Kiro Binding

The Kiro binding generates Kiro-compatible agent configurations from your .at topology. Kiro is AWS's AI development environment.

Scaffolding

agentopology scaffold my-topology.at --target kiro

Generated File Structure

.kiro/
  agents/
    researcher.md         # Agent specification
    writer.md
    reviewer.md
  steering/
    topology.md           # Steering document for the topology
  specs/
    code-review.md        # Spec file for the topology workflow
  config.json             # Kiro configuration

Agent Specifications

Each agent generates a markdown spec file with structured sections:

# Researcher

## Description
Research specialist that gathers context from the codebase for the code-review pipeline.

## Model Tier
Standard (sonnet-class)

## Tools
- Read
- Grep
- WebSearch

## Instructions
Analyze the codebase using available tools. Focus on:
1. Identifying relevant files and modules
2. Understanding code patterns and architecture
3. Gathering dependency information

Produce a structured research summary for downstream agents.

Steering Documents

The steering document provides Kiro with high-level topology context:

# Code Review Topology

## Pattern
Pipeline

## Agents
| Agent | Role | Model Tier |
|---|---|---|
| Researcher | Gathers codebase context | Standard |
| Writer | Produces code changes | Standard |
| Reviewer | Reviews and approves changes | High |

## Flow
1. Researcher analyzes the codebase
2. Writer produces changes based on research
3. Reviewer evaluates changes
4. If revisions needed, Writer re-executes (max 2 iterations)

Spec Files

Topology workflows generate Kiro spec files that describe the end-to-end process:

# Code Review Spec

## Goal
Perform automated code review using a three-agent pipeline.

## Steps
1. Research phase: Gather context about the target code
2. Writing phase: Produce improvements or changes
3. Review phase: Evaluate changes for correctness and quality

## Acceptance Criteria
- All relevant code files have been analyzed
- Changes follow existing codebase patterns
- Reviewer has approved or revision limit reached

Configuration File

The config.json ties the agent specs together:

{
  "topology": "code-review",
  "pattern": "pipeline",
  "agents": {
    "researcher": {
      "spec": "agents/researcher.md",
      "tools": ["Read", "Grep", "WebSearch"]
    },
    "writer": {
      "spec": "agents/writer.md",
      "tools": ["Read", "Write"]
    },
    "reviewer": {
      "spec": "agents/reviewer.md",
      "tools": ["Read", "Grep"]
    }
  },
  "steering": "steering/topology.md"
}

Model Mapping

Kiro manages model selection through AWS Bedrock. The binding includes model tier recommendations:

AgenTopology	Kiro Recommendation
`opus`	High-capability tier (Claude Opus via Bedrock)
`sonnet`	Standard tier (Claude Sonnet via Bedrock)
`haiku`	Fast tier (Claude Haiku via Bedrock)

Limitations

Kiro's multi-agent capabilities depend on the current feature set. The binding generates spec and steering files that work with Kiro's document-driven approach.
Flow orchestration is described in steering documents but not directly executable by Kiro.
Hook events are documented in the steering file but require manual implementation.
MCP server configurations are translated to Kiro's tool integration format where supported.
The scale block maps to AWS infrastructure hints in the configuration but does not auto-provision resources.

Kiro

On this page