Skip to main content

Creating Skills

Folder: .github/skills/tsh-creating-skills/ Used by: Copilot Engineer

Provides naming conventions (gerund form), description guidelines, body structure, progressive disclosure patterns, templates, and validation checklists for creating new skills (SKILL.md).

Core Design Principles​

  • Conciseness — The context window is a shared resource. Only add context the LLM doesn't already have.
  • Progressive disclosure — Discovery tier (~100 tokens): name + description. Activation tier (<5000 tokens): full SKILL.md body. Resource tier (on demand): templates, examples.
  • Separation of concerns — A skill defines HOW to perform a task. It must NOT define WHO the agent is (agent territory) or WHAT triggers the workflow (prompt territory).

Naming Conventions​

  • Skill directories use gerund form: tsh-<gerund-subject>/ (e.g., tsh-code-reviewing/, tsh-creating-prompts/)
  • The name field in SKILL.md frontmatter must match the directory name
  • The tsh- prefix identifies artifacts from The Software House's copilot-collections

Skill Folder Structure​

tsh-<gerund-subject>/
├── SKILL.md            # Main skill file (required)
├── <name>.example.md   # Example output (optional)
└── <name>.template.md  # Output template (optional)

Validation Checklist​

  • Directory name uses gerund form with tsh- prefix
  • name field in frontmatter matches directory name
  • Description is concise (1–2 sentences)
  • Body stays under 5000 tokens
  • No personality or behavioral content (agent territory)
  • No project-specific conventions (instructions territory)

Connected Skills​

  • tsh-creating-agents — For creating agents that load this skill.
  • tsh-creating-prompts — For creating prompts that trigger workflows using this skill.