AI-native Development 2026-09-01 Update Details

The 2026-09-01 update includes the following changes.

Updates to the Context Assets for AI Agents

We have added a group of skills for implementing user modules in Java (the JavaEE development model). Until now the skills targeted the script development model (JSSP). AI agent support now reaches development that requires building things out in Java.

• Implementation support for the Java development model
  • Added skills covering each area of Java implementation: layer structure, database access, IM-Workflow integration, publishing REST APIs, and authorization and account management.
• Expanded screen design (IMDS) support
  • Substantially expanded the templates for export and import screens, along with the reference material for screen components and utilities.

Skill	What it does
java-im-architecture	Designing the layer structure, dependency rules, and exception hierarchy, with implementation examples that span all layers
java-im-service-layer	Implementing the service layer, including transaction boundaries and exception conversion
java-im-mirage-usage	Implementing entities, DAOs, and 2WaySQL with im_mirage, the database access foundation
java-im-workflow-usage	Java implementations of IM-Workflow processing (action processing, arrival processing, branch conditions, and so on)
java-im-web-api-maker-usage	Publishing REST APIs with Web API Maker and switching authentication and authorization methods
java-im-authz-usage	Operating authorization settings (resources, subjects, policies) and checking permissions
java-im-account-usage	Operating account information: login settings, locking, attributes, and role assignment
java-im-role-usage	Operating roles: hierarchies, category management, and search
java-im-profile-usage	Retrieving, registering, and deleting profile images
java-im-storage-usage	File operations on storage (public, session, and system areas)
java-im-lock-usage	Exclusive control in a distributed environment
java-im-identifier-usage	Issuing unique identifiers (IDs)

The skill assets are published here.

Newly Added Implementation Support for the Java Development Model

You can now have an AI agent generate code while adhering to your design policy, even when implementing business logic in Java.

The skills we added fall into three broad groups.

• The foundation of the design — Provides the structural rules for which layer a given piece of processing belongs in, and what to observe when crossing layers. Alongside implementation examples, it also shows the ways of writing code that you should avoid.
• Database access and integration with business features — Covers how to implement the backbone of a business system: database operations, workflow integration, and publishing REST APIs.
• Calling the platform features — Generates code that calls intra-mart platform features — authorization, accounts, roles, files, exclusive control, and ID issuance — in the correct manner.

Coding conventions for Java (structure, naming, Javadoc, logging, and so on) are deployed at the same time, so the way code is written stays consistent even when several people implement it.

These assets are placed when you choose to use Java at project creation. They are independent of the existing assets for the script development model, so if you do not use Java, the configuration remains as before.

We also provide the implementation guidelines as documentation. For details, see the Java Implementation Guide.

Expanded Screen Design (IMDS) Skills

We added the screen patterns and components that come up most often in practice to the skill that generates screens conforming to the intra-mart Design System.

• Templates for export and import screens — Added support for screens that advance through steps in stages, and for displaying a summary of import results.
• Organized the reference material for screen components and layout — Classified it into three groups (individual components, adjustments such as margins and text color, and standard layouts that combine several components) so that what you need is easier to find.
• Refinements for list and input screens — Added guidance on the aspects that affect usability: what to display when there are no records, tables that keep their headings fixed while scrolling, retaining search conditions, and how to indicate required fields.

We also added checks to the mechanism that mechanically inspects whether a generated screen conforms to the conventions.

Revised the Conventions for Screen Structure

We revised the outermost structure of generated screens so that it does not duplicate the frame provided by the theme.

For components embedded in portal screens (portlets), we added a convention that assigns an identifier unique to each screen, so that multiple components placed on the same screen do not interfere with one another.

Other Improvements

• Made the IM-Workflow definition generation skill common — Because it is now used from both the script development model and the Java development model, we renamed it from jssp-im-workflow-generator to base-im-workflow-generator. It also supports the registration method used when IM-Workflow processing is implemented in Java.
• Support for locales other than Japanese, English, and Simplified Chinese — Previously the assets were sometimes not placed in environments using other locales. We changed this so that the English version is placed in those cases.

Release of the Visual Studio Code Extensions

We have published the VS Code extensions for intra-mart Accel Platform on the Visual Studio Code Marketplace.

Creating and deploying projects, which used to require typing commands, and investigating scripts, which used to rely on log output, can now be done directly from the VSCode UI.

Extension	What it does
intra-mart e Builder for Accel Platform (Core)	Creating and managing projects, updating assets, deploying to a server
intra-mart e Builder for Accel Platform (Debug)	Debugging server-side scripts, running tests and checking coverage
intra-mart e Builder for Accel Platform	An extension pack that installs the two above together

• Installing the extension pack gets you every extension you need in a single operation.
• The UI is available in Japanese, English, and Simplified Chinese, following the display language of VSCode.
• Accel CLI is bundled with Core, so you do not need to install it separately. Updating the extension also updates the bundled CLI.
• VSCode 1.127.0 or later is required. Because the extensions run builds and tests, they do not work in an untrusted folder.

