Skill Sets
In AI-native development with an AI agent, intra-mart Accel Platform-specific rules and knowledge are built into the project in advance, in the form of rules files and skill sets, for the AI agent to reference.
Skill Set Structure
When you create a project with Accel CLI, a collection of skill sets is placed in the project root.
When automatically generating iAP assets through AI-native development, the AI agent automatically loads and uses the skill sets it needs.
<project root>
├─ .claude
│ ├─ rules - Rules files (Claude Code edition)
│ └─ skills - Skill sets (Claude Code edition)
├─ .codex - Configuration files for Codex
├─ .agents
│ ├─ rules - Rules files (Codex edition)
│ └─ skills - Skill sets (Codex edition)
└─ .github
├─ instructions - Rules files (GitHub Copilot edition)
└─ skills - Skill sets (GitHub Copilot edition)
Rules Files (rules / instructions)
In AI-native development, the AI agent may implement code following general JavaScript/HTML rules due to its nature. However, iAP's Script Development Model (JSSP) has its own implementation rules, so the general rules alone may not produce working code.
By having the AI agent reference rules files that describe these platform-specific implementation rules, the generated artifacts are guided to follow the correct rules.
| File | Key Content |
|---|---|
jssp-overview.md | Overview of JSSP structure (.js + .html pairs) and processing sequence |
jssp-file-structure.md | Output destination definitions for artifacts and routing table mapping definitions |
jssp-code-style.md | Rhino usage notes and coding conventions for function container implementation |
jssp-naming.md | Naming conventions for variables, constants, and functions |
jssp-function-container.md | Assembly rules and patterns for function container implementation |
jssp-presentation-page.md | Assembly rules and patterns for presentation page implementation |
jssp-error-handling.md | Error handling definitions for exception scenarios |
jssp-security.md | Coding conventions for preventing vulnerabilities such as XSS and CSRF |
jssp-logging.md | Definitions for log and console output |
jssp-performance.md | Definitions for development environment tuning |
jssp-2way-sql.md | Definitions for 2WaySQL and database transactions |
jssp-testing.md | Definitions for unit testing |
jssp-accessibility.md | Definitions for accessibility support |
Skill Set List
When using AI-native development with an AI agent, platform-specific knowledge that is not part of general training data is often missing, leading to speculative implementations that may not work on iAP.
The skill sets document knowledge and implementation techniques for features specific to iAP. By having the AI agent reference these skills, it is guided to correctly use intra-mart's proprietary features.
Code Generation
The "Code Generation" skill sets define the knowledge and implementation techniques for generating assets that actually run on intra-mart, such as business screens and IM-Workflow.
By selecting the appropriate skill based on the type of artifacts needed, you can reduce omissions in the initial implementation.
| Skill | Purpose | Example Trigger Keywords |
|---|---|---|
jssp-page-generator | Generate new business screens (.html + .js) and routing XML | "Create a XX screen", "List screen", "Implement a form" |
jssp-im-job-generator | Generate job scheduler batch processes (.js) | "Create a batch process", "Scheduled execution", "Nightly batch" |
jssp-im-workflow-usage | Generate IM-Workflow integration programs (application/approval screens, action processing, etc.) | "Application screen", "Approval screen", "Action processing", "Case initiation" |
base-im-workflow-generator | Generate IM-Workflow import XML definitions (content, routes, flows, etc.) | "Create a workflow definition", "WF import XML" |
jssp-im-cache-usage | Cache the results of lookups or expensive computations in a JSSP function container using the Cache API | "Cache this", "Speed up this calculation" |
jssp-im-logic-generator | Generate IM-LogicDesigner import ZIP definitions (logic flows, routing) | "Create a logic flow", "IM-LogicDesigner flow" |
jssp-im-logic-usage | Generate code that calls an existing IM-LogicDesigner routing definition (logic flow API) from a JSSP presentation page | "I want to call a logic flow API", "I want to call logic/api", "I want to use an existing routing definition" |
jssp-im-oauth-generator | Bulk-generate OAuth-authenticated REST-API assets (scope / resource URL / client detail XML + JSSP) using the im_oauth provider feature | "Create an OAuth REST-API", "Define OAuth scopes", "Add oauth-client-resources-config" |
jssp-im-master-usage | Generate implementation code for imACMSearch dialogs (user, organization, company, etc.) | "Add user search", "Organization selection dialog", "Master search" |
jssp-im-contents-search-generator | Generate custom IM-ContentsSearch crawlers (jobs that register/delete Solr indexes) and custom content display templates | "Create a crawler", "Register in Solr", "Add content search" |
jssp-highcharts-usage | Chart rendering using Highcharts | "Display a chart", "Draw a graph", "Line chart", "Pie chart" |
jssp-imds-theme | Usage of imds-compliant HTML components (tables, buttons, dialogs, etc.) | "Create a table", "Implement a dialog", "Form UI" |
Localization
The "Localization" skill sets define the knowledge and implementation techniques for adding multi-language support to business screens, jobs, and other implementations.
