AWS Agent Toolkit Skills: Complete Guide to Core Capabilities and Usage

The AWS Agent Toolkit provides 15 core skills—including serverless, containers, IAM, observability, and Bedrock integrations—that enable AI coding agents to execute AWS development tasks through self-contained markdown definitions stored in plugins/aws-core/skills/.

The AWS Agent Toolkit is an open-source repository (aws/agent-toolkit-for-aws) that equips AI coding agents with domain-specific expertise for AWS development and operations. These AWS Agent Toolkit skills are packaged under the aws-core plugin and loaded on demand by agents such as Claude Code, Codex, Cursor, and Kiro to handle everything from authentication to infrastructure deployment.

What Are AWS Agent Toolkit Skills?

Each skill is a self-contained markdown definition (SKILL.md) located in plugins/aws-core/skills/<skill-name>/ that contains concise capability descriptions, step-by-step instructions for agents, and reference materials. According to the source code, a typical skill directory includes:

  • SKILL.md – The master skill definition containing capability descriptions and agent instructions
  • references/ – Markdown files with detailed guidance on specific patterns or services
  • assets/ – Helper scripts, templates, or executable code (e.g., powertools-handler.py for Lambda)

This architecture allows agents to retrieve only relevant guidance for the current task, reducing noise while ensuring access to vetted best practices.

Core Skill Categories

The AWS Agent Toolkit organizes its 15 core skills by functional domain. Each skill is implemented as a dedicated directory under plugins/aws-core/skills/ with its own SKILL.md file.

Authentication and Security

Compute and Infrastructure

SDK Integration

Operations and Observability

AI and Reusable Components

How to Install and Use Skills

Agents consume AWS Agent Toolkit skills through the aws-core plugin. The installation process loads the skill definitions into the agent's context.

Installing the Plugin


# Add the repository to the marketplace (Cursor, Kiro, etc.)

npx skills add aws/agent-toolkit-for-aws/skills

# Install the core plugin

/plugin install aws-core@claude-plugins-official

Invoking Specific Skills

Once installed, agents invoke skills by referencing their directory name. The skill returns ready-to-deploy code or configuration based on its reference materials.

Generate a Lambda function using the serverless skill:

skill aws-serverless create-lambda \
  --runtime python3.11 \
  --handler handler.handle \
  --description "Process S3 upload events"

This command processes the request against aws-serverless/references/lambda.md and returns a SAM/CloudFormation snippet plus a minimal handler file.

Query cost optimization using the billing skill:

skill aws-billing-and-cost-management recommend-savings \
  --service ec2 \
  --region us-east-1

The skill reads references/service-optimization.md to return Savings Plans versus Reserved Instances comparisons.

Access Python SDK patterns:

skill aws-sdk-python-usage list-s3-objects \
  --bucket my-data-bucket \
  --prefix logs/

This generates a boto3 script with proper pagination and error handling as defined in references/s3.md.

Configure Bedrock models:

skill amazon-bedrock create-model \
  --model-id anthropic.claude-v2 \
  --instance-type ml.g5.xlarge

The skill produces the required IAM policy, runtime configuration, and sample invoke script from references/sdk-converse-api-python.md.

Key Implementation Files

The AWS Agent Toolkit's architecture relies on specific configuration and hook files:

  • plugins/aws-core/.cursor-plugin/plugin.json – Plugin manifest used by Cursor to expose core skills
  • plugins/aws-core/.mcp.json – MCP Server configuration defining AWS MCP endpoint connectivity
  • plugins/aws-core/hooks/secret-safety.py – Security hook that redacts secrets before transmission back to the agent
  • plugins/aws-core/skills/*/SKILL.md – Master definitions for each of the 15 core skills
  • plugins/aws-core/skills/*/references/*.md – Detailed documentation referenced by skills during execution
  • plugins/aws-core/skills/*/assets/* – Executable scripts and templates (e.g., validation scripts for CloudFormation)

Summary

The AWS Agent Toolkit provides a comprehensive skill set for AI-assisted AWS development:

  • 15 core skills covering authentication, compute, SDKs, observability, and AI services
  • Markdown-based architecture where each skill in plugins/aws-core/skills/ contains a SKILL.md definition and supporting references
  • Agent compatibility with Claude Code, Codex, Cursor, and Kiro through the aws-core plugin
  • Security enforcement via the secret-safety.py hook that redacts sensitive credentials
  • On-demand loading that retrieves only relevant guidance, ensuring agents use current best practices

Frequently Asked Questions

What file structure defines an AWS Agent Toolkit skill?

Each skill is a directory under plugins/aws-core/skills/ containing a SKILL.md file with capability descriptions and agent instructions, a references/ subdirectory with detailed markdown documentation, and an optional assets/ folder with helper scripts or templates.

How do AI agents consume AWS Agent Toolkit skills?

Agents install the aws-core plugin via commands like /plugin install aws-core@claude-plugins-official, then invoke specific skills by name (e.g., skill aws-serverless create-lambda). The skill processes the request against its reference materials and returns implementation code or configuration guidance.

Which AWS Agent Toolkit skill should I use for serverless development?

Use the aws-serverless skill, located at plugins/aws-core/skills/aws-serverless/SKILL.md, which provides guidance for Lambda functions, API Gateway, Step Functions, and event sources, including ready-to-deploy SAM templates and handler code.

Are AWS Agent Toolkit skills secure for handling AWS credentials?

Yes. The toolkit includes a security hook at plugins/aws-core/hooks/secret-safety.py that automatically redacts secrets and credentials before they are transmitted back to the agent, while the signing-in-to-aws and aws-secrets-manager skills provide secure patterns for authentication and secret handling.

Have a question about this repo?

These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:

Share the following with your agent to get started:
curl -s "https://instagit.com/install.md"

Works with
Claude Codex Cursor VS Code OpenClaw Any MCP Client

Maintain an open-source project? Get it listed too →