The installation steps and how to use each feature are summarized in Using Visual Studio Code.

Development Support Module (velbench-1.2.0)

We have released version 1.2.0 of the Development Support Module (velbench). This release centers on the work needed to run correctly in multi-node environments and on improvements that let the developers using a staging environment complete the preparation themselves.

Support for Multi-Node (Cluster) Configurations

We fixed a defect where deploying to a staging environment and running tenant environment setup would fail in environments running application servers on multiple nodes.

The cause was that the internally held information was refreshed only on the node that accepted the deployment, leaving the other nodes referring to the old assets. Because the result varied depending on which node the processing was routed to, it surfaced as a defect that was difficult to reproduce.

• Discarding the held information is now distributed to every node.
• We revised the order in which assets are expanded and switched over, so that the target of the switch is guaranteed to have its assets present. If expanding or switching fails, the previous deployment is kept as is. No half-finished state remains.

We also resolved an issue where internal resources were not released when a staging environment was deleted.

Improvements to Owner Management

Managing the developers who may operate a staging environment (its owners) can now be done from the UI without guesswork.

• You can now search for and select a user — Previously you typed in a user code, which meant looking it up in advance, and a mistyped code could still be registered. We changed it to a method where you search by keyword and choose from the candidates, preventing incorrect registration.
• The list now shows user names — Previously only user codes were shown. You can now see at a glance who is registered.

Added Authorization for Issuing API Keys

An API key is required in order to use velbench's features from Accel CLI or from an agent. Developers can now issue an API key themselves simply by being granted the development support role. Previously they could not reach the API key issuance screen and had to ask a tenant administrator to configure it separately.

Bug Fixes and Hardening

• Hardened screen display — On the staging environment screens, the conversion applied to displayed values excluded certain characters (quotation marks), leaving room for unintended content to be embedded in the screen. We changed it to convert every target character.
• Improved application server portability — The way internally used files were referenced was, strictly speaking, not in line with the specification. It worked without issue on Resin, but we corrected it because it could cause problems on other application servers. There is no change in behavior.
• Corrected the declaration of prerequisite modules — We re-declared the modules that are actually required. IM-Workflow is no longer a mandatory prerequisite module.

Container-Based Development Environment Assets

We updated the configuration of the verification environment built with Docker.

• The Development Support Module is now included by default — velbench-1.2.0.imm is now bundled by default.
• The debugger module is now included by default — The debugger module (im_eb_js_debugger.imm) required for debugging server-side scripts is bundled by default as well. You do not need to incorporate it yourself or rebuild the WAR.
• Documented how to open the debug port — We documented the steps for opening the debug port, which are needed for remote debugging, in the appendix.

Documentation

New Java Implementation Guide

We added the guidelines for implementing user modules in Java (the JavaEE development model) as a new category.

It explains the structural thinking behind which layer a given piece of processing belongs in and what to observe when crossing layers, together with implementation examples. For database access, it covers how to use im_mirage, the foundation, from defining entities through writing SQL to handling transactions.

We also prepared a chapter collecting the ways of writing code that you should avoid. It can be used as a set of review criteria.

New "Using Visual Studio Code" Section

We added a category that explains everything from installation to how to use each feature of the VS Code extensions we published.

• Installing the extensions and confirming that they were installed correctly
• Creating and updating projects, and deploying to a server
• Debugging server-side scripts (including preparing the server side)
• Running tests and checking coverage

The structure lets you skip the relevant pages if you do not perform debugging.

Added Guidance on Automatic Fixes Triggered by E2E Test Failures

We added an operational example to the DevOps guidelines in which a failing screen-operation test (E2E test) is automatically recorded as an issue, which is then connected through to investigation and correction by AI.

Deploying to the verification environment, running the tests, detecting a failure, registering the issue, having AI correct it, and proposing the change all happen without human intervention. Once you incorporate the proposed correction, the same flow is verified again on the next update.

So that it makes sense even if you have never used GitHub Actions, we included a diagram and a table that explain in plain terms what each stage is doing.

Added a Disclaimer

We added a page covering the handling of generated output, the license for the assets we provide, and the conditions for using the extensions.

It states that how you apply agent-driven automation to your development process is something to decide and carry out at your own discretion and responsibility. For details, see the Disclaimer.

Added a License Page

A page summarizing the license terms for each of the assets we provide has been added.

