Skills
Agent skills are curated packages of instructions, code scripts, and reference materials that help AI coding agents complete specific AWS tasks. Skills bridge the gap between what AI models know from their training data and what is needed to work effectively with AWS — especially for newer services, complex multi-service workflows, and tasks where best practices matter.
Topics
What is a skill?
A skill is a directory containing a SKILL.md file and optional supporting files.
The SKILL.md file includes a brief description and a set of instructions that tell the
agent how to complete a task — which steps to follow, which AWS APIs to call, which mistakes
to avoid, and how to verify the result. Skills can also include reference files with deeper guidance
on specific subtasks, code scripts for deterministic operations, and slash commands that let you
invoke the skill directly.
Skills are designed to be lightweight and modular. Each skill focuses on a single task or domain, and agents load only the skills they need for the current task. A skill typically consumes a few thousand tokens when loaded — far less than the equivalent documentation — because it contains only the information the agent needs to act, not background context the agent already has.
How agents discover and use skills
There are three ways agents get access to skills:
-
Bundled with a plugin — Each plugin includes a curated set of skills that are available to the agent immediately after installation. The agent can use these skills without any network calls or additional setup.
-
Installed locally — You can download individual skills from the Agent Toolkit for AWS repository on GitHub
and add them to your agent's skills directory. -
Discovered at runtime via the AWS MCP Server — Agents can search for and retrieve skills on demand through the AWS MCP Server, without any local installation. The agent uses the
search_documentationtool to find relevant skills and theretrieve_skilltool to load them into context.
When an agent uses a skill — regardless of how it was installed — the process is the same:
-
The agent reads the skill's description to determine if it is relevant to the current task.
-
If relevant, the agent loads the full instructions from
SKILL.md. -
The agent follows the skill's procedures, loading reference files only as needed.
-
After the task is complete, the agent releases the skill content from context.
This progressive disclosure approach means that having many skills available does not slow your agent down or consume unnecessary context. Only the skills relevant to the current task are loaded.
Types of skills
The Agent Toolkit for AWS includes several types of skills:
-
Service decision guides help agents choose the right AWS service for a use case. For example, a database decision guide helps agents recommend DynamoDB, Aurora, or DSQL based on the workload requirements.
-
Step-by-step procedures provide tested workflows for common tasks like creating S3 Tables, setting up AWS Glue ETL pipelines, configuring IAM policies, and deploying serverless applications.
-
Troubleshooting guides provide diagnostic procedures for common errors, with steps to identify the cause and resolve the issue. For example, a CloudFormation deployment troubleshooting skill covers the top failure patterns and how to fix them.
-
SDK usage guides provide language-specific best practices for the AWS SDKs, covering common mistakes that models consistently get wrong (like DynamoDB marshalling in JavaScript or pagination patterns in Python).
Available skills
Each plugin includes a curated set of skills covering the most common workflows for that domain.
The full set of skills — including domain-specific skills for individual AWS services —
is available on GitHub
To see what skills are available from within your agent, ask: "What AWS skills do you have available?" or "Search for AWS skills related to databases."
You can also browse and install skills from the command line:
npx skills add aws/agent-toolkit-for-aws/skills
Using skills with and without the AWS MCP Server
Skills work best with the AWS MCP Server, which provides authenticated API access, sandboxed script execution, and enterprise controls like CloudWatch metrics and IAM context keys. For production workflows, use the AWS MCP Server.
Skills also work without the AWS MCP Server. When agents do not have access to the MCP Server, they can run the same AWS operations using the AWS CLI directly. Each skill includes instructions that work with both approaches.
