Kiro Review Agents

Specialized code review configurations for the Kiro CLI, adapted from the Claude Code PR Review Toolkit. Use them together for broad PR coverage or individually for focused reviews.

Prerequisites

• Kiro CLI installed and configured
• Git

Setup

Copy the .kiro directory into your project root:

cp -r path/to/kiro-starter-kit/.kiro your-project/

Or copy individual agent files from .kiro/agents/ and .kiro/agents/prompts/ if you have an existing .kiro directory.

Quick Start

Switch to the orchestrator and tell it what to review:

/agent swap review-orchestrator

The orchestrator selects the right agents based on your changes, runs them in parallel, and returns an aggregated summary ranked by severity.

Each prompt should specify:

1. Which files to review — a git diff, a file list, staged changes, etc.
2. Which reviewers to run — all agents, or a specific subset

Example Prompts

# All agents on uncommitted changes
Review the uncommitted changes from `git diff --name-only` with all agents

# All agents on staged changes
Review the staged files from `git diff --cached --name-only` with all agents

# All agents on a branch diff
Review the changes between this branch and main from `git diff --name-only main...HEAD` with all agents

# All agents on specific files
Review these files with all agents: src/auth/login.ts, src/utils/validation.ts

# Specific agents on specific files
Run the silent-failure-hunter and pr-test-analyzer on `git diff --name-only`
Check src/auth/login.ts with the code-reviewer

Running a Single Agent

To run one agent in isolation instead of the full orchestrated review:

/agent swap code-reviewer

Then specify the files:

Review src/auth/login.ts and src/utils/validation.ts

Available Agents

Agent	Purpose	Use When...
review-orchestrator	Coordinates all agents and aggregates results	Running a full PR review
code-reviewer	Code quality, bug detection, guideline compliance	Any code changes
pr-test-analyzer	Test coverage quality and completeness	Adding features or modifying tests
silent-failure-hunter	Silent failures, catch blocks, fallback behavior	Changing error handling code
type-design-analyzer	Type encapsulation and invariant enforcement	Adding or modifying types
performance-reviewer	Algorithmic complexity and resource management	Working with data processing, DB queries, or loops
pci-compliance-reviewer	PCI-DSS compliance (Requirements 3, 4, 6, 10)	Touching payment processing or card data
comment-analyzer	Comment accuracy and long-term maintainability	Adding or modifying documentation
code-simplifier	Code clarity and maintainability	After other reviews pass (runs last)

Severity Scale

All agents score issues on the same 1-100 scale:

Score	Level	Meaning
80-100	Critical	Blocks merge
50-79	Important	Should fix before merge
20-49	Suggestion	Worth fixing
1-19	Nitpick	Optional

Customization

Agents automatically read project-specific guidance from two locations:

• .kiro/steering/*.md — Project conventions and rules (e.g., naming conventions, import patterns, error handling policies). Create markdown files here and agents will enforce them during reviews.
• .kiro/skills/**/SKILL.md — Reusable skill definitions that agents can reference.

Agents also expect these files in your project root for additional context:

• AGENTS.md — Project coding guidelines
• .editorconfig — Formatting standards