It lists the license that applies to Accel CLI, the context for AI agents, the IM-Juggling library, Docker Stacks, the Development Support Module, the debugger module, and the Visual Studio Code extensions, along with the libraries bundled with each. For details, see the License page.

Along with this, the explanation of the context asset license that used to live in the disclaimer has been consolidated into the license page.

Other

• Updated the download link for the Development Support Module to velbench-1.2.0.
• Revised the order of the categories, placing the Java Implementation Guide after the Module Reference.

Accel CLI

Added the accel preset Command

You can now name a combination of asset configuration settings, save it, and share it across your team.

What you save is recorded inside the project, so committing it to version control lets everyone on the team reproduce the same configuration by name.

• You no longer have to re-select a configuration each time for each kind of work, such as "for screen development" or "for server-side development."
• You can specify it when running accel update, or accel switch described below, skipping the interactive questions.
• In addition to the name, you can give it a display label and a description in your own language.

Added the accel switch Command

We added a command that switches the configuration of already-deployed assets without downloading anything.

Until now, the only way to change the configuration was accel update. Because update also pulls in the latest assets, changing only the configuration still involved network access and an asset update. accel switch reconfigures while keeping the assets as of the point they were deployed, so no unintended updates are mixed in.

• You can place only the skills you need for the work at hand. This lets you narrow the amount of information you feed an agent on a per-task basis.
• It does not touch files you have edited. Edited files that fall within the scope of the switch are left unchanged and reported in a list at the end of the processing.
• Dependencies are re-fetched only when necessary. If you are only swapping skills, it finishes right away.
• By specifying one of the presets above, you can switch without re-selecting the configuration.

Introduced an Asset Cache

Fetched assets are now kept in your environment and reused across commands and across projects. From the second time onward, already-fetched assets are not downloaded again.

• You can check what is held and how much space it uses with accel cache info, and delete it with accel cache clear. Deleting it does not affect your projects.
• Using it from several working directories at the same time causes no problems.

Automatic Generation of the Agent Integration Settings (MCP)

The configuration file that agents use to connect to iAP is now generated automatically.

The previous approach rewrote existing settings, which meant you had to have prepared those settings in advance. We changed it to generate the settings themselves from a template supplied with the assets, so the connection target and credentials are filled in automatically. You do not need to write the configuration file yourself.

• It generates the file in the format appropriate for Claude Code, GitHub Copilot, or Codex.
• It does not touch settings you added yourself. Running it repeatedly does not create duplicates, and only the settings made unnecessary by a configuration change are removed.
• Updating assets or switching the configuration regenerates the settings to match.
• If the connection details are not yet set, nothing is generated and you are directed to run accel login.

Groundwork for Operating from a UI

We have made the processing that the CLI provides available to be called directly from a program. The VS Code extensions described above use this mechanism to run the same processing as the CLI.

There is no change in behavior when using it as a command.

Improved Safety

• Suppressed the display of credentials — When a communication error occurs, credentials are now stripped from the connection details included in the message. They no longer remain in error output or execution logs.
• Neutralized displayed content — Strings originating from assets were previously written to the screen as-is, leaving room for the display to be spoofed. We changed this to strip control characters before displaying them.

Preventing Operational Mistakes

• Detection of unknown options — When an undefined option is specified, the command now stops with an error before processing begins. Previously, processing could proceed without you noticing the mistake in your input.
• Better guidance in proxy environments — When a connection through a proxy fails, the tool now tells you which environment variables to check.
• Improved log output — When writing execution results to a file, decorative control characters are no longer mixed in.

How to Update from the 2026-08-01 Release Assets

Run the following command in the project root.

bunx @intra-mart/accel@latest update --asset-ref=2026-09-01-release

Caution

Do not omit --asset-ref=2026-09-01-release.

If dependency resolution fails in Maven

In this release we have replaced some of the assets distributed through the Maven repository. An error like the following may occur when you build after updating.

[ERROR] Failed to execute goal on project my-accel-project: Could not resolve dependencies for project com.example:my-accel-project:im_module:0.1.0
[ERROR] dependency: jp.co.nttdata:im_nd_universalextractor:jar:2.2.2 (compile)
[ERROR]         jp.co.nttdata:im_nd_universalextractor:jar:2.2.2 was not found in https://maven.intra-mart.jp/ during a previous attempt. This failure was cached in the local repository and resolution is not reattempted until the update interval of intra-mart-maven has elapsed or updates are forced

In that case, move aside (rename) or delete the .m2/repository directory under your home directory, then build again.

mv ~/.m2/repository ~/.m2/repository.bak

If accel update is not available, replace the assets with accel detach → accel attach.

If you want to use the assets for the Java development model, add Java in the feature selection when updating.