Create Skill

Creating a New Cortex Developer Skill

This skill guides the end-to-end creation, overlap evaluation, authoring, validation, and ecosystem registration of new AI agent skills in Cortex Framework V7, or the creation of custom instruction extensions inside existing skill custom/ directories.

CRITICAL OPERATIONAL RULE: You are an end-to-end deployment agent, not just a text generator. A skill creation task is NOT complete until the mandatory Skill Anatomy Validation Gate has been executed with a 100% pass rate. You MUST NOT stop after creating the files.

CRITICAL ISOLATION RULE (ZERO-TRUST FOR CORE PLATFORM): New skills must reside exclusively inside .agents/skills/<skill_name>/. Never modify platform core runtime code (cortex-framework-core/src/common/) while authoring skills unless explicitly instructed by the user.

Step 1: Overlap Analysis & Pathway Selection

Before creating a new skill, evaluate existing skills to prevent redundancy and fragmentation.

1. Perform Skill Overlap Assessment

Dynamically inspect the existing skills in .agents/skills/ (and the registry in .agents/AGENTS.md). Read their frontmatter descriptions to evaluate whether the requested capability overlaps with an existing workflow.

2. Select Pathway: Custom Extension vs. Standalone Skill

Refer to skill_anatomy_guide.md and authoring_best_practices.md for the decision matrix:

• Path A (Custom Extension via custom/): If the proposed capability extends or customizes an existing lifecycle phase (e.g. organization-specific naming standards, custom test assertions, compliance gates, specialized dataset targets), DO NOT create a new skill. Instead, proceed to Step 2A to create custom instructions in <existing_skill>/custom/.
• Path B (New Standalone Skill): If the capability introduces a genuinely new domain or distinct lifecycle phase not covered by any existing skill, proceed to Step 2B to scaffold and author a new standalone skill.

Step 2A: Author Custom Extension in custom/ (Path A)

1. Identify target existing skill (e.g. create_data_product, validate_data_product).
2. Scaffold custom instructions using the script:

   python3 external-skills/.agents/skills/create_skill/scripts/scaffold_skill.py --custom-for <existing_skill_name> --custom-filename custom_instructions.md
   

3. Author the custom rules using custom_instruction_template.md.
4. Verify that the file resides in .agents/skills/<existing_skill>/custom/.

Step 2B: Scaffold & Author Standalone Skill (Path B)

1. Scaffold Directory Structure:

   python3 external-skills/.agents/skills/create_skill/scripts/scaffold_skill.py <skill_name> --description "<Brief description under 1024 chars>"
   

2. Author SKILL.md: Use the template in skill_template.md.
   • YAML Frontmatter: Include name (matching directory) and description ($\le 1024$ chars).
   • Operational Rules: Define non-negotiable trust boundaries and quality gates at the top.
   • Numbered Steps: Clearly structure steps (Planning, Implementation, Validation, Reporting).
   • Link Integrity: Ensure every relative link points to a real existing file.
3. Add Supporting Assets & References:
   • Place templates in assets/.
   • Place detailed reference guides in references/.
   • Place CLI scripts in scripts/.
4. Ensure custom/.keep exists in the new skill folder.

Step 3: Ecosystem Registration (Standalone Skills)

Register the new skill across the workspace:

1. Register in external-skills/.agents/AGENTS.md: Add an entry under Lifecycle Slash Commands mapping the slash command to the skill file.

2. Register in external-skills/.agents/skills/using_cortex_skills/SKILL.md: Update the discovery decision tree and table to include the new skill.

3. Document in external-skills/README.md: Add the skill and command to the index tables.

Step 4: The Validation Quality Gate (MANDATORY)

DO NOT ASK FOR PERMISSION. Immediately execute the anatomy validator:

uv run python3 tests/unit/validate_skills.py

Verify that:

• The validator reports PASSED with 0 errors and 0 warnings.
• Frontmatter name and description are valid.
• All relative links resolve cleanly.
• If unit tests exist, execute uv run pytest tests/unit.

Step 5: Generate Completion Report

Generate a final completion report using skill_creation_report_template.md and present the summary to the user.