Loading image...Kiro
  • CLI
  • Web
  • Powers
  • Enterprise
  • Pricing
  • Docs
SIGN INDOWNLOADS
Loading image...Kiro
Loading image...Kiro
Product
  • About Kiro
  • CLI
  • Web
  • Powers
  • Pricing
  • Downloads
For
  • Enterprise
  • Startups
  • Students
Community
  • Overview
  • Showcase
  • Discord
  • Events
Resources
  • Documentation
  • Blog
  • Changelog
  • FAQs
  • Report a bug
  • Suggest an idea
  • Billing support
Social
Site TermsLicenseResponsible AI PolicyLegalPrivacy PolicyCookie Preferences
  1. Docs
  2. Web
  3. Steering

Steering the agent


Steering gives the agent persistent knowledge about your codebase through markdown files. Part of Kiro Web (Preview), steering files ensure the agent consistently follows your established patterns, libraries, and standards instead of explaining your conventions in every session.

Steering files

The agent automatically looks for steering files in the .kiro/steering/ folder at the root of your repository. These are markdown files that define your team's standards, architecture decisions, and conventions.

Common uses for steering files:

  • Coding conventions and style guidelines
  • Architecture patterns and design decisions
  • Technology stack preferences and version requirements
  • Testing approaches and coverage expectations
  • PR description templates and commit message formats

For detailed guidance on creating steering files, see the steering documentation. Steering files work the same way across Kiro IDE, Kiro CLI, and Kiro Web.

Teaching through code reviews

You can also steer the agent by leaving feedback on pull requests. When you comment on a PR with guidance like "always use our standard error handling" or "follow our naming conventions," the agent learns and applies those patterns to future work across all your repositories.

Only your feedback (the user who created the task) influences the agent's learnings. Other reviewers' comments don't affect what the agent learns.

Steering during a session

During any session, you can steer the agent in real time by providing direction in the chat:

  • "Use the repository's existing error handling pattern"
  • "Follow the same approach as the UserService class"
  • "Make sure to add integration tests, not just unit tests"

In autonomous mode, the agent asks clarifying questions upfront — your answers act as steering for that task. In the default mode, you can steer continuously as you iterate together.

Page updated: April 21, 2026
Autonomous mode
Sandbox