You can localize artifacts from the start or after the business screen or job is largely complete.
| Skill | Purpose | Example Trigger Keywords |
|---|---|---|
jssp-localize-support | Externalize hardcoded strings to message properties (ja/en/zh_CN) | "Localize", "i18n", "Add multi-language support" |
Verification
The "Verification" skill sets define the knowledge and implementation techniques for checking the quality of generated artifacts.
| Skill | Purpose | Example Trigger Keywords |
|---|---|---|
jssp-code-review | Automated code review | "Run a code review", "Show review results" |
jssp-jest-test | Create unit tests for function containers (.js) using jest-on-rhino | "Write tests", "Add unit tests", "Use a mock" |
jssp-page-verifier | Verification of generated screens | "Verify the screen", "Check screen behavior" |
jssp-playwright-test | E2E testing with Playwright | "Run E2E tests", "Playwright test" |
jssp-security-check | Running security checks | "Run a security check", "Check for vulnerabilities" |
test-spec-generator | Generate a test perspective list and test case sheets (xlsx/HTML) from the specifications (under spec/) | "Create test case sheets", "Create a test perspective list" |
Production Setup
The "Production Setup" skill sets define the knowledge and implementation techniques for generating the tenant environment setup assets that accompany a user module.
Setup assets for roles, authorization, menus, jobs, and DDL/DML are generated with multi-language support (ja/en/zh_CN).
| Skill | Purpose | Example Trigger Keywords |
|---|---|---|
jssp-tenant-setup-generator | Bulk-generate a full set of tenant environment setup assets (config XML, roles, authorization, menus, jobs, DDL/DML, extension importers) | "Create tenant environment setup assets", "Create Importer import assets", "Create setup XML" |
jssp-sample-setup-generator | Bulk-generate a full set of sample data setup assets (config XML, roles, authorization, menus, jobs, DDL/DML, extension importers) | "Create sample data setup assets", "Create assets for loading sample data", "Create sample data for a trial environment" |
Menus: Only "Global Navigation (PC edition)" is generated. Authorization: Only "Screen/Process" and "IM-LogicDesigner REST API" are generated.
Skill Set Customization
Existing skill sets can be improved as you operate them. You can write back knowledge gained from fixing issues, or add new project-specific skills.
Sharing Tacit Knowledge
When the AI agent successfully corrects artifacts through the work described in AI-native Development Workflow's Improving Accuracy section, it is recommended to save the prompt used at the time.
When multiple developers share the same project, other developers may encounter the same issues in generated artifacts. Sharing the prompt used for the fix can help prevent the same issue from occurring again.
After correcting artifacts, you can also enter the following prompt before clearing the context to save the fix approach and implementation notes within the project.
To prevent the bug I just fixed from occurring again, please save the knowledge to SKILL.md.
By saving this knowledge, you can share it with other team members when the skill sets are under version control, and continue to use it even when the development environment changes.
Creating Custom Skill Sets
When you want the AI agent to acquire knowledge not covered by existing skill sets — such as project-specific libraries or company-wide common APIs — you can create custom skill sets.
Skill sets are created as Markdown files. By placing the files in the designated directory, the AI agent automatically references them in response to keywords in prompts.
<project root>
├─ .claude
│ └─ skills
│ └─ my-custom-skill - Add custom skill (Claude Code edition)
│ └─ SKILL.md - Write the skill contents
├─ .agents
│ └─ skills
│ └─ my-custom-skill - Add custom skill (OpenAI Codex edition)
│ └─ SKILL.md - Write the skill contents
└─ .github
└─ skills
└─ my-custom-skill - Add custom skill (GitHub Copilot edition)
└─ SKILL.md - Write the skill contents
Skill files should describe instructions, implementation examples, and notes for the AI agent. The following is a structural example of a skill file.
---
name: my-custom-skill
description: A description that tells the AI agent what this skill does and what kinds of prompts (keywords) should trigger it. Write it in a form such as "Use when 'create XX' or 'write a sample of XX' is mentioned."
allowed-tools: Bash, Read, Write, Glob
---
# my-custom-skill
## Overview
Describe the purpose and use of this skill.
## Usage
Describe specific implementation steps or how to call APIs.
## Implementation Example
Provide code samples.
## Notes
Describe implementation constraints and cautions.
The YAML frontmatter (the block enclosed by --- at the top) holds the skill metadata.
The AI agent scans only the description of each skill — without reading the body — and matches it against the user's prompt to decide whether to trigger the skill.
Writing the description so it clearly states "what the skill does" and "what keywords should trigger it" makes the skill more likely to be invoked at the intended moment.
| Field | Role |
|---|---|
name | Skill identifier. Should match the directory/file name (kebab-case) |
description | The skill's description and trigger keywords. Used by the AI agent to decide whether to select this skill |
allowed-tools | Tools that the skill is permitted to use while running (Claude Code edition only; optional) |
By including created skills in version control, all team members can provide the same knowledge to the AI agent.