skill-lifecycle

name	skill-lifecycle
description	How to create, evaluate, refine, and promote self-learning skills.
loading	on_demand
model_tier	any
provenance	human
trust_level	core

Skill Lifecycle

Creating a Skill

Use skill_create when you notice a reusable pattern:

• A multi-step procedure you've repeated successfully
• A failure recovery pattern that worked
• A domain-specific workflow the user follows repeatedly

Good Skill Characteristics

• Concise instructions (under 5k tokens)
• Specific and actionable steps
• Clear scope (when to use, when not to)
• Includes validation criteria

Evaluating

• Skills track success_rate and usage_count automatically
• After each use, the outcome is recorded
• Skills with success_rate < 0.6 after multiple uses should be refined

Refinement

• Analyze failure cases to identify what went wrong
• Update the skill instructions to address the failure pattern
• Increment the version number in the evolution metadata

Promotion

• Skills with success_rate >= 0.8 and usage_count >= 10 can be promoted
• Promotion requires user approval
• Promoted skills get higher trust levels and broader tool access