AI-native Development 2026-08-01 Update Details

The 2026-08-01 update includes the following changes.

Updates to the Context Assets for AI Agents

We have added the following groups of skills.

• Business process management (IM-BPM)
  • Added skills for IM-BPM features, such as having an AI agent create a business flow diagram (BPMN).
• Testing
  • Added a skill that automatically generates a list of test perspectives and detailed test case sheets from the specifications.
• Environment setup (sample data)
  • Added a skill that automatically generates, in a single pass, a complete set of samples for trial and demo use.
• Generating code that calls IM-LogicDesigner routing
  • Added a skill that generates the code for calling an IM-LogicDesigner routing definition (logic flow API) from the script development model.

Skill	What it does
bpm-docs-generator	Creates specifications from a business flow diagram; converts them into a format for your internal knowledge base
bpm-scripts-generator	Generates the screens and processing for a business process from the specifications
bpm-xml-reflector	Reflects the finalized specifications and program content back into the business flow diagram
test-spec-generator	Creates a list of test perspectives and test case sheets (Excel / for viewing in a browser)
jssp-sample-setup-generator	Generates a complete set of sample data for trial and demo environments
jssp-im-logic-usage	Generates code that calls IM-LogicDesigner routing

The skill assets are published here.

Newly Added Development Support for Business Process Management (IM-BPM)

You can now assemble a system starting from a business flow diagram (BPMN) — a diagram that depicts the flow of work, such as "who approves what, and when."

Specifically, we have added the following three support features.

• Automatically creating specifications from a business flow diagram — Reads the flow of work, the branching conditions, and the tasks depicted in the diagram, and writes them out as specifications a person can read. The specifications created can also be converted into a format that can be registered directly into your internal knowledge base (intra-mart Knowledge).
• Automatically generating programs from the specifications — Based on the specifications created, generates the screens and processing needed to actually run the business process.
• Reflecting decisions back into the business flow diagram — Writes the content finalized in the specifications and programs back into the original business flow diagram. This prevents the diagram and the real thing from diverging.

As a result, you can carry out the whole sequence — business flow diagram → specifications → program → reflection back into the business flow diagram — with the support of AI. We have also provided a mechanism that mechanically inspects the business flow diagram for errors in how it is written, reducing rework in later stages.

Added Automatic Creation of Test Documents (Test Perspective Lists and Test Case Sheets)

From the "conditions to be satisfied" written in the specifications, you can now automatically create a list of test perspectives and detailed test case sheets for each screen.

• Because it follows the layout and formatting of your existing test case sheets, you can continue to use your organization's established format.
• In addition to Excel format, you can also choose an HTML format for viewing in a browser, which works even in environments where no dedicated tool is installed.

This substantially shortens the creation of test documents, which used to be time-consuming manual work.

Added Automatic Generation of Sample Data Setup Assets

You can now automatically generate, in a single pass, the complete set of sample data needed to stand up a trial or demo environment.

It prepares everything at once: user permission settings, menus, scheduled processing, initial data, the registration of components (portlets) on the portal screen, and the import of workflows and business logic. It generates all three languages (Japanese, English, and Simplified Chinese) at the same time.

This reduces both the effort of preparing demo and trial environments and the trouble caused by missed settings.

Added Code Generation That Reuses Existing Business Logic

You can now automatically generate the code for calling a REST API (routing definition) already created in IM-LogicDesigner from a screen you are newly building.

Because the input and output information needed for the call is read automatically from the system, the person in charge does not have to look up the specifications again. This makes it easier to proceed with development that takes full advantage of existing assets.

Quality and Stability

• Moved automatic inspection of generated output into a subagent — We changed the review and correction work that follows screen generation so that it is carried out by a dedicated subagent. Because less information is handled in a single exchange, quality stays consistent even on large-scale development.
• Resolved omissions when importing workflow definitions — Resolved an issue where, when an approval route had multiple branching conditions, some of them could fail to be imported.
• Made component registration on the portal screen reliable — Permission settings are now aligned automatically so that components display correctly on the administration screen. In addition, import failures caused by names exceeding the character limit are now detected in advance.
• Bug fixes — Fixed errors in how the configuration file for GitHub Copilot was written, and filled in some missing type definitions.

