# How Plugins Simplify AWS Agent Toolkit Skill Integration

> Simplify AWS Agent Toolkit skill integration with self-contained plugins. Leverage decoupled logic and auto-registration to easily add new functionality without framework code changes.

- Repository: [Amazon Web Services/agent-toolkit-for-aws](https://github.com/aws/agent-toolkit-for-aws)
- Tags: how-to-guide
- Published: 2026-06-27

---

**AWS Agent Toolkit decouples skill logic from the core engine by encapsulating capabilities into self-contained plugins that auto-register through manifest files, eliminating the need to modify framework code when adding new functionality.**

The AWS Agent Toolkit achieves seamless AWS Agent Toolkit skill integration by treating plugins as first-class extension points. Rather than embedding business logic directly into the agent core, the framework delegates capabilities to discrete packages under the `plugins/` directory. This modular approach allows developers to install, validate, and invoke complex AWS workflows through declarative configuration files rather than imperative code changes.

## Modular Registration via Self-Describing Plugins

Each plugin operates as an autonomous unit within the repository's `plugins/` directory. The `aws-core` plugin exemplifies this pattern: its **[`plugins/aws-core/README.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-core/README.md)** enumerates 15+ default skills including **`aws-sdk-python-usage`**, **`aws-serverless`**, and **`observability`** while declaring the plugin's purpose and supported agent platforms.

This self-documentation ensures that the agent loader can introspect available capabilities without executing arbitrary code. When a plugin is installed, the toolkit immediately registers all declared skills with the internal registry, making them available for agent invocation.

## Automatic Skill Discovery Through SKILL.md Manifests

Inside each plugin, individual skills are defined by **[`SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/SKILL.md)** files located in skill-specific subdirectories. For example, **[`plugins/aws-core/skills/aws-serverless/SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-core/skills/aws-serverless/SKILL.md)** contains the skill name, description, input schema, and the Python routine that implements the logic.

The toolkit's loader scans the plugin directory tree and parses each manifest automatically. This eliminates manual registration steps—adding a new skill requires only placing a properly formatted [`SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/SKILL.md) file into the plugin structure. The core framework remains untouched, achieving true plug-and-play extensibility.

## MCP Server Configuration Bundling

Beyond skill definitions, plugins provision **MCP Server** configurations that enable real-time documentation search and AWS service invocation. The `aws-core` plugin ships with pre-configured endpoints for **`call_aws`** and **`run_script`** operations, allowing agents to interact with AWS services immediately after installation.

This bundling removes setup friction. Users do not need to configure service endpoints or authentication chains separately; the plugin encapsulates both the skill logic and the infrastructure required to execute it.

## One-Command Installation and Validation

Installing capabilities follows a streamlined workflow. Users execute a single command to fetch the plugin, followed by a reload operation to refresh the skill registry:

```bash

# Install the core plugin for Claude

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

# Or for Codex

codex plugin marketplace add aws/agent-toolkit-for-aws

# Reload to discover new skills

/reload-plugins

```

After reloading, the agent discovers all contained skills and exposes them through the **`retrieve_skill`** tool. This workflow applies equally to specialized plugins like `aws-agents` and `aws-agents-for-devsecops`, which provide skills such as **`agents-optimize`** and **`setup-security-agent`** via their respective manifest files at **[`plugins/aws-agents/skills/agents-optimize/SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-agents/skills/agents-optimize/SKILL.md)** and **[`plugins/aws-agents-for-devsecops/skills/setup-security-agent/SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-agents-for-devsecops/skills/setup-security-agent/SKILL.md)**.

## Pre-Registration Schema Validation

Before a plugin enters the registry, **[`tools/validate.py`](https://github.com/aws/agent-toolkit-for-aws/blob/main/tools/validate.py)** enforces structural integrity. This validation script checks that plugin manifests contain required fields including `name` and `plugins`, ensuring that newly added skills conform to the expected schema.

By catching malformed configurations during the validation phase—rather than at runtime—the toolkit prevents registration errors and guarantees that the `retrieve_skill` tool receives properly structured metadata.

## Practical Usage Example

Once installed, skills become callable through the agent's standard interface:

```python
from agent_toolkit import Agent

# Initialize agent with plugin-provided skills

agent = Agent()

# Invoke the aws-serverless skill

result = agent.run_skill(
    skill_name="aws-serverless",
    inputs={"action": "deploy", "template": "my-app.cdk.json"}
)
print(result)

```

The agent resolves the skill name against the plugin registry, loads the implementation defined in the corresponding [`SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/SKILL.md), and executes the Python routine with the provided inputs—all without requiring import statements or configuration boilerplate in the calling code.

## Summary

- **Plugin encapsulation** separates skill logic from the core agent framework, residing in the `plugins/` directory with self-describing [`README.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/README.md) files.
- **Automatic registration** occurs when the loader parses [`SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/SKILL.md) manifests from paths like [`plugins/aws-core/skills/aws-serverless/SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-core/skills/aws-serverless/SKILL.md), eliminating manual setup.
- **MCP Server bundling** provides immediate AWS service connectivity through configurations included in each plugin.
- **One-command install** via `/plugin install` or `codex plugin marketplace add` followed by `/reload-plugins` instantiates all plugin capabilities.
- **Schema validation** via [`tools/validate.py`](https://github.com/aws/agent-toolkit-for-aws/blob/main/tools/validate.py) ensures manifest integrity before skills enter the registry.

## Frequently Asked Questions

### What file defines a specific skill within a plugin?

Each skill is defined by a [`SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/SKILL.md) file located in the skill's subdirectory within the plugin folder. For example, [`plugins/aws-core/skills/aws-serverless/SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-core/skills/aws-serverless/SKILL.md) specifies the skill's name, description, input schema, and implementation routine. The toolkit's loader scans for these files to build the skill registry automatically.

### How do I add a new skill without modifying the core agent code?

Create a new [`SKILL.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/SKILL.md) file inside an existing or new plugin directory under `plugins/`. The [`tools/validate.py`](https://github.com/aws/agent-toolkit-for-aws/blob/main/tools/validate.py) script will validate the manifest, and upon running `/reload-plugins`, the agent automatically discovers and registers the skill. No changes to the core framework are required because the plugin system handles discovery and binding.

### What validation ensures plugin quality before registration?

The [`tools/validate.py`](https://github.com/aws/agent-toolkit-for-aws/blob/main/tools/validate.py) script performs pre-registration checks on plugin manifests, verifying required fields such as `name` and `plugins`. This validation guarantees that skills conform to the expected schema and that the registry only contains properly structured definitions, preventing runtime errors during skill invocation.

### Can plugins include infrastructure configurations for AWS services?

Yes. Plugins bundle MCP Server configurations that provision AWS service endpoints, enabling capabilities like `call_aws` and `run_script` immediately after installation. The `aws-core` plugin demonstrated in [`plugins/aws-core/README.md`](https://github.com/aws/agent-toolkit-for-aws/blob/main/plugins/aws-core/README.md) includes these configurations, allowing agents to invoke AWS services without additional manual setup.