snowflake-labs/skill-optimiser

Skill Optimiser

Comprehensive guide for creating, auditing, and optimizing skills following official Anthropic best practices. This skill encodes all rules, patterns, and checklists from "The Complete Guide to Building Skills for Claude" (33-page PDF, January 2026).

Important

Before making any changes to a skill, READ the existing SKILL.md fully. Understand its current structure before applying optimizations.

Workflow

1. Determine the Task

Identify which operation the user needs:

• Create: Build a new skill from scratch
• Audit: Review an existing skill for compliance with best practices
• Optimize: Improve an existing skill's triggering, structure, or instructions
• Fix: Address a specific issue (undertriggering, overtriggering, instructions not followed)

2. Gather Context

• Read the target skill's SKILL.md (if it exists)
• Identify the skill category (see references/patterns.md > Skill Categories)
• List all files in the skill folder to understand its structure

3. Apply Best Practices

Consult the reference files for detailed guidance:

• references/structure-rules.md for file structure, naming, and YAML frontmatter rules
• references/description-guide.md for writing effective descriptions and trigger phrases
• references/instructions-guide.md for writing effective SKILL.md body content
• references/patterns.md for skill categories and implementation patterns
• references/testing-checklist.md for the complete validation checklist
• references/troubleshooting.md for diagnosing and fixing common issues

4. Validate Changes

After making changes, run through the quick checklist in references/testing-checklist.md to ensure compliance.

Audit Procedure

When auditing an existing skill, check these in order:

1. Folder name: kebab-case, no spaces/capitals/underscores
2. File name: Exactly SKILL.md (case-sensitive)
3. No README.md inside the skill folder
4. YAML frontmatter: Has --- delimiters, name matches folder, description includes WHAT + WHEN + triggers
5. No XML angle brackets in frontmatter
6. Description under 1024 characters
7. SKILL.md body under 5,000 words (move excess to references/)
8. Instructions are specific and actionable (no vague "validate things properly")
9. Error handling included for common failure modes
10. Examples provided for key workflows
11. Progressive disclosure: detailed docs in references/, not inline

Quick Reference: Description Formula

[What it does] + [When to use it] + [Key capabilities/triggers] + [Negative triggers]

Good example:

description: "Deploy the Route Deviation Analysis demo: create projection views
from SYNTHETIC_DATASETS.UNIFIED, run 3-step ETL pipeline, and register React
dashboard pages. Use when: setting up route deviation demo, detour analytics,
fleet deviation analysis. Do NOT use for: general fleet tracking, real-time GPS
monitoring, or non-deviation routing tasks. Triggers: deploy route deviation,
deploy detour analytics, setup deviation analysis, route deviation demo."

Bad examples:

• "Helps with projects." (too vague)
• "Creates sophisticated multi-page documentation systems." (missing triggers)
• "Implements the Project entity model with hierarchical relationships." (too technical, no user triggers)

Quick Reference: SKILL.md Body Template

# Skill Title

Brief 1-2 sentence summary of what this skill does end-to-end.

## Prerequisites / Important

CRITICAL items that must be verified before starting.

## Configuration

| Parameter | Default | Description |
|-----------|---------|-------------|

## Execution Rules

Numbered list of non-negotiable rules.

## Workflow

### Step 1: ...
### Step 2: ...

## Troubleshooting

### Error: [Common error]
Cause: ...
Solution: ...