User Approval

Kyew requires explicit user approval before skills become active. This ensures you maintain control over what Claude learns from your interactions.

Why Approval Matters

Without approval requirements, the system could:

• Learn incorrect patterns from failed attempts
• Codify temporary workarounds as permanent practices
• Create skills that conflict with your preferences
• Auto-apply guidance you haven't verified

With approval:

• You review all generated knowledge
• Only validated skills become active
• You maintain editorial control
• Skills reflect your actual preferences

The Approval Workflow

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Draft     │────▶│   Review    │────▶│   Active    │
│             │     │             │     │             │
│ (Generated) │     │ (Pending)   │     │ (Approved)  │
└─────────────┘     └─────────────┘     └─────────────┘
                           │
                           ▼
                    ┌─────────────┐
                    │  Rejected   │
                    │             │
                    │ (With note) │
                    └─────────────┘

1. Generation (Draft)

When you generate a skill:

"generate skill from CORS debugging memories"

The skill is created with status: draft. It won't be suggested.

2. Review (Pending)

View pending skills:

"show pending skills"

Review the content:

• Are the instructions accurate?
• Is the scope appropriate?
• Would this help in future similar situations?

3. Approve or Reject

Approve to activate:

"approve skill skill-abc123"
"approve the CORS debugging skill with note 'verified against docs'"

Reject if not useful:

"reject skill skill-abc123 because instructions are too vague"

4. Active

Approved skills:

• Are suggested when context matches
• Track success/failure outcomes
• Can be deprecated later if outdated

Review Guidelines

Check Accuracy

• Are the instructions technically correct?
• Do they match current best practices?
• Are there any outdated references?

Verify Usefulness

• Would you actually use this guidance?
• Is it specific enough to be actionable?
• Does it cover the right scenarios?

Consider Scope

• Is the skill too broad or too narrow?
• Does it overlap with existing skills?
• Are the triggers appropriate?

Managing Approved Skills

Deprecation

When skills become outdated:

"deprecate skill skill-abc123 because the API changed"

Deprecated skills are no longer suggested.

Updates

Edit skills that need corrections:

"update skill skill-abc123 to fix the typo in step 3"

Updates create new versions, preserving history.

Rollback

If an update caused problems:

"rollback skill skill-abc123 to version 1"

Code Tool Approval

Code tools have an additional approval requirement because they execute custom JavaScript:

1. Create code tool (draft status)
2. Review the code for security
3. Approve to enable execution

This prevents malicious or buggy code from running automatically.