If you would tell a human or a machine about it more than once, put it in the appropriate context mechanism. Don't repeat yourself—codify it.
What are you trying to accomplish?
Give Instructions
Define how the AI should behave, what tone to use, or general guidelines
Add Capabilities
Connect to external tools, APIs, or data sources
Automate Actions
Trigger behaviors automatically based on events or conditions
Save Reusable Patterns
Store prompts, workflows, or agent configurations
Who should these instructions apply to?
Every Conversation I Have
Personal preferences that should always apply to my AI interactions
Any Coding Agent in This Repo
Build steps, test commands, conventions, and project context for agents working in your codebase
Specific Tasks or Domains
Expertise that can be invoked when needed
What kind of capability do you need?
External Systems
Connect to APIs, databases, or external services via Model Context Protocol
Smart, Contextual Activation
Dynamically activate capabilities based on conversation keywords (Kiro Powers)
Domain Expertise
Specialized knowledge or multi-step workflows
When should this automation trigger?
On Specific Events
React to triggers like new emails, schedule, or system events
Project-Level Agent Context
Repo-specific instructions any coding agent should follow when working here
On-Demand Actions
Manual invocation of pre-configured workflows
What do you want to reuse?
My Personal Preferences
How I want the AI to always interact with me
Task-Specific Expertise
Best practices, workflows, or specialized knowledge
Repo-Level Agent Instructions
Build commands, conventions, and project context that any coding agent should know
Steering File
Your personal AI preferences that apply to either all or select conversations. Think of this as your "AI knows me" file—tone preferences, communication style, recurring contexts about your role, team, or projects.
Perfect For:
Your communication style preferences (formal, casual, concise, detailed)
Your role, team structure, and recurring context
Personal preferences you'd tell every AI assistant (in certain situations)
Default behaviors (always cite sources, avoid certain phrases)
Information you reference constantly (company values, your workflow)
agents.md File
A predictable, agent-focused companion to your README. While READMEs are for humans (quick starts, project descriptions, contribution guidelines), AGENTS.md contains the detailed, precise context that coding agents need to work effectively in your repo — build steps, test commands, coding conventions, and project-specific instructions that would clutter a README or aren't relevant to human contributors.
Relationship to README
AGENTS.md doesn't replace your README — it complements it. READMEs stay concise for humans. AGENTS.md gives agents a clear, predictable place for the extra detail they need.
Perfect For:
Build, lint, and test commands agents should run (and common flags or gotchas)
Project structure explanations and where key code lives
Coding conventions and style rules specific to the repo
Dependency or environment setup steps agents need to know
Repo-specific workflows that aren't obvious from the code alone
Kiro Power
Key Advantage
Instead of loading all MCP tools at once, Powers activate dynamically based on keywords in your conversation. This keeps your AI fast and focused while still having access to everything it needs.
Pre-built or custom capabilities within Kiro that extend what the AI can do. Powers are action-oriented—they give the AI new abilities to interact with systems, data, or perform operations. They're essentially smart wrappers around MCP servers that only load when relevant.
Perfect For:
Built-in Kiro integrations (Slack, calendar, project management)
Custom actions specific to your Kiro setup
One-off or on-demand capabilities you invoke as needed
Team-wide capabilities everyone should have access to
Actions that need to happen within Kiro's ecosystem
Making MCP tools available without the overhead of loading everything upfront
Skill
Packaged domain expertise, best practices, or specialized workflows. Skills are knowledge modules that can be invoked when needed, providing the AI with deep context and instructions for specific tasks or domains.
Task-specific guidance that's reused across projects
Teachable expertise that others on your team should also use
MCP Server
Model Context Protocol servers that connect AI to external systems, APIs, or data sources. MCP servers are the bridge between your AI and the outside world—enabling reads, writes, and actions in other systems.
Relationship to Powers
MCP servers provide the underlying functionality, while Kiro Powers offer a way to access those MCP tools dynamically. Think of MCP as the raw capability and Powers as the smart activation layer.
Perfect For:
Connecting to external APIs (GitHub, Linear, databases)
Reading/writing to file systems or cloud storage
Integrating with proprietary or custom internal tools
Giving AI access to real-time data sources
Building custom integrations not available as Powers
Hook
Event-triggered automations that activate AI workflows based on specific conditions or schedules. Hooks are your "if this, then AI does that" automations.
Perfect For:
Responding to events (new email arrives, PR created, Slack mention)