skill-creation

name	skill-creation
description	Use when you've identified a recurring pattern worth abstracting into a skill
trigger_pattern	When 2+ similar processes exist or a recurring pattern is identified
tags	["meta","creation","induction"]
version	1
is_protected	true
q_value	0.5
q_update_count	0

Skill Creation

Iron Law

NO SKILL WITHOUT 2+ SIMILAR PROCESSES.

A skill must be abstracted from concrete evidence, never from a single instance or pure speculation.

When to Create a Skill

Decision Tree:
├── Found 2+ similar Processes in Neo4j?
│   └── YES → Candidate for skill creation
├── Same pattern applied successfully 2+ times?
│   └── YES → Candidate for skill creation
├── Only 1 instance of the pattern?
│   └── NO → Wait for more evidence
├── Pattern is project-specific?
│   └── NO → Only if generalizable beyond the project
└── Pattern is a one-time debugging step?
    └── NO → Debugging steps are NOT skills

Creation Process

Step 1: Gather Evidence

Use ProcessSearchTool to find similar processes:

processes = search_processes("trigger pattern description")

Require:

• At least 2 processes with similar triggers
• Processes should be generalizable (is_generalizable=true)
• Processes should have positive outcomes

Step 2: Abstract the Pattern

From the concrete processes, extract:

1. Generalized trigger: What situation triggers this skill?
2. Action template: Step-by-step procedure (abstracted from specifics)
3. Expected outcome: What should happen when applied correctly?
4. Boundary conditions: When does this skill NOT apply?

Step 3: Quality Checklist

Before creating, verify:

• Pattern is generalizable (works beyond original context)
• At least 2 source processes
• Clear trigger condition (not vague)
• Actionable steps (not just observations)
• Expected outcome defined
• No sensitive information (passwords, API keys)
• Name follows kebab-case convention
• Description starts with "Use when..."

Step 4: Create

Use SkillCreateTool with:

• name: kebab-case, descriptive (e.g., "debug-memory-leak")
• description: "Use when [trigger condition]"
• content: Markdown with steps, examples, anti-patterns
• tags: 2-5 relevant tags
• source_process_ids: IDs of the source processes

Naming Conventions

• Use kebab-case: debug-memory-leak, NOT debugMemoryLeak
• Start with action verb: debug-, extract-, optimize-, handle-
• Be specific: debug-memory-leak, NOT debug-issue
• Keep it short: 2-4 words

Content Structure

# Skill Name

## When to Use
[Clear trigger condition]

## Steps
1. [First step]
2. [Second step]
3. [...]

## Examples
[Concrete example of applying this skill]

## Anti-patterns
[What NOT to do]

## Boundary Conditions
[When this skill does NOT apply]

Anti-patterns

• Creating skills from single instances
• Creating skills that are too vague ("be careful")
• Creating skills that are too specific ("fix line 234 in file X")
• Creating skills without source process provenance
• Creating duplicate skills (search first!)