Workflow Patterns: Combining Skills and Prompts

How to compose skills with prompts for multi-step workflows.

Skills vs Prompts

Component	Purpose	Location	When Used
Skills	Domain expertise, SDK patterns, coding standards	.github/skills/	Task matches skill description
Prompts	Reusable templates for specific actions	.github/prompts/	User invokes explicitly or agent selects

Skills provide knowledge. Prompts provide structure.

Composing Skills + Prompts

Example 1: Adding a New API Endpoint

Task: "Add a PUT endpoint for updating user profiles"

Components needed:

• Skill: fastapi-router-py — FastAPI patterns, response models, auth
• Skill: pydantic-models-py — Request/response schema patterns
• Prompt: add-endpoint.prompt.md — Step-by-step endpoint creation template

Workflow:

1. Agent loads fastapi-router-py skill
   → Learns: Router patterns, dependency injection, response models
   
2. Agent loads pydantic-models-py skill
   → Learns: Base/Create/Update/Response model pattern
   
3. Agent uses add-endpoint.prompt.md structure:
   → Step 1: Define Pydantic models (UserUpdate, UserResponse)
   → Step 2: Create router with PUT handler
   → Step 3: Add authentication dependency
   → Step 4: Write tests

Example 2: Building a Data Service with Storage

Task: "Create a document service backed by Cosmos DB"

Components needed:

• Skill: azure-cosmos-db-py — Cosmos DB service patterns
• Skill: azure-identity-py — Authentication with DefaultAzureCredential
• Skill: pydantic-models-py — Model definitions

Workflow:

1. Agent loads azure-identity-py skill
   → Establishes: Use DefaultAzureCredential, never hardcode credentials
   
2. Agent loads azure-cosmos-db-py skill
   → Learns: Service layer pattern, partition key strategy, parameterized queries
   
3. Agent loads pydantic-models-py skill
   → Learns: Model variant pattern (Base, Create, Update, Response, InDB)
   
4. Agent implements:
   → DocumentBase, DocumentCreate, DocumentResponse models
   → DocumentService class with CRUD operations
   → Proper error handling and logging

Example 3: Frontend Component with State

Task: "Create a workflow editor with draggable nodes"

Components needed:

• Skill: react-flow-node-ts — Custom React Flow nodes
• Skill: zustand-store-ts — State management
• Prompt: create-node.prompt.md — Node component template
• Prompt: create-store.prompt.md — Store creation template

Workflow:

1. Agent uses create-store.prompt.md with zustand-store-ts skill
   → Creates: workflowStore with nodes, edges, actions
   
2. Agent uses create-node.prompt.md with react-flow-node-ts skill
   → Creates: Custom node components with handles, TypeScript types
   
3. Agent integrates components
   → Connects store to React Flow canvas
   → Adds drag-drop functionality

Multi-Step Workflow Template

For complex tasks, structure the workflow explicitly:

## Task: [Description]

### Step 1: [Setup]
- Load skills: [skill-1], [skill-2]
- Verify: [check]

### Step 2: [Implementation]
- Use prompt: [prompt-name]
- Apply patterns from: [skill-name]
- Verify: [check]

### Step 3: [Integration]
- Connect components
- Verify: [check]

### Step 4: [Testing]
- Write tests following TDD pattern
- Verify: All tests pass

Prompt Templates in This Repository

Prompt	Purpose	Combines Well With
add-endpoint.prompt.md	Create REST API endpoints	fastapi-router-py, pydantic-models-py
create-store.prompt.md	Create Zustand stores	zustand-store-ts
create-node.prompt.md	Create React Flow nodes	react-flow-node-ts
code-review.prompt.md	Review code changes	Any skill for domain context

Best Practices

1. Load Skills Before Prompts

Skills establish context and patterns. Prompts structure the execution.

CORRECT:
1. Load azure-cosmos-db-py (establishes patterns)
2. Apply add-endpoint.prompt.md (structures implementation)

WRONG:
1. Apply add-endpoint.prompt.md (no domain context)
2. Load azure-cosmos-db-py (too late, patterns not applied)

2. Limit Active Skills

Keep 2-3 relevant skills active at once. More causes context dilution.

GOOD: fastapi-router-py + pydantic-models-py + azure-cosmos-db-py
BAD:  Loading all 127 skills "just in case"

3. Chain Skills for Complex Domains

When a task spans multiple domains, chain skills in logical order:

Authentication → Data Layer → API Layer → Frontend
azure-identity-py → azure-cosmos-db-py → fastapi-router-py → react-flow-node-ts

4. Use Prompts for Repeatability

If you do the same workflow often, create a prompt template:

---
name: new-azure-endpoint
description: Create a new Azure-backed API endpoint with full stack
---

## Steps

1. Load skills: azure-identity-py, [data-skill], fastapi-router-py
2. Create Pydantic models for request/response
3. Implement service layer with Azure SDK
4. Create FastAPI router with endpoint
5. Add authentication dependency
6. Write tests