| name | skill-creator |
| description | Guide for creating RoleX skills. Use when creating a new skill or updating an existing skill that extends a role's capabilities. Skills are SKILL.md files referenced by a procedure โ the procedure summary is loaded at activate time, the full SKILL.md is loaded on demand via skill(locator). |
Feature: RoleX Skill Creator
Create skills that follow the RoleX capability system.
A skill is a SKILL.md file referenced by a procedure entry via ResourceX locator.
Progressive disclosure: procedure summary at activate โ full SKILL.md via skill(locator) โ execution via use.
Scenario: What is a RoleX skill
Given a role needs operational capabilities beyond its identity
Then a skill is a SKILL.md file containing detailed instructions
And a procedure is a Gherkin summary that references the skill via locator
And the procedure is loaded at activate time โ the role knows what skills exist
And the full SKILL.md is loaded on demand via the skill tool
Scenario: Skill directory structure
Given a skill lives under the skills/ directory
Then the structure is:
"""
skills//
โโโ SKILL.md (required โ detailed instructions in Gherkin)
โโโ resource.json (required โ name, type: "skill", tag)
โโโ references/ (optional โ loaded on demand)
โโโ .md
"""
And SKILL.md has YAML frontmatter with name and description
And the body uses Gherkin Feature/Scenario format
And references contain domain-specific details loaded only when needed
Scenario: The procedure-skill contract
Given a procedure is trained to a role
Then the Feature description contains the ResourceX locator for the skill
And the Feature body summarizes what the skill can do
And example procedure:
"""
Feature: Role Management
https://github.com/Deepractice/DeepracticeX/tree/main/skills/role-management
Scenario: What this skill does
Given I need to manage role lifecycle
Then I can born, teach, train, retire, and die roles
"""
And the description line is the locator โ the skill tool resolves it via ResourceX
Feature: Skill Creation Process
How to create a new RoleX skill step by step.
Scenario: Step 1 โ Understand the scope
Given you are creating a skill for a specific domain
When you analyze what operations the skill should cover
Then identify the concrete commands, parameters, and workflows
And determine what the role needs to know vs what it already knows
And only include information the role cannot infer on its own
Scenario: Step 2 โ Write SKILL.md
Given you understand the scope
When you write the SKILL.md
Then start with YAML frontmatter (name + description)
And write the body as Gherkin Features and Scenarios
And each Feature covers a logical group of operations
And each Scenario describes a specific workflow or decision point
And use Given/When/Then for step-by-step procedures
And keep SKILL.md under 500 lines โ split into references if needed
Scenario: Step 3 โ Create resource.json
Given SKILL.md is written
When you create the resource manifest
Then include name, type "skill", and tag version
And example:
"""
{
"name": "my-skill",
"type": "skill",
"tag": "0.1.0"
}
"""
Scenario: Step 4 โ Train the procedure
Given the skill directory is ready
When you train the procedure to a role
Then use the train operation with Gherkin source
And the Feature description MUST be the ResourceX locator for the skill
And the Feature body summarizes capabilities for activate-time awareness
And example:
"""
locator: "!individual.train"
args:
individual: "sean"
id: "my-skill"
content: "Feature: My Skill\n https://github.com/org/repo/tree/main/skills/my-skill\n\n Scenario: When to use\n Given I need to do X\n Then load this skill"
"""
Scenario: Step 5 โ Test the skill
Given the procedure is trained
When you activate the role
Then the procedure summary should appear in the activation output
And calling skill with the locator should load the full SKILL.md
And the loaded content should be actionable and complete
Feature: SKILL.md Writing Guidelines
Rules for writing effective skill content in Gherkin.
Scenario: Context window is a shared resource
Given the SKILL.md is loaded into the AI's context window
Then keep content concise โ only include what the role cannot infer
And prefer concrete examples over verbose explanations
And use Gherkin structure to organize โ not to pad length
Scenario: Gherkin as instruction format
Given RoleX uses Gherkin as its universal language
When writing SKILL.md body content
Then use Feature for logical groups of related operations
And use Scenario for specific workflows or procedures
And use Given for preconditions and context
And use When for actions and triggers
And use Then for expected outcomes and next steps
And use doc strings (triple quotes) for command examples and templates
Scenario: Progressive disclosure within a skill
Given a skill may cover many operations
When some details are only needed in specific situations
Then keep core workflows in SKILL.md
And move detailed reference material to references/ files
And reference them from SKILL.md with clear "when to read" guidance
Scenario: Frontmatter requirements
Given the frontmatter is the triggering mechanism
Then name must match the procedure name
And description must explain what the skill does AND when to use it
And do not include other fields in frontmatter
