User Config Manager

User Config Manager

Audit, organize, and optimize the user's global VS Code AI configuration ecosystem.

Scope

This skill manages user-level AI customization files.

VS Code user data folder:

• Windows: %APPDATA%/Code/User/
• macOS: ~/Library/Application Support/Code/User/
• Linux: ~/.config/Code/User/

Note: Workspace-level equivalents exist in .github/agents/, .github/instructions/, and .vscode/mcp.json.

Deployed Alex Portable Configuration

The following Alex-derived content is deployed user-wide (April 2026):

Portable Instructions (~/.copilot/instructions/)

User-Level Agents (~/.copilot/agents/)

These provide Alex-like behavior in all workspaces.

Triggers

• User asks to review or organize their Copilot configuration
• Before major Alex brain updates (avoid conflicts)
• When user reports inconsistent AI behavior across workspaces
• Quarterly maintenance alongside memory curation

Audit Procedure

Step 1: Inventory User Config

Read all user config files:

# User memories
memory view /memories/

# Custom agents (list globalStorage directory)
# Check for *-agent/ directories

# User prompts
# List %APPDATA%/Code/User/prompts/

# MCP config
# Read %APPDATA%/Code/User/mcp.json

Step 2: Memory Scope Check

For each memory file, verify content belongs at user level:

3-Workspace Test: Content must be useful in 3+ different workspaces.

Step 3: Agent Analysis

For each custom agent, check:

1. Tool scope: Are tools appropriately constrained?
2. Model selection: Is specific model justified?
3. Handoff patterns: Do handoffs make sense?
4. Overlap: Does agent overlap with workspace agents?

Step 4: Instruction Conflict Detection

Compare user instructions (prompts/*.instructions.md) with workspace instructions:

Step 5: MCP Server Audit

For each MCP server in mcp.json:

1. Reachability: Can the server be contacted?
2. Usage: Is the server actively used?
3. Overlap: Do workspace MCP servers duplicate functionality?
4. Security: Are credentials properly managed?

Step 6: Generate Report

# User Config Health Report

**Date**: [Date]
**User**: [Name from user-info.md]

## Summary

| Component | Files | Issues | Health |
|-----------|-------|--------|--------|
| Memories | N | N | 🟢/🟡/🔴 |
| Agents | N | N | 🟢/🟡/🔴 |
| Instructions | N | N | 🟢/🟡/🔴 |
| MCP Servers | N | N | 🟢/🟡/🔴 |

## Issues Found

### Critical
- [List critical issues]

### Warnings
- [List warnings]

### Recommendations
- [List optimization suggestions]

## Actions Taken
- [List any auto-fixes applied]

Templates

Custom Agent Template

---
name: AgentName
description: What this agent specializes in
argument-hint: What users should provide when invoking
model: claude-sonnet-4
target: vscode
user-invocable: true
disable-model-invocation: false
tools: ['search', 'read', 'web', 'vscode/memory']
agents: []
handoffs: []
hooks: {}
---

You are [AGENT NAME], a specialized assistant that [PURPOSE].

## Core Responsibilities

- Responsibility 1
- Responsibility 2

## Rules

- Rule 1: Never do X
- Rule 2: Always do Y

Use #tool:web/fetch to reference tools inline.

## Output Format

[Describe expected output structure]

User Memory Structure Template

# User Memory

## Identity
- Name, role, and professional context
- Communication preferences (formality, length)

## Writing Style
- Formatting preferences (em dashes, lists)
- Terminology preferences

## Tool Preferences
- CLI tool preferences (npm vs pnpm)
- Diagram tool preferences (Mermaid settings)

## Workflow
- Git habits (commit frequency, branch naming)
- Review preferences

## Cross-Workspace Access
- MCP server usage patterns
- Authentication patterns

Tool Set Template

{
  "tool-set-name": {
    "tools": ["tool1", "tool2", "tool3"],
    "description": "Brief description of what these tools do together",
    "icon": "appropriate-icon-name"
  }
}

Integration with Other Skills

Scope Boundaries

This skill does NOT manage:

• Workspace-level .github/ brain files (that's brain-qa)
• VS Code settings.json (that's user preference, not AI config)
• GitHub Copilot Enterprise settings (org-managed)