Instruction Creator
This skill guides the creation of effective custom instruction files that help GitHub Copilot follow project conventions and domain-specific logic.
Environment Detection
Before executing any skill operations, determine the active environment and set the skills_source path:
Environment Identification
# Detect environment
if os.name == 'nt': # Windows
# Could be native Windows or VS Code running in Windows
skills_source = os.path.expanduser(r'~\.copilot\skills')
elif 'WSL_DISTRO_NAME' in os.environ or os.path.exists('/mnt/c'):
# Running inside WSL
skills_source = os.path.expanduser('~/.copilot/skills')
else:
# Generic Unix/Linux
skills_source = os.path.expanduser('~/.copilot/skills')
Using skills_source Throughout
Once detected, use skills_source as the base path when referencing skill files:
- Skill Path:
{skills_source}/<skill-name>/SKILL.md - Script Path:
{skills_source}/<skill-name>/scripts/<script>.py - Reference Path:
{skills_source}/<skill-name>/references/<file>.md
Workflow
-
Detect Environment: Determine if running on Windows or WSL and set
skills_source.import os skills_source = (os.path.expanduser(r'~\.copilot\skills') if os.name == 'nt' and 'WSL_DISTRO_NAME' not in os.environ else os.path.expanduser('~/.copilot/skills')) -
Define Scope: Identify the purpose and the files the instructions should apply to (e.g.,
**/*.ts). -
Initialize: Use the initialization script to create the boilerplate.
# Auto-detect environment python3 init_instruction.py "my-new-instruction" \ --description "Description here" \ --applyTo "**/*.ext" # Or specify skills path explicitly python3 init_instruction.py "my-new-instruction" \ --skills-path "~/.copilot/skills" \ -o ./instructions -
Draft Content: Follow the referenced guidelines to fill in the sections.
-
Use Templates: Refer to standard templates for structures in your environment.
-
Semantic Linking: Link to relevant Skills using runtime-resolved paths:
To perform [Task], execute the [Skill Name](SKILL.md) from `{skills_source}/<skill-name>/`. -
Validate: Test the instructions with Copilot to ensure they are clear and actionable.
Core Principles
- Policy Maker: Instructions define the "What" and "How" (decision logic and standards).
- The 4 Sections: Effective instructions include: Overview, Structure, Guidelines, and Resources.
- Concise & focused: Avoid fluff; use imperative language. Limit files to ~500 lines.
- Actionable: Provide concrete "Good" and "Bad" code examples and clear steps.
- Linked: Connect instructions to the skills that execute them and internal automation scripts.
Required Frontmatter
Every instruction file must include YAML frontmatter with the following fields:
---
description: 'Brief description of the instruction purpose and scope'
applyTo: 'glob pattern for target files (e.g., **/*.ts, **/*.py)'
---
Frontmatter Guidelines
- description: Single-quoted string, 1-500 characters, clearly stating the purpose.
- applyTo: Glob pattern(s) specifying which files these instructions apply to.
- Single pattern:
'**/*.ts' - Multiple patterns:
'**/*.ts, **/*.tsx, **/*.js' - Specific files:
'src/**/*.py' - All files:
'**'
- Single pattern:
File Structure
A well-structured instruction file should include these sections as applicable:
- Title and Overview: Clear
#title and short purpose statement. - Workflow: Step-by-step procedures (the “How”).
- Semantic Links: Links to Skills that execute tasks (the “Mechanism”).
- Rules & Constraints: Decision logic and boundaries.
- Best Practices: Recommended patterns and approaches.
- Code Standards: Naming, formatting, style.
- Common Patterns: Frequent implementations.
- Security: Security considerations (if applicable).
- Performance: Optimization guidance (if applicable).
- Testing: Testing standards and verification steps (if applicable).
Writing Style
- Use clear, concise language.
- Write in imperative mood (“Use”, “Implement”, “Avoid”).
- Be specific and actionable.
- Avoid ambiguous terms like “should”, “might”, “possibly”.
- Use bullets and lists for readability.
Examples and Snippets
Provide labeled examples:
### Good Example
```language
// Recommended approach
code example here
```
### Bad Example
```language
// Avoid this pattern
code example here
```
Patterns to Avoid
- Overly verbose explanations.
- Outdated guidance.
- Abstract rules without concrete examples.
- Contradictory advice.
- Copy-paste from docs without added context.
Runtime Variables
When executing this skill, use these variables to reference paths dynamically:
| Variable | Description | Example (Windows) | Example (WSL) |
| --------------- | ----------------------------------- | ----------------------------------- | ----------------------------------- |
| skills_source | Base path to personal skills folder | ~\.copilot\skills | ~/.copilot/skills |
| skill_path | Path to specific skill | {skills_source}\diataxis\SKILL.md | {skills_source}/diataxis/SKILL.md |
Usage in References
## Workflow
1. Analyze requirements.
2. Execute Diataxis Skill from `{skills_source}/diataxis/SKILL.md`.
3. Validate output.
Usage in Scripts
import os
skills_source = os.path.expanduser(r'~\.copilot\skills') if os.name == 'nt' else os.path.expanduser('~/.copilot/skills')
skill_script = os.path.join(skills_source, 'instruction-creator', 'scripts', 'init_instruction.py')
Validation Checklist
- Test with Copilot on real prompts.
- Verify code examples compile or run.
- Confirm
applyTomatches intended files. - Verify
skills_sourcedetection works in target environment.
Maintenance
- Review instructions when dependencies or frameworks change.
- Refresh examples to match current best practices.
- Remove deprecated patterns.
- Keep glob patterns aligned with project structure.
- Update environment detection logic if new platforms are supported.
Resources
- Guidelines: Refer to
references/guidelines.mdin the instruction-creator skill folder. - Templates: Refer to
references/templates.mdin the instruction-creator skill folder. - Scripts: Use
scripts/init_instruction.pyfrom the instruction-creator skill folder.
Note: All references are relative to {skills_source}/instruction-creator/.