AWS Storage

Overview

This skill provides domain expertise for choosing among AWS storage services, selecting storage classes, optimizing cost, and routing to resources for operating storage services. It covers object storage (S3 General Purpose buckets and their storage classes, S3 Express One Zone on directory buckets, S3 Tables, S3 Vectors), file storage (Amazon EFS, S3 Files, FSx for Lustre, FSx for NetApp ONTAP, FSx for OpenZFS, and FSx for Windows File Server), block storage (EBS volume types and EC2 instance store), and the data-movement and protection services that connect them (DataSync, Storage Gateway, Transfer Family, and AWS Backup). It does not advise on databases or analytics query engines. It works with or without the AWS MCP server; when available, the AWS MCP server (opens in a new tab) is recommended for verifying current specifications and pricing, and all guidance also works with the standard AWS CLI. For deep single-service tasks, route to the specialized skills listed in the Routing section below.

How to Handle User Queries

When this skill is triggered, classify the user’s request and follow the appropriate path.

Rules

These apply to all responses regardless of path:

1. You MUST verify current numbers. When the AWS MCP server is available, use search_documentation and read_documentation to cross-check before citing specifics. When quoting costs, you MUST include a link to the relevant pricing page. When quoting performance metrics, you MUST include a link to the relevant product page. Otherwise, verify against linked AWS documentation pages or use the AWS CLI to confirm current values. Where a reference file directs you to documentation for a current value, you MUST retrieve that value from the linked page before answering. Do not substitute a remembered figure, and do not offer an approximation or a range in place of a retrieved value. If retrieval is not possible in the current environment, name the value you could not verify rather than citing one from memory.
2. You MUST retrieve the relevant service reference file from the Routing section below before answering questions about that service. AWS storage specifications, limits, and service capabilities change frequently. You MUST NOT answer from memory alone. You MUST surface relevant troubleshooting guidance and ‘gotchas’ from reference files in your response. Justify recommendations by workload fit, not by mentioning that a reference file ‘explicitly’ mentions a workload for a given service.
3. You MUST include cost implications when recommending services or approaches. Do not wait for the user to ask. Do not compare services on storage charges alone; per-object fees such as metadata charges can materially change TCO. For deep cost analysis, monitoring, or optimization beyond storage selection, route to the billing-and-cost-management (opens in a new tab) skill.
4. You MUST have clarity on the user’s need when making a recommendation. Match the specificity of your response to the specificity of the request. When the query determines the storage category and the relevant services, retrieve information and recommend directly. When the query is not fully specified, YOU MUST mention the assumptions and limitations of your recommendation and include the additional questions that would confirm or change it. Ask follow up questions in place of a recommendation only when the query does not let you determine the storage category at all. Recommend the best-fit service for the workload even when it falls outside this skill’s scope; add relevant in-scope options as alternatives.

---

Step 1: Classify Intent

Determine what the user needs:

You MUST follow the interaction logic in the corresponding path instructions below.

Ambiguous cases: If the user’s primary ask is a recommendation, it is SELECT regardless of context. If they need help executing a known plan, it is INVESTIGATE.

---

Step 2a: SELECT Path

The user needs help choosing. You MUST use the following decision factors to inform your recommendation, asking questions to fill gaps that would change the choice.

Decision Factors:

Use the Storage Options table to identify candidate services, not to cut services; keyword matches against the Common Workloads column are not the only answers. You MUST retrieve the reference files for each of the candidate services using the Routing section below.

Considering these factors, recommend specific AWS storage service(s) and:

1. You MUST include clear rationale tied to the user’s stated requirements.
2. You MUST present alternatives where the choice is close or dependent on unspecified information, explaining the tradeoff.

---

Step 2b: INVESTIGATE Path

The user knows their service or approach and needs operational help.

1. Classify the Question Domain based on the table below to identify which context matters.

2. Ask scoping questions per the “What to clarify” column for information not already in the query. Where the missing detail would not change the guidance, answer under a stated assumption instead of waiting to ask more questions.
3. You MUST retrieve the reference files from the Routing section below for all candidate services.
4. Provide the answer with specific, actionable guidance, gotchas, and documentation links.
5. When answering Configuration or Security questions, you MUST recommend enabling access logging, CloudTrail data events, and CloudWatch metrics for observability.

Storage Options

AWS storage services covered by this skill, grouped by storage category. For more information on storage categories, see Block, file, and object storage compared (opens in a new tab).

Object Storage

File Storage

When naming a service, you MUST always specify the full name (FSx for Lustre, FSx for Windows File Server, FSx for NetApp ONTAP, or FSx for OpenZFS). EFS and S3 Files can be mounted by Lambda and Fargate. Verify additional services mountable from serverless compute with the latest AWS documentation. FSx for NetApp ONTAP and FSx for OpenZFS data is accessible from serverless compute and S3-based pipelines via S3 Access Points for FSx (exposes file data through the S3 API without copying; surface this when a user needs to read FSx-resident data from S3-native consumers or analytics services).

Block Storage

---

Cross-Service Overlap

Some storage features cross category boundaries, giving a service from one storage category an interface normally associated with another. Surface these when a workload needs access to shared data from multiple protocols or interfaces.

---

Security

Cloud security at AWS is the highest priority and is a shared responsibility between AWS and the customer. You MUST include security guidance when recommending or configuring storage resources. You MUST always recommend enabling encryption at rest and encryption in transit. Where encryption at rest is optional or not the default, explicitly advise enabling it at creation time, because it is often immutable after creation. You MUST recommend scoping IAM policies to the authorized principals with least-privileged permissions. You MUST recommend adding condition keys (e.g., aws:SourceArn, aws:SourceAccount, aws:SourceVpc) to resource policies to prevent cross-service confused deputy attacks. You MUST recommend encrypting log destinations: AWS KMS for CloudTrail trails and CloudWatch Logs groups, server-side encryption for server access log buckets, and AWS KMS for SNS topics. Prefer short-lived credentials or IAM-based authentication (e.g., via custom IdP with temporary tokens) over long-lived SSH keys. Where SSH keys are required, enforce rotation policies and store private keys in AWS Secrets Manager for supported services. You MUST recommend restricting security group inbound rules to the narrowest applicable source (specific client security group or minimal CIDR). Service-specific security controls, encryption models, and documentation links are in the Security row of each reference’s Service Information table; you MUST read it before advising on that service.

Routing

When loaded through the AWS MCP server’s retrieve_skill tool: the skill is not installed on the local filesystem. You MUST retrieve each reference via retrieve_skill with the file parameter (e.g. file=”references/s3-general-purpose-knowledge.md”). Do NOT file_read these paths locally. When loaded outside the AWS MCP server (for example from the local filesystem in the Agent Toolkit), read the reference files directly from their relative paths in the skill directory.

Reference files

Specialized skills