Explyt 5.5: Skills, New Agent Modes, Auto Review, and Edit Scope for Predictable Changes

In Explyt 5.5, working with the agent becomes more controllable, safer, and more extensible. SKILLs provide full support for an open skills standard: you can store them in your project, reuse them across tasks and tools, and the agent will automatically apply the right skills when needed.

We also introduced new agent modes: Plan and Review. Auto Review enables quick checks of generated changes, Edit Scope limits the agent’s editing area, and you can now create custom agent modes tailored to specific roles and workflows. We added support for Claude 4.6 Opus, and refreshed the UI.

Skills support

We delivered full support for the open SKILLs standard. A Skill is designed to solve a specific narrow task, and the agent uses it automatically when needed. You can add a new Skill from the new chat menu, as shown in the video. A Skill is stored in the project as a set of Markdown files in the .explyt/skills folder, where SKILL.md contains metadata and the system prompt. Write Skills manually: this gives a noticeable quality boost (roughly 15–20% in internal measurements) compared to generating them automatically.

What's the difference between Rules, Skills, Workflows, Agent modes

Agent modes: Plan and Review

Plan mode shows a detailed plan of steps required to complete the request and lets you adjust it as you go, so you can guide the agent at each stage. It is useful when you need to align the sequence of actions before a complex task — for example, migrating to a new framework version or creating a microservice from scratch. The plan is broken down into subtasks with acceptance criteria and saved in the repository as a set of files so other team members can discuss it.

Review mode can only read files, so it is safe by default. It is useful when you need to review changes made by the agent or another person. For code, the Review agent can run anti-pattern checks, for example, ensure that the code does not use try/catch on a base exception type (such as Throwable). It uses IDE inspections to highlight potential issues and reduce the risk of accumulating technical debt. Other AI tools do not use such IDE inspections, so they cannot provide any guarantees and may miss technical debt. This is how we ensure that agents introduce less technical debt into your project.

Ability to add your own agent modes

You can now create your own agent modes by clicking a special button in the agent selection menu. This lets you build an agent for a specific task, such as a technical writer for creating documentation in a project. By gradually testing it in practice and adding new Skills, you can create a full-fledged specialist for your team. In such an agent, you can explicitly define the available tools and a list of preinstalled Skills to focus it on a specific scenario (for example, security or documentation). We support a subset of the Claude agents format, so you can take any Claude agent and use it without changes. Support for the format will expand in the next releases.

Auto Review for generated changes

After the agent makes changes, you can not only review them manually but also run Auto Review. This button starts a separate chat with the Review agent, passing the user task and the changes made by the agent. It is useful when you need to quickly verify the changes before creating a pull request and sending it for team review.

Ability to limit the agent edit scope to specific files

We added a setting that limits the agent's editing area to only selected files and folders: the agent can propose and apply changes within the defined scope but will not go outside of it. This reduces the risk of accidental edits in sensitive parts of the repository and makes work more predictable in large projects. It is useful when you delegate a local task to the agent — for example, documentation updates, refactoring a specific module, or changes only in tests.

Even with Edit Scope enabled, the agent can still read all project files, but it can only edit the attached ones.

If you need to forbid the agent from reading or writing certain files in the project entirely, not just in a single task, you can use the readignore/writeignore files.

UI updates

Chat history popup and agent manager in a separate window

Chat history now opens as a popup from the central agent chat panel, without occupying a separate tab. If you want to manage multiple agents at once, you can expand this popup into a separate Explyt Agents window. There you will see agents that are running and those that have finished, and you can open their chats.

User messages are always pinned at the top

User messages are now pinned at the top of the chat so the task context is always in view. Long messages collapse to three lines, while attachments stay fully visible, making viewing more convenient even in long dialogues. It is useful when a request includes coding style requirements or a list of acceptance criteria and you need to keep them in sight.

Notification when the agent finishes its work

When the agent finishes work and the user is not in its chat or the tools window is hidden, a notification will now pop up. This helps you avoid missing the moment when user action is required. It is useful when the agent is working on a long task while you are working in other IDE tabs. We will add notifications in the operating system soon.

New LLM support

For Personal plan users, we added support for Claude 4.6 Opus. It is useful when you need stronger reasoning and text generation for complex tasks in code or documentation.

Install the update now and see how Explyt transforms the way you build, test, and fix code.