/creating-workflows

name: orchestration:creating-workflows
description: Use when user says "create workflow", "create a workflow", "design workflow", "orchestrate", "automate multiple steps", "coordinate agents", "multi-agent workflow". Creates orchestration workflows from natural language using Socratic questioning to plan multi-agent workflows with visualization.

Creating Orchestration Workflows

I'll help you create powerful orchestration workflows that coordinate multiple Claude Code agents. I use Socratic questioning to understand your needs and generate optimal workflow syntax.

When I Activate

I automatically activate when you:

• Describe a multi-step process you want to automate
• Mention "workflow", "orchestration", "automate", "coordinate agents"
• Ask "how do I create a workflow?"
• Want to connect multiple agents or tasks
• Ask about automating repetitive processes

My Process

1. Understanding Your Intent

**CRITICAL**: I use AskUserQuestion tool for ALL questions. NO plain text numbered lists.

I'll ask strategic questions to understand:

• What problem you're solving
• What your goal is
• What scope you have in mind
• **What external data sources you need** (APIs, web scraping, databases)

2. Detecting Temp Script Needs

I automatically scan for these triggers:

• **External APIs**: Reddit, Twitter, GitHub, ProductHunt, etc.
• **Web Scraping**: Extracting data from websites
• **Data Processing**: Analyzing 10+ items, statistical analysis
• **Authentication**: Any service requiring API keys

**If detected → I'll proactively create temp scripts for you**

3. Identifying the Pattern

I'll determine if your workflow is:

• **Sequential**: One step after another (`->`)
• **Parallel**: Multiple tasks at once (`||`)
• **Conditional**: Based on results (`~>`)
• **Semantic Routing**: Dynamic switch statements based on LLM interpretation (`route() =>`)
• **Scheduled**: Runs automatically in the background (`@schedule(...)`)
• **Hybrid**: Combination of above

3. Designing the Workflow

I'll help you define:

• Which agents to use (built-in or custom)
• How data flows between steps
• Error handling strategy
• Review checkpoints

4. Generating Syntax

I'll create clean, readable workflow syntax like:

# Simple sequential workflow
explore:"Analyze codebase" ->
implement:"Add feature" ->
test:"Run tests"

# Parallel with merge
[security-check || style-check || performance-check] ->
general-purpose:"Consolidate findings"

Question Approach

For more details on my questioning strategy, see [socratic-method.md](socratic-method.md).

**Quick overview:**

• **Vague requests**: I ask about problem → scope → constraints
• **Specific requests**: I confirm pattern → ask about customization
• **Medium requests**: I explore scope → clarify details

Common Patterns

I have templates for common scenarios. See [patterns.md](patterns.md) for complete catalog.

**Popular patterns:**

• Feature implementation (explore → implement → test → review)
• Bug fixing (investigate → fix → verify)
• Security scanning (scan → review → fix → verify)
• Documentation (analyze → write → review)
• Refactoring (analyze → refactor → test → validate)

Custom Agents

When your workflow needs specialized expertise, I can create **temp agents** for you.

**Temp agents are:**

• Created automatically during workflow design
• Saved in `./temp-agents/` directory (in current working directory)
• Cleaned after workflow design (with user confirmation)
• Can be promoted to permanent agents if useful

**When I create temp agents:**

• You need domain-specific expertise (e.g., security scanner)
• Task requires specific output formats
• Multiple workflows might benefit (I'll suggest making it permanent)

See [temp-agents.md](temp-agents.md) for examples and guidelines.

Credential Security (CRITICAL)

When workflows require API credentials or sensitive data, I follow strict security practices:

Security Checklist

1. **Never hardcode credentials** in workflow files 2. **Use config files** for credentials (e.g., `config/reddit-credentials.json`) 3. **Provide example files** (e.g., `config/reddit-credentials.json.example`) 4. **Verify .gitignore** includes credential files before workflow runs 5. **Check git status** to ensure credentials aren't staged

Required Workflow Phase

For workflows requiring credentials, I add a **Phase 0: Security Verification** step:

# Phase 0: Security & Credential Verification
general-purpose:"SECURITY CHECK - Verify credentials are properly configured.

1. Check if credentials file exists
2. Verify .gitignore includes the credentials file
3. Check git status for accidental staging
4. Abort if credentials might be committed

If NOT protected:
- WARN user
- Offer to add to .gitignore
- Only proceed after confirmation":credentials_status

.gitignore Patterns

These patterns MUST be in .gitignore:

# API credentials - NEVER commit
config/*-credentials.json
config/credentials*.json
**/credentials.json
**/secrets.json
*.credentials.json
.env
.env.local

User Notification

When I detect a workflow needs credentials, I: 1. Ask user if they have credentials set up 2. Provide setup instructions with example file 3. Verify .gitignore protection 4. Add security verification phase to workflow

Temp Scripts (CRITICAL)

**Temp scripts** are Python/Node.js scripts I create for tasks that Claude Code tools can't handle directly.

When I Create Temp Scripts

I **automatically** create temp scripts when you need:

1. **External API calls** - Reddit, Twitter, GitHub, ProductHunt 2. **Web scraping** - Extracting data from websites 3. **Data processing** - Pandas analysis, JSON parsing at scale 4. **Database queries** - SQL, NoSQL operations 5. **Batch operations** - Processing 10+ files 6. **Third-party libraries** - NumPy, BeautifulSoup, requests

How It Works

**You say**: "Fetch 10 Reddit posts about startups"

**I create**:

g