AWS Kiro¶

AWS Kiro is Amazon's AI coding assistant with agentic capabilities available as both an IDE extension and CLI.

Kiro CLI¶

Kiro CLI is the command-line interface that provides:

Custom Agents: Define specialized agents for workflows
Sub-Agents: Spawn parallel agents for complex tasks
MCP Integration: Connect external tools via Model Context Protocol
Steering Files: Guide agent behavior with context documents
Smart Hooks: Automate workflows with lifecycle triggers

Installation¶

# Install via npm
npm install -g @anthropic-ai/kiro-cli

# Or via Homebrew
brew install kiro-cli

Starting Kiro CLI¶

# Default agent
kiro-cli

# With specific agent
kiro-cli --agent release-coordinator

Custom Agents¶

Agents are JSON files stored at ~/.kiro/agents/{agent-name}.json.

Agent Configuration¶

{
  "name": "my-agent",
  "description": "A custom agent for my workflow",
  "tools": ["read", "write", "shell"],
  "allowedTools": ["read", "shell"],
  "resources": ["file://README.md", "file://.kiro/steering/**/*.md"],
  "prompt": "You are a helpful coding assistant",
  "model": "claude-sonnet-4"
}

Configuration Fields¶

Field	Type	Description
`name`	string	Agent identifier (required)
`description`	string	Agent purpose (required)
`tools`	array	Available tool capabilities
`allowedTools`	array	Tools that run without prompting
`resources`	array	Context files (glob patterns supported)
`prompt`	string	System instructions for the agent
`model`	string	Claude model to use
`mcpServers`	object	MCP server configurations
`includeMcpJson`	boolean	Inherit global MCP config

Creating Agents¶

Interactive CLI:

kiro-cli agent create --name my-agent

Within chat session:

> /agent generate

Agent Management¶

# Swap agents mid-session
> /agent swap release-qa

# List available agents
> /agent list

Sub-Agents¶

Kiro CLI supports spawning sub-agents for complex tasks with parallel execution.

How Sub-Agents Work¶

Task Assignment: Describe a task and Kiro determines if sub-agents are appropriate
Initialization: Sub-agent created with context from agent configuration
Autonomous Execution: Independent task completion
Progress Updates: Live notifications of work status
Result Return: Completed results returned to main agent

Spawning Sub-Agents¶

> Use the security-scanner agent to audit the authentication module
> Use the qa-agent to verify test coverage

Sub-agents can run in parallel for faster completion.

Sub-Agent Tool Access¶

Available	Not Available
read	web_search
write	web_fetch
shell	introspect
MCP tools	thinking
	todo_list

Steering Files¶

Steering files provide context and guidelines for agents.

Structure¶

.kiro/
└── steering/
    ├── project-guidelines.md
    ├── coding-standards.md
    └── release-process.md

Referencing in Agents¶

{
  "name": "my-agent",
  "resources": [
    "file://.kiro/steering/**/*.md",
    "file://README.md"
  ]
}

Resource Prefix

Resources require the file:// prefix and support glob patterns.

MCP Server Configuration¶

Global Config¶

~/.kiro/mcp.json:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/mcp-server-filesystem"]
    }
  }
}

Per-Agent Config¶

{
  "name": "my-agent",
  "mcpServers": {
    "custom-server": {
      "command": "./my-server",
      "args": ["--config", "config.json"]
    }
  },
  "includeMcpJson": true
}

Tool Reference¶

Tool	Description
`read`	Read file contents
`write`	Write/create files
`shell`	Execute shell commands
`glob`	Find files by pattern
`grep`	Search file contents
`web_search`	Search the web
`web_fetch`	Fetch web content

Model Options¶

Model ID	Description
`claude-sonnet-4`	Balanced performance (default)
`claude-opus-4`	Most capable
`claude-haiku`	Fast, lightweight

Converting from Canonical¶

aiassistkit provides adapters for converting between canonical and Kiro formats:

import (
    "github.com/agentplexus/assistantkit/agents/core"
    "github.com/agentplexus/assistantkit/agents/kiro"
)

// Create canonical agent
canonical := &core.Agent{
    Name:         "scanner",
    Description:  "Security scanner",
    Instructions: "You are a security expert...",
    Model:        "sonnet",
    Tools:        []string{"Read", "Grep", "Bash"},
}

// Convert to Kiro format
adapter := &kiro.Adapter{}
kiroAgent := adapter.FromCore(canonical)
// Result:
// {
//   "name": "scanner",
//   "model": "claude-sonnet-4",
//   "tools": ["read", "grep", "shell"]
// }

// Write to user agents directory
kiro.WriteUserAgent(canonical)

Tool Mapping¶

Canonical	Kiro
Read	read
Write	write
Edit	write
Bash	shell
Glob	glob
Grep	grep
WebSearch	web_search
WebFetch	web_fetch

Model Mapping¶

Canonical	Kiro
sonnet	claude-sonnet-4
opus	claude-opus-4
haiku	claude-haiku

Kiro Powers vs Kiro CLI¶

Kiro has two extension mechanisms:

Feature	Kiro Powers	Kiro CLI Agents
Format	MCP server bundle	JSON config
Sub-agents	No	Yes
Parallel execution	No	Yes
Marketplace	kiro.dev/powers	Manual install
Use case	IDE extensions	CLI workflows

For multi-agent workflows like release automation, Kiro CLI agents are recommended over Powers.

Example: Release Agents¶

A set of release-focused agents:

# Install release agents
cp release-coordinator.json ~/.kiro/agents/
cp release-qa.json ~/.kiro/agents/
cp release-security.json ~/.kiro/agents/

# Use coordinator
kiro-cli --agent release-coordinator

# Spawn sub-agents for review
> Use the release-qa agent to verify tests
> Use the release-security agent to audit dependencies