Kiro IDE 1.0: Agent Focus Mode, Custom Agents, and a Capability-Based Permissions System

Kiro’s IDE hit 1.0 on June 25. The release focuses on giving developers tighter control over what agents can do and a new layout for people who spend most of their day directing agents rather than writing code manually.

Agent Focus Mode

Agent Focus Mode is an experimental alternative layout where conversation occupies the main panel and sessions run in a sidebar on the left. It’s designed for workflows where you’re orchestrating multiple agents at once rather than editing files directly.

From Focus Mode you can start independent parallel sessions, see their status at a glance, and review file changes as inline diffs as they come in. Toggling back to the traditional IDE view carries your work over in both directions, so nothing is lost if you switch mid-session.

Kiro frames this as a chat-first view for developers who prefer directing agents through dialogue. Whether that’s actually more efficient than the traditional layout depends heavily on how you work, and the “experimental” label means the interface may still change. But it’s a meaningful alternative for users who find themselves primarily instructing agents rather than making manual edits.

Capability-Based Permissions

The 1.0 permissions system evaluates every file read, shell command, and MCP tool call against a set of rules you define. When the agent tries to do something outside those rules, it prompts for consent instead of proceeding.

You can declare tool access with tags in a custom agent’s configuration: read, write, shell, web. The system is scoped per agent, so a specialized code-review agent can be locked to read-only file access while a deploy agent has broader shell permissions.

Previously, Kiro’s permission model was coarser. This gives teams a way to build and share agents with explicit, auditable capability boundaries, which matters for organizations where agents are running commands in shared or production-adjacent environments.

Custom Agents

Custom agents are now defined in Markdown files with a tag-based tool declaration syntax. You specify what tools the agent can access (file reads, shell commands, MCP server calls), write a system prompt in the same file, and you’re done.

These files can be committed to version control and shared across a team, which keeps agent configurations alongside the rest of the project rather than buried in local settings. You can also configure MCP servers inline in the agent file.

Natural Language Hooks

Hooks in Kiro automate tasks in response to events, like running tests when a file is saved or posting a diff when a PR is ready. Previously you wrote them in a structured format. Now you can describe what you want in plain language and Kiro generates the hook configuration for you.

The output is a JSON file in .kiro/hooks/ in the v1 hook format, which means you can read and edit the result directly if the generated config needs adjustments.

Session Export and Dockable Chat

Conversations can now be saved as zip files containing a Markdown transcript and metadata, useful for audits, documentation, or sharing a troubleshooting session with a colleague.

Chat panels can also open as full-width editor tabs, which makes multi-monitor setups more practical. The panel and tab versions stay in sync.

---

Kiro 1.0 is available at kiro.dev. The Agent Focus Mode docs are at kiro.dev/docs/experimental/focus-mode and the full 1.0 changelog is at kiro.dev/docs/whats-new-1-0.

Source: Kiro 1.0 release notes / Kiro IDE changelog