claude-science-hpc-integrations --skill aws-hpc-slurm-ssm-connector. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is AWS Hpc Slurm Ssm Connector popular?
It lives in bioteam/claude-science-hpc-integrations, which has 5 GitHub stars and 1 forks.
Related skills
Schrodinger AWS Hpc Ssm Connector
5bioteam/claude-science-hpc-integrations
Connect Claude Science to a remote Schrödinger Suite jobserver running on an AWS ParallelCluster (or any Slurm HPC) that is reachable only through AWS SSM, and submit/monitor/harvest Schrödinger jobs from it. Use this skill whenever the user mentions Schrödinger, jobserver, jsc, testapp, Maestro, Glide, Desmond, FEP+, or MMLIBS licensing on a remote cluster; whenever they want to reach an HPC head node "over SSM", "through Session Manager", "in a private subnet", or via an ssh_config ProxyCommand; or whenever an SSH compute provider that proxies over SSM fails to connect. It covers reducing SS
Novnc
9Vincent-the-gamer/skills
Deploy noVNC — a browser VNC client — using websockify-js. Use when the user wants to view or control a VNC server in the browser, to set up noVNC + websockify, or needs GBK clipboard support in noVNC.
Octo Operator
10meshmakers/octo-claude-skills
Use this skill for developing and operating the OctoMesh Communication Operator — the .NET 10 KubeOps Kubernetes operator that watches CommunicationPool CRDs, connects to the Communication Controller over SignalR (/operatorHub), and runs helm upgrade/uninstall for Adapter and Application workloads. Covers edge vs central deployment modes, the Helm values layering model (context < base < overrides), secret injection tiers, CRD generation, the TUnit + Microsoft.Testing.Platform test workflow on .NET 10, and the local kind cluster developer loop. Also covers the octo-helm-core chart repository (o
Octo Devtools
10meshmakers/octo-claude-skills
Drives OctoMesh local development through the octo-tools PowerShell cmdlets — building repos in dependency order, starting and stopping services (always non-interactively for agents via Start-Octo -nonInteractive $true / Stop-Octo), managing Docker or local kind Kubernetes infrastructure, syncing and branching git repos, NuGet package propagation, certificates, and octo-cli context/auth setup. Use it for any OctoMesh dev-environment or DevOps task. Trigger on build, compile, feature environment, New-OctoEnvironment, Invoke-BuildEnvironment, lane isolation, dotnet build, start services, stop se
Create Pasqal Cloud Account
9pasqal-io/agent-marketplace
Get access to Pasqal Cloud - create the account on portal.pasqal.cloud and optionally couple it to Google Cloud Marketplace, which is what makes real QPUs and the advanced emulators available. For requests such as "create a Pasqal Cloud account", "I need a PASQAL_PROJECT_ID", "subscribe to Pasqal on GCP Marketplace", "how do I run on FRESNEL", or "my project only shows EMU_FREE". Access only, not running experiments. Unsure which route fits? Use choose-access-method.
Octo Deploy
10meshmakers/octo-claude-skills
Guide the promotion of an OctoMesh-powered app from local dev to the shared test-2 environment — publish the CK model and blueprint to the shared GitHub catalogs (octo-ckc / octo-bpm), push the app image to <registry>, register a test-2 octo-cli context, install + deploy on the target tenant, and verify everything WITHOUT kubectl access. Covers both the repeatable CI lane (azure-pipelines publishing on main) and the manual first-time lane, plus test-2-specific triage (server-side catalog cache refresh, ImagePullBackOff via lastDeploymentError, Tailscale reachability). Staging and production ar
Browse the most popular Claude skills, all devops & cloud skills, or write your own with the skills generator.