AI Tuning Skill

Optimize GitHub Copilot, Claude Code, and MCP configurations.

Trigger Phrases

•"improve my CLAUDE.md"
•"better copilot instructions"
•"tune AI for this project"
•"add MCP servers"
•"optimize AI prompts"

Effective AI Instructions Principles

•Be Specific: Vague -> vague results
•Show Examples: Code > descriptions
•State Constraints: What NOT to do
•Organize Hierarchically: General -> specific
•Include Commands: Quick reference

CLAUDE.md Structure

markdown

# CLAUDE.md

## Project Overview
[What, architecture, technologies]

## Project Structure

[directory tree]

code


## Build Commands
```bash
# Install
[command]

# Test
[command]

# Lint
[command]

Code Style Requirements

[Formatter, linter, key rules with examples]

Architecture Guidelines

[Patterns, layer rules]

Important Patterns

[Code examples]

code


## copilot-instructions.md Structure

```markdown
# GitHub Copilot Instructions

## Project Context
[Brief description, tech stack]

## Code Generation Guidelines

### [Language] Patterns
[Conventions, type annotations, imports]

### Examples
```[language]
// GOOD
[example]

// AVOID
[counter-example]

Common Patterns

[Reusable code]

Commands

[Quick reference]

code


## MCP Configuration

```json
{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@context7/mcp-server"]
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "${workspaceFolder}"]
    },
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}

Optimization Techniques

1. Context Density

markdown

# Instead of verbose:
"We use Python 3.12 as the language.
We use ruff for linting.
Testing is done with pytest."

# Write dense:
"Python 3.12 | ruff (lint+format) | pytest"

2. Example-Driven

markdown

# Instead of:
"Use type annotations."

# Write:
"Type annotations required:
```python
def process(items: list[str]) -> dict[str, int]: ...
```"

3. Constraints Section

markdown

## Constraints
- Max line length: 100 chars
- No star imports
- Error messages assigned to variables
- All public functions need docstrings

4. Command Quick Reference

markdown

| Action | Command |
|--------|---------|
| Test | `uv run pytest` |
| Lint | `uv run ruff check .` |
| Format | `uv run ruff format .` |

Validation

bash

# Check AI files exist
[ -f "CLAUDE.md" ] && echo "OK CLAUDE.md"
[ -f ".github/copilot-instructions.md" ] && echo "OK Copilot"
[ -f ".vscode/mcp.json" ] && echo "OK MCP"

# Check sections in CLAUDE.md
grep "^## " CLAUDE.md

# Check code examples
grep -c '```' CLAUDE.md

Quality Metrics

Metric	Target
Has examples	Yes (3+ code blocks)
Has commands	Yes
Organized	Yes (## headers)
Specific	No vague terms
Current	Tool versions updated

Pattern Library Development

Include reusable patterns:

python

# Pattern: Error Handling
def fetch(id: str) -> User:
    """Fetch user by ID.

    Raises:
        UserNotFoundError: If not found.
    """
    result = db.query(User).filter_by(id=id).first()
    if result is None:
        raise UserNotFoundError(f"User {id} not found")
    return result

AI File Audit

bash

echo "=== AI Configuration Audit ==="

for f in CLAUDE.md .github/copilot-instructions.md .vscode/mcp.json; do
  if [ -f "$f" ]; then
    echo "OK $f exists ($(wc -l < "$f") lines)"
  else
    echo "MISSING $f"
  fi
done