This release improves agent reliability when creating and overwriting long files. We fixed a bug that prevented the agent from recovering from failures, improving overall stability.
Changelog
This release brings spec task management, intelligent code diagnostics, AI-powered commit messages, and refined agent interactions to streamline your development process.
New Features
Support for Sonnet 4.5 (experimental): You can now send your prompts to Sonnet 4.5 via the chat box. Sonnet 4.5 is also available in Auto, our agent which uses a variety of task-specific foundation models under the hood
Spec MVP tasks: During spec creation, you can now mark tasks (including unit tests) as optional to prioritize core features while keeping comprehensive task lists
Diagnostics Tool: Added diagnostics tool to provide agent with syntax, type, and semantic errors, improving agent accuracy and reducing lint errors
Git Commit Message Generation: Draft commit messages using the agent directly from the source control pane with a single click. Learn more →
Repetitive Tool Use Detection: Agent now detects repetitive tool usage patterns and prompts users to continue or stop, optimizing credit consumption
Unified Credits System improvements: Unified credits system with Auto agent capabilities enhancements and bug fixes. This feature provides users a single credit pool and streamlined usage tracking. Learn more →
User Experience Improvements
Long Running Commands: Enhanced executeBash tool to detect and avoid commands like dev servers, test runners, interactive commands etc.
Usage Meter: Usage meter now properly refreshes when switching between accounts within the same session
Credit Usage Persistence: Improved credit usage tracking across account switches and session management
Context Popup Positioning: Improved positioning of the context popup triggered by
#in chat - now consistently appears at the top-left of the input containerTheme Compatibility: Enhanced visibility of the overage toggle and improved account dashboard access in Light High Contrast, Kiro Light, Light+, and Quiet Light themes
Bug Fixes and Improvements
Remote Workspace Reliability: Fixed issues with extension loading in untrusted workspaces when connecting to remote hosts
Updated Model Selection Setting: Introduced new
kiroAgent.modelSelectionsetting, replacing the deprecatedkiroAgent.agentModelSelectionfor model management
This release addresses critical security vulnerabilities and bug fixes.
Security Improvements
V8 Engine Security: Merge Chromium 138.0.7204.251 to address CVE-2025-10585, a zero-day vulnerability addressing type confusion issue in V8 JavaScript and WebAssembly engine.
Command Execution Security: Fixed vulnerability on PowerShell that could allow malicious commands to execute without user confirmation
Bug Fixes
License Access: Fixed render issue in License page
This release brings Auto, our new agent, enhanced usage tracking, and editor improvements to elevate your development experience.
New Features
Note: These features are rolling out gradually between now and October 1st, and users will be migrated to a single pool of credits and new limits. Your limits will reset to the new monthly plan limits once migrated.
Auto: Added Auto to the model options. Auto is an agent that uses a mix of different frontier models such as Sonnet 4 combined with specialized models, and optimization techniques such as intent detection, caching, and more
Unified Limits: Support for new paid plans with unified limits and fractional credit consumption
Usage Dashboard: Support to view usage based on single Kiro credit pool
Usage Meter: Add credit consumption status bar item to allow easier usage tracking
Bug Fixes and Improvements
Editor Upgrade: Upgraded editor to CodeOSS 1.103.2
Account Management: Enable Kiro account deletion for BuilderID
Authentication: Improvements to login reliability and logout flow for better signing experience
UI Improvements: Hide trust command UI on command cancellation
This release includes editor performance improvements, enhanced reliability, and bug fixes to enhance your development experience.
Performance & Reliability
Code OSS v1.100.3 Upgrade: Updated to Code OSS v1.100.3 that adds improved stability and performance improvements
Faster Response Times: Optimized request handling makes interactions more efficient and improves performance
Improved Session Management: Enhanced backend processing reduces interruptions and allows for longer continuous work sessions
Enhanced User Experience
Enhanced Tool Call Recovery: When tool invocations result in repeated failures, Kiro now prompts for your intervention instead of continuing indefinitely, protecting your usage limits and preventing loops
Enhanced Security Patterns: Expanded detection of dangerous shell commands with improved pattern recognition. Risky commands require manual review unless you explicitly trust them. Learn more about security patterns →
Model Update: Removed Claude Sonnet 3.7 from the model selector and set Claude Sonnet 4 as the default for all users. Sonnet 4 delivers improved performance and capabilities across all interactions.
Bug Fixes
Git Repository Initialization: Fixed issue that prevented Git setup for new projects without existing repositories
Model Selection Persistence: Fixed chat model selection not being saved when restarting Kiro IDE
This release brings key improvements to accessibility, account management, and overall platform reliability.
🎟️ Waitlist Code Support
Kiro is now accessible to more developers through waitlist codes. If you have an waitlist code, you can register and start building with AI-powered development workflows alongside our growing community.
💳 Paid Plans Now Available
Paid plans are now live, giving you access to higher usage limits. Choose the plan that fits your development needs and scale your AI-powered workflows.
📊 Usage Tracking & Billing Dashboard
Introducing comprehensive usage tracking to help you monitor your Kiro consumption. The new billing dashboard displays real-time spec and vibe request usage, plan management options, and overage controls for paid plans.
Additional Improvements
Security & Trust
Enhanced Command Trust: Improved handling of piped commands (
ls -la | grep ".js") with clearer trust requirements for command chains.Workspace Protection: Improved file system tool restrictions to better prevent accidental external modifications
Command Boundary: Commands with potentially dangerous patterns now require manual approval unless explicitly trusted.
Performance & Reliability
Automatic Retries: Failed backend requests are automatically retried for improved reliability
Optimized Model Efficiency: Improvements to optimize frontier model usage with Sonnet
User Experience
Light Mode Improvements: Resolved theming issues for better visual consistency
Clearer Login Errors: More helpful error messages when authentication fails
Preview release
Kiro, a new agentic IDE that helps you do your best work with spec-driven development.
Features
Specs
Structured artifacts that formalize the development process for complex features in your application. Learn more

Hooks
Set up event-driven automation like updating documentation or generating tests on events such as file save. Learn more

Steering
Define steering files to guide agent behaviors. Learn more

Agentic chat
Build features from chat. Turn on Autopilot mode for hands-off execution or approve each change manually. Learn more

MCP
Integrate with your favorite tools and services. Learn more
