skill-creator-patterns

Skill Patterns

Choose the pattern that best fits your skill's purpose. Most skills use one primary pattern, sometimes combining elements from others.

Pattern 1: Sequential Workflow

Use when: Users need multi-step processes executed in a specific order.

Key techniques:

• Explicit step ordering with numbered steps
• Dependencies between steps (output of step N feeds step N+1)
• Validation gates between steps
• Rollback instructions for failures

Structure:

## Workflow: [Name]

### Step 1: [Action]
[What to do]
Expected output: [What success looks like]

### Step 2: [Action]
Requires: Output from Step 1
[What to do]

### Step 3: [Action]
[What to do]

## Rollback
If Step 2 fails:
1. [Undo Step 1]
2. [Report error]

Real-world examples: deployment pipelines, onboarding workflows, data migration scripts, release processes.

Pattern 2: Reference Knowledge

Use when: Claude needs domain expertise to apply during regular work.

Key techniques:

• Core principles up front
• Decision tables for common choices
• Code examples showing correct vs incorrect patterns
• Anti-patterns section with WRONG/CORRECT comparisons
• Progressive disclosure (main concepts in SKILL.md, details in references/)

Structure:

# [Domain] Reference

## Core Principles
1. [Most important rule]
2. [Second rule]
3. [Third rule]

## Decision Guide

| Situation | Approach | Why |
|-----------|----------|-----|
| [Case A] | [Do X] | [Reason] |
| [Case B] | [Do Y] | [Reason] |

## Patterns

### [Pattern Name]
When to use: [condition]

[Code example]

## Anti-Patterns

### [Bad Pattern]
WRONG:
[code]

CORRECT:
[code]

Why: [explanation]

Real-world examples: coding conventions, API design guides, testing standards, architectural patterns.

Pattern 3: Iterative Refinement

Use when: Output quality improves with multiple passes.

Key techniques:

• Generate initial draft
• Quality check against explicit criteria
• Refinement loop with re-validation
• Clear stopping condition

Structure:

## Phase 1: Initial Draft
1. [Generate first version]
2. [Save to file]

## Phase 2: Quality Check
Review against these criteria:
- [ ] [Quality criterion 1]
- [ ] [Quality criterion 2]
- [ ] [Quality criterion 3]

## Phase 3: Refinement
For each failing criterion:
1. Identify the specific issue
2. Fix it
3. Re-check

Repeat until all criteria pass or 3 iterations completed.

## Phase 4: Finalization
1. Apply final formatting
2. Generate summary of changes

Real-world examples: code review, report generation, documentation writing, test coverage improvement.

Pattern 4: Context-Aware Routing

Use when: The same goal requires different approaches depending on context.

Key techniques:

• Decision tree based on input characteristics
• Different tool selection per branch
• Fallback options
• Transparency about choices made

Structure:

## Decision Tree

### Check 1: [Condition]
- If [A]: proceed to Path A
- If [B]: proceed to Path B
- Otherwise: proceed to Default Path

### Path A: [Name]
[Specific instructions for this case]

### Path B: [Name]
[Specific instructions for this case]

### Default Path
[Safe fallback instructions]

## Reporting
Always explain which path was chosen and why.

Real-world examples: file storage routing, environment-specific deployment, language-specific code generation.

Pattern 5: Multi-Tool Coordination

Use when: Workflows span multiple tools or MCP servers.

Key techniques:

• Clear phase separation by tool/service
• Data passing between phases
• Validation before moving to next phase
• Centralized error handling

Structure:

## Phase 1: [Service A]
1. [Action using Service A tools]
2. Capture: [output to pass forward]

## Phase 2: [Service B]
Input: [data from Phase 1]
1. [Action using Service B tools]
2. Capture: [output to pass forward]

## Phase 3: [Service C]
Input: [data from Phase 2]
1. [Action using Service C tools]

## Error Handling
- Phase 1 failure: [recovery steps]
- Phase 2 failure: [recovery steps]
- Phase 3 failure: [recovery steps]

Real-world examples: design-to-development handoff (Figma + Drive + Linear + Slack), CI/CD pipeline orchestration, cross-platform content publishing.

Pattern 6: Forked Research

Use when: Heavy exploration that should not pollute the main conversation context.

Key techniques:

• context: fork in frontmatter
• agent: Explore for read-only research
• Focused task in SKILL.md body
• Results summarized and returned to main conversation

Structure:

---
name: deep-research
context: fork
agent: Explore
---

Research $ARGUMENTS thoroughly:

1. Find relevant files using Glob and Grep
2. Read and analyze the code
3. Identify patterns and relationships
4. Summarize findings with specific file:line references

Real-world examples: codebase exploration, dependency analysis, architecture review, security audit.

Choosing the Right Pattern

Your skill needs to...	Best pattern
Execute steps in order	Sequential Workflow
Teach Claude domain knowledge	Reference Knowledge
Produce high-quality output	Iterative Refinement
Adapt to different inputs	Context-Aware Routing
Coordinate multiple tools/services	Multi-Tool Coordination
Explore without polluting context	Forked Research

Many skills combine patterns. A deployment skill might use Sequential Workflow as the primary pattern with Context-Aware Routing to handle different environments, and Iterative Refinement for the validation step.

Skill Sizing Guidelines

Skill Size	Lines in SKILL.md	Supporting Files	Use Case
Small	< 100	None	Single concept, simple reference
Medium	100-300	0-2	Workflow, moderate reference
Large	300-500	2-5	Complex workflow, extensive reference

If SKILL.md exceeds 500 lines, split content into supporting files. Keep SKILL.md focused on core instructions and navigation.