Prompt Export

Transform design artifacts into a sequenced prompts.md file that guides coding agents through incremental development.

Why This Exists

Breaks development into sequenced prompts with checkpoints, preventing agents from losing focus or making assumptions.

Input Requirements

This skill expects upstream context from one or more of:

• Problem framing document
• PRD
• UX spec (flows, screens, interactions)
• Any combination of the above

If the user hasn't completed upstream work, suggest they start with problem-framing or prd-generation first.

Workflow

Step 1: Gather Artifacts

Ask the user to provide their design artifacts. Accept any format—markdown, docs, images of sketches, bullet points.

Step 2: Extract Key Elements

From the provided artifacts, identify:

Step 3: Clarify Gaps

If critical information is missing, ask targeted questions:

• "What tech stack do you want? (React, vanilla JS, etc.)"
• "Any specific libraries or frameworks to use or avoid?"
• "Mobile-first, desktop-first, or responsive?"
• "Any authentication or data persistence needs?"

Keep questions minimal—infer where possible.

Step 4: Decompose Into Prompts

Break the project into a logical sequence. Typical prompt sequence:

1. Project setup — scaffold, dependencies, config
2. Data/state foundation — types, schemas, state structure
3. Core component(s) — one prompt per major component or screen
4. Feature implementation — one prompt per feature
5. Integration — connect components, wire up flows
6. Polish — error states, loading states, edge cases
7. Final review — cleanup, optimization, documentation

Adapt based on project complexity. Simple projects might be 3-4 prompts. Complex ones might be 10+.

Step 5: Generate prompts.md

Output a single file the user can reference as they work through Claude Code.

Output Format

Automatically save the output to prompts.md (in project root) using the Write tool while presenting it to the user.

# [Project Name] — Development Prompts

## Context
[2-3 sentences: what this is, who it's for, core problem it solves]

**Tech stack:** [framework, styling, key libraries]
**Key constraints:** [what's in/out of scope]

---

## Prompt 1: Project Setup
> [Prompt text the user pastes into Claude Code]

**Goal:** [What this accomplishes]
**Checkpoint:** [How to verify it worked before moving on]

---

## Prompt 2: [Name]
> [Prompt text]

**Goal:** [What this accomplishes]
**Checkpoint:** [How to verify]

---

## Prompt 3: [Name]
> [Prompt text]

**Goal:** [What this accomplishes]
**Checkpoint:** [How to verify]

---

[Continue for all prompts...]

---

## Final Checklist
- [ ] [Success criterion 1]
- [ ] [Success criterion 2]
- [ ] [Success criterion 3]

Prompt Writing Guidelines

Each prompt should:

• Be self-contained — agent shouldn't need to read previous prompts
• Reference existing files — "In src/components/Header.jsx, add..."
• Specify acceptance criteria — what "done" looks like
• Stay focused — one logical unit of work

Good prompt:

"Create a TaskList component in src/components/TaskList.jsx that displays an array of tasks. Each task shows a title, due date, and completion checkbox. Clicking the checkbox toggles completion state. Use Tailwind for styling. The component receives tasks as a prop."

Bad prompt:

"Build the task management features."

Adaptation Guidelines

Simple project (3-5 prompts):

• Setup → Core UI → Features → Polish

Medium project (5-8 prompts):

• Setup → Data layer → Screen 1 → Screen 2 → Feature A → Feature B → Integration → Polish

Complex project (10+ prompts):

• Consider grouping into phases with phase summaries
• Add dependency notes ("Run after Prompt 3")
• Include rollback hints for risky steps

Handoff

After presenting the prompts, inform the user:

"Work through these prompts sequentially with Claude Code—each builds on the last. Check the checkpoint before moving to the next prompt."

Offer to:

• Adjust prompt granularity (more/fewer steps)
• Reorder sequence
• Add detail to specific prompts

Note: File is automatically saved to prompts.md in the project root.