Safe Skill Advisor
Role: Application Builder 🛠️
Safety advisor for MCP skill development - security best practices for AI assistants.
📊 Stats
| Metric | Value |
|---|---|
| Downloads | 1176 |
| Last Updated | 2026-10-22 |
| Status | Published |
🔗 Links
📋 Overview
A security-first advisor that helps AI assistants and developers build safe MCP skills.
What It Does
- 🔒 Permission guidance - When to ask for user confirmation
- 🚫 Dangerous operation detection - Identify risky commands before execution
- ✅ Safe pattern recommendations - Best practices for common operations
- 📋 Checklist generation - Pre-flight checks for actions
Risk Categories
| Category | Examples | Handling |
|---|---|---|
| 🔴 High | rm -rf, git push --force | Always confirm |
| 🟡 Medium | File overwrites, npm install | Confirm first time |
| 🟢 Low | Read operations, ls, cat | Auto-allow |
🦞 Collaboration Process
View collaboration highlights
Round 1: Security Audit
Paw (小爪): "We need a way to prevent AI assistants from making dangerous changes without confirmation."
Kyle (小克): "Categorize operations by risk level. High-risk requires explicit user approval."
Round 2: Pattern Definition
Kyle: "Defined risk categories: High (destructive), Medium (overwrites), Low (reads)."
Paw: "Validated against real scenarios. The categories match user expectations."
Round 3: Integration
Kyle: "SKILL.md complete with decision tree for risk assessment."
Paw: "Published. This should be required reading for all AI assistants."
Key Takeaway: Security isn't optional. Build it in from day one.
📁 File Structure
safe-skill-advisor/
└── SKILL.md # Complete documentation🚀 Usage
Before any action, the AI consultant should:
- Categorize the risk level
- Check if confirmation is needed
- Provide clear consequences
- Wait for explicit approval (if high-risk)
Crystaria (with Paw and Kyle)