matteotitta Workflow Design
Design multi-step AI workflows with chained prompts for B2B SaaS marketing deliverables.
Claude Code Triggers
Invoke this skill when user says:
- "Design a workflow"
- "Create a process for [task]"
- "Build a prompt chain"
- "Sequence prompts for [goal]"
- "Create an AI workflow"
- "Multi-step process for [task]"
- "Chain prompts together"
Do NOT invoke when:
- User wants a single prompt → Use
prompt-designskill - User wants to execute a prompt → Just run it directly
- User wants code/app built → Use
vibe-codingskill
Input Requirements
Required Inputs
| Input | Description | Source |
|---|---|---|
| Task description | What the workflow should accomplish | User specifies |
| Desired outcome | What the final output should be | User specifies |
Optional Inputs (improve quality)
| Input | How It Helps |
|---|---|
| Existing prompts | Builds on proven patterns |
| Integration needs | Determines variable chaining |
| Tool context | Adapts for specific AI (Claude, GPT, etc.) |
| Quality criteria | Defines success metrics per stage |
Input Validation Checklist
Before proceeding, verify:
- Task description is clear
- Desired outcome is defined
- Stages can be identified
If inputs are missing: Ask for task description and desired outcome.
Core Frameworks (voice-locked invariants)
Workflow Patterns
| Pattern | Structure | Use When |
|---|---|---|
| Linear | Stage 1 → 2 → 3 → 4 | Research → Strategy → Execution |
| Parallel + Merge | [A + B + C] → Merge → D | Independent research streams |
| Iterative Loop | Stage → Check → Refine → Check | Quality-critical outputs |
| Fork | Stage 1 → [A or B] based on condition | Different paths for different inputs |
Variable Chaining Rules
| Rule | Description |
|---|---|
| Explicit naming | {{stage1_output}} not just {{output}} |
| Format matching | Output format of N must match input format of N+1 |
| Validation | Check variables exist before using |
| Fallback | Define behavior if variable is empty |
Process pointer
Step-by-step phases, flowchart, and pre-built workflow chains live in the premium reference.
Summary: validate inputs → map workflow → generate stage prompts (each using 7-section architecture) → create variable registry → write execution guide → compile single markdown document.
Anti-Hallucination Guardrails
- Use exact 7-section structure for each stage. Don't invent new sections.
- Mark all variables. Every user input and stage output must be
{{marked}}. - Include quality gates. Every stage needs verification criteria.
- Test variable chains. Ensure outputs match expected inputs.
- Reference existing patterns. Check reference files before creating new workflow stages.
Quality
Pre-delivery checklist, worked example (Transcript-to-LinkedIn workflow), and anti-examples live in the premium reference.
Minimum bar: diagram complete, variable registry covers every variable, all stages carry a 7-section prompt, quality gates defined between stages, execution guide present.
Skill Auto-Update Protocol
Feedback signal detection, pattern triggers, and self-update format live in the premium reference.
Integration with Other Skills
Each stage uses the 7-section prompt architecture from the prompt-design skill. For individual prompt creation, use prompt-design.
| Skill | Relationship | Usage |
|---|---|---|
| prompt-design | Foundation | Each workflow stage uses 7-section prompt architecture |
| All content skills | Uses output | Workflows feed into skill execution |
| vibe-coding | Related | For building tools that use workflows |
| transcript-analysis | Provides input | Interview workflows |