skill-creator

Skill Creator

Template-driven skill creation with evaluation criteria.

When to Use

• Creating new domain-specific skills
• Standardizing workflow patterns
• Documenting best practices for reuse

Do NOT Use

• For one-off tasks (use direct implementation)
• When existing skill covers the need

Process

┌─────────────────────────────────────────────────┐
│  1. DEFINE: Identify trigger conditions         │
│  2. DRAFT: Write SKILL.md from template         │
│  3. ADD: Include evaluation criteria            │
│  4. TEST: Validate skill loads correctly        │
└─────────────────────────────────────────────────┘

Directory Structure

.agents/skills/<skill-name>/
  SKILL.md          # Required: skill definition
  examples/         # Optional: usage examples
  templates/        # Optional: file templates
  evals.md          # Optional: evaluation criteria

SKILL.md Template

---
name: <skill-name>
description: "<one-line description of when to use this skill>"
---

# <Skill Name>

## When to Use
- Bullet list of trigger conditions
- Be specific about use cases

## Do NOT Use
- Anti-patterns and edge cases
- When other skills are more appropriate

## Process
Step-by-step instructions with:
1. Numbered steps
2. Code examples where helpful
3. Clear decision points

## Parameters
| Parameter | Default | Description |
|-----------|---------|-------------|
| key | value | explanation |

## Examples
Concrete usage scenarios

## References
Links to related files (@file.md style)

Step-by-Step Creation

1. Define Trigger Conditions

Identify when this skill should activate:

• What user requests match?
• What code patterns trigger it?
• What context is required?

2. Write SKILL.md

Create the file following the template:

mkdir -p .agents/skills/<skill-name>
touch .agents/skills/<skill-name>/SKILL.md

Key sections:

• Frontmatter: name and description (required)
• When to Use: Specific trigger conditions
• Process: Clear, actionable steps
• Examples: Concrete usage patterns

3. Add Evaluation Criteria

If skill produces artifacts, define success criteria:

# Evaluation Criteria

| Criterion | Pass | Fail |
|-----------|------|------|
| Correctness | Output compiles | Syntax errors |
| Completeness | All steps done | Missing steps |
| Quality | Follows patterns | Deviates from style |

4. Test Skill Loading

# Verify frontmatter parses
grep -E "^name:|^description:" .agents/skills/<skill-name>/SKILL.md

# Check file structure
ls -la .agents/skills/<skill-name>/

Frontmatter Requirements

Field	Required	Format
name	Yes	kebab-case, matches directory
description	Yes	One line, describes trigger

Naming Conventions

Pattern	Example	Use Case
<domain>-<action>	rust-development	Domain-specific work
<tool>-<task>	github-ci-guardrails	Tool-specific tasks
swarm-<focus>	swarm-testing-quality	Swarm coordination
self-<action>	self-fix-loop	Automation loops

Example: Creating a Skill

User: "Create a skill for managing database migrations"

Agent: [Skill Creator activated]
1. DEFINE: Triggers on "migration", "schema change", "db upgrade"
2. DRAFT: Created .agents/skills/db-migrations/SKILL.md
3. ADD: Added evals.md with success criteria
4. TEST: Verified frontmatter and structure

Created skill with:
- When to use: Schema changes, version bumps
- Process: 1) Generate migration 2) Test 3) Apply
- Evals: All migrations reversible, tests pass

Quality Checklist

• Frontmatter has name and description
• Name matches directory name
• Clear trigger conditions
• Process has numbered steps
• Examples are concrete
• No duplicate of existing skill