Development Environment

• Automatically adding the libraries required for IM-BPM — When you choose a configuration that uses IM-BPM, the required libraries are now incorporated automatically.

Change in What We Provide

• On the feature that prevents leaks of confidential information — For environments that use git, we made a change that automatically adds the hooks configuration, in order to prevent API keys and similar items from being published by mistake.

Release of the @intra-mart/juggling-core Library

We have published @intra-mart/juggling-core to the npm registry. This is a library for operating IM-Juggling using TypeScript. By using this library, you can automate operations on an IM-Juggling project — adding user modules, generating the war file, applying patches, and so on — without going through screen operations.

Development Support Module (velbench-1.1.0)

We have released version 1.1.0 of the Development Support Module (velbench). This release adds features such as deploying assets, running jobs, and setting up sample data in the staging environment.

With the changes made in July, the range of what you can verify in the staging environment has expanded as follows.

What you can verify	As of end of June	As of end of July
Screen behavior	○	○
Checking the provided APIs (routing)	○	○
Checking the logs	○	○
Running the tenant environment setup	○	○ (multiple modules supported)
Running jobs (automated processing)	×	○
Loading sample data	×	○
Deploying multiple files at once	×	○

The central theme of this month's update is that the range of "work you can complete within your own workspace, without touching the shared environment" has expanded.

Job Execution

You can now run jobs from the screen and check the results within the staging environment.

• The "Jobs" tab on the staging detail screen (newly added in this release)
  • The jobs defined in the assets are listed, and you can select and run them from there
  • Two job types are supported: Java jobs and script (JSSP) jobs
  • For jobs that do not appear in the list, you can add a row by manually entering the type and path and run them
  • You can specify the parameters passed to the job at run time.

Sample Data Setup

You can now check and run the sample data setup definitions contained in the assets from the screen. You can bring the sample data used during development into the staging environment and set it up.

• Note: What is run is reflected in the database of the actual tenant and cannot be undone. This point is also stated explicitly as a warning in the confirmation dialog shown before execution.

• The "Sample Data Setup" area on the staging detail screen (newly added in this release)
  • Lists the sample data setup configurations present in the assets
  • Checking the definitions for each configuration file
  • Selective execution at the section level
  • Bulk execution covering all configurations
• After execution, the counts of "succeeded / failed / total" and the result for each definition are shown in a list

Support for Deploying Multiple User Modules

Until now, only a single archive file could be handled in one deployment. You can now deploy multiple archive files together in a single operation. For example, you can deploy and apply a project-shared module and an application module to the staging environment together.

• You can specify additional files to deploy from the Accel CLI command.
• You can now upload multiple files from the "Deployment" tab on the staging detail screen.
  • When you select and upload multiple files, they are combined into a single deployment
  • You can check the list of file names included in that deployment on the deployment detail screen
• When running the tenant environment setup, you can apply the assets of multiple modules together.

Bug Fixes

Support for Referencing Libraries Inside Staging Assets

When assets brought into the staging environment contained JAR libraries, there was a restriction whereby classes inside those libraries could not be resolved from the script development model (server-side JavaScript). We have now redefined the entry point for class resolution for staging assets against the staging scope, so classes in JARs inside the assets can be referenced from screen scripts.

Fixed Retrieval of the List Row Count on SQL Server

Fixed a defect where the row count for the staging asset list could not be retrieved correctly when SQL Server was used as the database.

Limited the Scope of the Screen Assistance Display

The staging environment has a feature that displays annotations while an application screen is open. This automatically embedded a feature for manipulating annotations into the screen, but annotations could end up being added to responses where they were not intended. Specifically, annotations were added even to non-HTML responses (for example, application/json), and parsing on the receiving side could fail. We have now corrected this so that annotations are added only to HTML responses.

• When the response type is not set, it is treated as HTML (ordinary screens have it unset at the point of the check, so excluding "unset" would stop annotations from being added to the screens that genuinely need them)
• Accompanying information such as the character encoding is ignored; the check uses only the type portion

Documentation

Complete Overhaul of the Documentation Structure

We have reorganized the entire documentation into a structure that follows the order in which readers work through it. Previously the guides were listed individually, so reaching the information you wanted meant moving across several guides. We have now reorganized them into purpose-based categories so that you can navigate from "what do I want to know."

