hyperpod-nccl

A read-only troubleshooting guide and diagnostic script for NCCL failures on SageMaker HyperPod GPU clusters. NCCL is software that lets GPUs communicate during distributed training, and HyperPod is AWS infrastructure for running such training across GPU machines.

In plain words
What is it for?
Use it to investigate training hangs, collective-operation timeouts, EFA or libfabric errors, startup coordination failures, shared-memory or locked-memory limits, NCCL version differences, and out-of-memory exits on EKS or Slurm clusters.
Why use it?
It helps separate communication failures from related cluster, networking, memory, container, and startup problems without changing the cluster or training job. It points from detected issues to relevant explanations.

Skill for Claude CodeCodex

Part of the sagemaker-ai plugin — 19 skills shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/awslabs/agent-plugins/hyperpod-nccl
Any agent
npx skills add awslabs/agent-plugins --skill hyperpod-nccl
Clone the repo
git clone --depth 1 https://github.com/awslabs/agent-plugins

Made for: Claude Code, Codex.

Or install sagemaker-ai, the plugin that ships this one along with the rest of its 19 skills.

Per session 150 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 3,035 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00150 $0.03035
Opus 5 $0.00075 $0.01517
Sonnet 5 $0.00030 $0.00607
Haiku 4.5 $0.00015 $0.00303

Measured 3d ago against content hash 5dea8bd1f3ec, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

hyperpod-nccl scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

The scan reads SKILL.md. This mod also ships 1 executable file (scripts/nccl-diagnose.sh), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

plugins/sagemaker-ai/skills/hyperpod-nccl/SKILL.md · 188 lines

How it starts

The opening of the file, as written. The whole thing — 188 lines — stays where its author put it; the contents beside it link to each section on GitHub.

HyperPod NCCL Debugger

Operating policy. Run read-only diagnostics yourself. Never run a command that changes cluster, node, or workload state — present each one as a Suggested command (run this yourself) block and wait for the customer. Destructive order: investigate → reboot → replace (replace destroys root + secondary volumes; not supported on Slurm controller nodes). Never discard training state on speculation.

Diagnose NCCL failures on SageMaker HyperPod (EKS and Slurm). scripts/nccl-diagnose.sh reads state via AWS APIs, kubectl, and SSM, then prints each issue as [FAIL] ... → references/<file>.md § <section>. Read-only.

Signal sourcing: list-cluster-events carries infrastructure-level state only (lifecycle, bootstrap, EFA health check, capacity, replacement, reboot, AMI rollback). It does not carry NCCL timeouts, GPU XID/ECC, or per-pod training signals — those come from pod logs, CloudWatch training streams, on-node SSM probes, and NCCL env audit. "No events" on a training-time NCCL issue is expected, not a clean bill of health.


Workflow

  1. Collect cluster name, region, namespace/job (EKS), exact NCCL error string.
  2. Run the diagnostic (always — the output drives everything else).
  3. For every [FAIL] line, Read the referenced section.
  4. Present finding, root cause, and the Suggested-command block with concrete values (instance IDs, SG IDs, namespaces) filled in from the script output. Wait for customer approval.
  5. Re-run the diagnostic to confirm.

If a finding has no matching section, report it as a bug — do not invent a fix.

Step 1: Authenticate kubectl (EKS)

EKS_ARN=$(aws sagemaker describe-cluster --cluster-name <HYPERPOD-NAME> --region <REGION> \
  --query 'Orchestrator.Eks.ClusterArn' --output text)
EKS_NAME=$(echo "$EKS_ARN" | awk -F'/' '{print $NF}')
aws eks update-kubeconfig --name "$EKS_NAME" --region <REGION>
kubectl get nodes

Step 2: Run the diagnostic

# Basic:
bash scripts/nccl-diagnose.sh --cluster <HYPERPOD-NAME> --region <REGION>

# Scope to an EKS job/namespace:
bash scripts/nccl-diagnose.sh --cluster <NAME> --region <REGION> --namespace <NS> --job <JOB>

# Force orchestrator:
bash scripts/nccl-diagnose.sh --cluster <NAME> --region <REGION> --orchestrator slurm

# Larger hardware sample (default 3):
bash scripts/nccl-diagnose.sh --cluster <NAME> --region <REGION> --sample-nodes 10

# Specific node only:
bash scripts/nccl-diagnose.sh --cluster <NAME> --region <REGION> --node i-0abc123def456

Read the full file on GitHub · 188 lines

Files

What ships with it

5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 188 lines · 150 tokens per session scan A 5dea8bd1f3ec

Subscribe to this mod's changes

hyperpod-nccl is a skill published in the GitHub repository awslabs/agent-plugins (876 stars, last pushed 6d ago), licensed Apache-2.0. It adds 150 tokens to every session and 3,035 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

analytics-opensearch-expertise

Amazon OpenSearch Service domain health assessment. Performs read-only, API-driven checks against a customer's OpenSearch domain(s) covering cluster health, node/shard configuration, performance metrics, security posture, and cost optimization signals. Activate this skill for requests about OpenSearch or Elasticsearch…

aws/tools-for-devops-agent · 136 tokens

msk-operations

Amazon MSK Provisioned operations, troubleshooting, and health assessment for Standard and Express brokers. Use whenever the user mentions Amazon MSK, MSK Provisioned, MSK Standard/Express brokers, Apache Kafka on AWS, kafka. / express. instance types, or the AWS/Kafka CloudWatch namespace. Covers MSK performance…

aws/tools-for-devops-agent · 226 tokens

rds-operation-review

Comprehensive Amazon RDS and Aurora operational review aligned with the AWS Well-Architected Framework and RDS/Aurora best practices. Use this skill when a user asks to review, audit, or assess RDS instances or Aurora clusters for best practices compliance, security posture, reliability, performance, cost…

aws/tools-for-devops-agent · 118 tokens

aws-health-events

ALWAYS use this skill in the beginning of any incident investigation, root cause analysis, or operational troubleshooting. This skill retrieves and analyzes AWS Health events (service issues, scheduled changes, and account notifications) to identify AWS-side events that may explain or correlate with observed…

aws/tools-for-devops-agent · 159 tokens

database-rds-devops

Database-level data-plane diagnostics for Aurora MySQL and Aurora PostgreSQL. Executes predefined read-only health check queries via RDS Data API to analyze buffer pool, connections, locks, replication, storage, performance, and index efficiency. Requires the rds-aidba MCP server for database-internal access beyond…

aws/tools-for-devops-agent · 75 tokens

eks-operation-review

Comprehensive Amazon EKS operational review aligned with the AWS EKS Best Practices Guide. Use this skill when a user asks to review, audit, or assess EKS clusters for best practices compliance, operational readiness, security posture, cost optimization, reliability, networking, scalability, or upgrade readiness.…

aws/tools-for-devops-agent · 99 tokens