• Fundamentals: The concepts of AI-native development and the prerequisite knowledge to have in hand before you start developing
• Environment Setup: Preparing an individual developer's environment and the verification environment shared by the team
• Tutorial: The flow of building business screens and workflows through dialogue with an agent
• Tutorial (Business Process First): The flow of developing starting from a business process diagram
• Development Support Module: The feature for bringing the assets you developed into the verification environment and checking their behavior
• Module Reference: The structure of the deliverables and the specification of each definition file
• DevOps Guidelines: Steps for building a mechanism that automates everything from development to distribution
• Samples: A complete set of specifications for business applications you can use as-is

We have also newly added an introductory chapter explaining the concepts of AI-native development, along with the steps for carrying existing assets created with a different development tool over to the new development environment.

If you are new, we recommend reading in the order "Fundamentals → Environment Setup → Tutorial."

Support for Development Starting from a Business Process Diagram

By simply handing over a diagram that depicts the flow of work (BPMN), you can now build both the specifications for realizing that work and business screens that actually run.

The flow has three stages.

1. From the flow of work depicted in the diagram, create specifications that identify the required screens, the roles of the people involved, and the items requiring further consideration
2. Based on the specifications, create business screens that actually run
3. Write the content you created back into the original diagram, bringing it to a state where it can be run as a business process as-is

As a result, you can use the business flow drawn by the business department directly as the starting point for development. Previously, a step was needed in which a person interpreted the business flow and translated it into specifications.

A concrete example based on opportunity management work is also included.

Please note that the generated specifications are only proposals. Review the content and add to or correct anything that differs from your requirements before using it.

Guidance for Automating the Path from Development to Distribution (DevOps)

We have added steps for building a mechanism that automatically inspects, assembles, and distributes the assets you create.

Because inspection and assembly run automatically every time a change is registered, you notice any problem at that moment. In addition, the latest assets that have passed verification can always be obtained from the same place, which prevents situations such as "I don't know which one is the latest" or "it worked on my machine but not in another environment."

We have prepared steps for each of two widely used platforms (GitHub Actions and GitLab + Jenkins). Organizations already using either one can apply them as-is.

We also introduce an operational example in which simply registering an issue you want addressed has an agent investigate the cause and go as far as submitting a proposed fix. It includes the key points for operating this safely, such as how to set an upper limit on the cost per issue.

• Note: These steps are only one example. You can also adopt only the parts you need, to suit your organization's operational policy.
• Note: We plan to keep updating this documentation.

Added Documentation on Using the @intra-mart/juggling-core Library

We have added an introduction to the @intra-mart/juggling-core library to the documentation, along with an explanation that includes sample code.

Other

• Updated the recommended versions of the prerequisite software needed for development.
• Regarding building the verification environment, we have stated explicitly that using container technology (Docker) is not a requirement. We have also organized the benefits of using it (reducing the effort of setup, sharing environments within the team, and extending to the automation described above).
• Updated the complete set of specifications for the business applications provided as samples (leave request and purchase of goods).

Accel CLI

Added the update Command

We have added a command that brings already-distributed assets up to date. It compares three things — "the content at the time of distribution," "the current content on the developer's machine," and "the latest content" — and decides how to handle each file.

The interactive update previously required answering all of roughly 12 questions every time. It now asks "Do you want to update while keeping your existing settings?" right after it starts, and choosing the default "Yes" lets you skip all the questions. The content review before anything is applied is still shown.

Situation	Behavior
The developer has made no changes	Overwritten with the latest content
The developer has edited it, and there is no change on the latest side	The content on their machine is kept
Both sides have changed the same place	Both versions are written side by side, leaving the decision to the developer

Additional notes:

• You can specify the same configuration items as init. Items you omit keep their current settings.
• In interactive mode, the current values are presented as the initial values, so you can change the configuration at the same time as updating.
• Whether the developer has made edits is determined using the file hash values recorded at distribution time.
• Content changes that accompany a configuration change (such as adding a module to use) are also applied as part of the update.
• Even if places awaiting a decision remain, the process itself is not treated as an abnormal termination (by design, so that work is not halted).
• The source the assets were retrieved from (the location of the distribution source, the branch, and the point in time) is now recorded in the project settings.

Removed the Dependency on an External Command (git)

Previously, version control software (Git) 2.25 or later was required in the user's environment. We implemented the necessary communication processing inside the tool itself and removed this prerequisite. The range of data retrieved (only the files needed) is the same as before, and it retains the characteristic that retrieval cost does not grow as the distribution source grows in size.

• Added support for in-house proxy environments (the various proxy configuration environment variables).
  • It reads the environment variables http_proxy / https_proxy / no_proxy. Differences in case are not distinguished.
• The content comparison processing has also been replaced with an internal implementation. The display format is identical to before.
• Only the version control initialization at project creation time still uses the external Git. In environments where it is not installed, a warning is displayed and processing continues.
• Added an upper limit on the volume of received data, and validation of the information received from the distribution source. This is to prevent malfunctions caused by unexpected data.

Performance Improvements

We improved the way skill assets are downloaded, cutting retrieval time to roughly one third of what it was. Even for large assets, the time spent downloading is greatly reduced.

Improvements to Deployment to iAP

Deploying Multiple Artifacts at Once

To support development setups where shared components are managed separately from the main body, --additional-file now lets you deploy additional artifacts at the same time.

• It can be specified repeatedly. The specified files are validated for existence, format, and duplication (including duplication with the main body), and processing is aborted if there is a problem.
• The default delivery order is "additional files (in the order specified), then the main body last." When paths overlap, the content of the main body takes precedence.
• In interactive mode the order is presented, and you can choose a different one if you want to change it. In automated runs the default order is always used.
• The file list is now displayed using paths relative to the project, making it easier to review.
• We also fixed a defect whereby the extension was applied twice for files of a particular format.

Setup After Deployment

After deployment, following the tenant environment configuration, loading the sample data is now proposed on the spot. It follows the same flow: present the list → confirm → all or selected → show the results. It is proposed regardless of whether the tenant environment configuration was run and regardless of whether it succeeded.

• If there was a fatal failure in the preceding step, it is shown as a warning and processing continues for the remaining modules. The policy is to avoid one failure halting everything.
• Specifying --tenant-setup / --sample-setup runs everything at once without interposing a confirmation (valid in both interactive and automated runs).
• If a specified process fails, it terminates abnormally, but it makes clear that "the delivery itself succeeded." This is so that you are not led to misjudge the cause. When the target configuration does not exist, only a warning is issued and it is treated as a normal termination.

Automatically Placing a Check That Prevents Accidentally Registering Connection Details

When you run the accel login command, the authentication key is written into a configuration file intended for AI development support tools. Registering this into a shared repository by mistake would lead to an external leak, so we made a change that builds in an automatic pre-registration check at the same time as the connection details are saved.

Consideration for existing environments:

• The checking mechanism is placed with a marker indicating that the tool manages it. Files without the marker (ones the developer prepared themselves) are not overwritten.
• Automatic enablement is limited to cases where no other mechanism is configured and no unmanaged file exists. Otherwise a warning is displayed and manual configuration is explained. This is a deliberate decision so as not to silently disable a third-party mechanism you have already adopted.
• The feature can be disabled in the project settings.

Displaying Messages from the Assets to the User

The assets being distributed can now define messages for users that depend on conditions. There are two levels, "notice" and "caution," and you can specify display conditions according to the configuration.

• Because they are collected within existing processing, there is no impact on processing time. Identical content is not displayed more than once.
• Notices about absence can also be displayed, such as "this asset is not provided for this configuration."
• They are displayed after file placement completes and before dependency installation. On an update, they are displayed even when there was nothing to change.
• If a definition contains an error, only that item is skipped, without halting processing.

How to Update from the 2026-07-01 Release Assets

If the accel update command is available, run accel update in the project root. If it is not available, replace the assets with accel detach → accel attach.