Build a Roadmap Factory with Claude

Configure a Claude Project that generates market-aligned learning roadmaps on demand.

Introduction

30 Second Summary

Creating educational content that keeps pace with the tech job market is like trying to hit a moving target. New tools emerge, skill demand shifts, and yesterday's hot topic becomes today's table stakes.

In this project, you will build a Claude Projects workspace that generates complete learning roadmaps on demand. Each roadmap comes with persona-driven project prompts ready to paste into a content generator.

What You'll Build

Open a new chat in your project, tell it which tech category you want, and watch it produce a full roadmap with 5-10 project prompts, market data citations, and a fish-out-of-water cover image concept in minutes.

By the end of this project, you'll have:

Generate market-aligned roadmaps for any of 21 tech categories by starting a single chat session.
Produce ready-to-paste project prompts with real-world persona scenarios, feasibility checks, and LinkedIn-worthy titles.
Refresh your roadmaps with current market data using web search verification built into the workflow.
Secret Mission: Add a self-audit engagement scorer that rates each generated roadmap against shareability, completion likelihood, and market relevance criteria.

Are there any prerequisites?

You need an active Claude Pro or Max subscription with access to Projects. Free accounts can create up to 5 projects, but Pro or Max gives you the expanded context and usage limits this project benefits from.

Before We Start

Before we dive in, let's lock in what you're about to build and why it matters for your content strategy. This step is your moment to commit to the outcome before any hands-on work begins.

Set Up Your Claude Project Workspace

Right now, every conversation you have with Claude starts from scratch. Previous prompts, context, and instructions disappear when you open a new chat.

A Claude Project solves this by giving you a persistent workspace where instructions carry over into every chat session. In this step, you'll create that workspace and configure it for roadmap generation.

In this step, get ready to:

Create a new Claude Project named "Roadmap Factory."
Enable web search in your account settings.
Confirm the project works by opening a chat inside it.

Create your Roadmap Factory project

Claude Projects are self-contained workspaces with their own chat histories and instructions. Every chat you open inside a project automatically inherits whatever instructions you set at the project level.

Navigate to claude.ai in your browser.
Hover over the left side of your screen to expand the sidebar.
Click Projects in the sidebar.
Click + New Project in the upper right corner.
Enter Roadmap Factory as the project name.
Leave the description field empty for now.
Click Create project.

You should see your new Roadmap Factory project open, with options to set project instructions and add knowledge.

What is a Claude Project?

A Project is a workspace where you define persistent instructions that apply to every chat inside it. Think of it as a dedicated assistant role that Claude steps into each time you start a new conversation in that project.

Project instructions are different from account-wide instructions. They only affect chats within this specific project, keeping your other Claude conversations unchanged.

🙋‍♀️ Don't see Projects in the sidebar?

Make sure you're hovering over the left edge of the screen to expand the sidebar. If you still don't see it, navigate directly to claude.ai/projects in your browser's address bar.

Projects are available on all Claude plans, including free accounts.

Help me find the Projects section in Claude

Enable web search in your settings

Your roadmap factory needs access to current market data. Web search lets Claude pull real-time information from the internet when generating roadmaps.

Click your initials in the lower left corner of the Claude interface.
Select Settings.
Navigate to Search and tools in the settings menu.
Confirm that Web search is toggled on.
Confirm that Research is also available and toggled on.

Both toggles should be in the "on" position. Close settings when you're done.

Why do we need web search?

Market trends shift constantly. New tools emerge, pricing changes, and job market data updates quarterly. Web search lets your roadmap factory verify that its recommendations are current every time you generate a new roadmap.

The Research tool builds on web search by conducting multi-step investigations across multiple sources. You must have web search turned on for Research to function.

🙋‍♀️ Don't see Search and tools in Settings?

The exact location of these toggles may vary slightly depending on your plan type. On Pro and Max individual plans, look under your account settings for a section related to tools, capabilities, or search. If web search is already enabled by default on your plan, you may not see a separate toggle for it.

You can also enable web search directly in a chat by clicking the + button in the lower left corner of the chat input and selecting Web search.

Help me enable web search on my Claude plan

Confirm the project is ready

Before you add instructions in the next step, confirm that the project workspace is functional.

Navigate back to your Roadmap Factory project (click Projects in the sidebar, then click Roadmap Factory).
Click to start a new chat inside the project.

Before you type anything, what do you expect to see? The project instructions are empty, so Claude should respond as its default self without any special role.

Type Hello! What project are you part of? and press Enter.

Claude responds normally without mentioning any specific role or project context. This is expected. The project workspace is functional, but it has no instructions yet. In the next step, you'll give it a defined role and behavior.

One roadmap per chat session

Each time you want to generate a new roadmap, start a fresh chat inside this project. This keeps each generation clean and avoids context overload from previous conversations.

Your workspace is ready. Next up, you'll write the core system prompt that transforms this blank project into a roadmap generation engine.

Write the Core System Prompt — Role & Context

Your Roadmap Factory workspace is ready, but right now it is an empty shell. Without a system prompt defining Claude's role, every chat inside that project produces generic, unfocused output.

In this step, you will write the foundational instructions that turn your project into a purpose-built roadmap engine. You will define Claude's identity, the categories it covers, and the structured workflow it follows in every session.

In this step, get ready to:

Define Claude's role as a NextWork Roadmap & Project Prompt Factory.
List the 21 content categories it covers.
Set up the session workflow and output constraints.

Write the role definition and categories

Project instructions act like a persistent briefing that every new chat in your project inherits automatically. The first thing your instructions need is a clear role definition telling Claude exactly what it is and what domains it covers.

Return to your Roadmap Factory project in claude.ai.
Click Set project instructions (or the pencil icon next to the instructions field if it already shows placeholder text).
Paste the following role definition and categories block into the instructions field:

# NextWork Roadmap & Project Prompt Factory

## Your Role

You are a NextWork Roadmap & Project Prompt Factory. Your job is to generate market-aligned learning roadmaps with ready-to-paste project prompts for NextWork's project generator. Each roadmap you produce is a themed learnlist of 5-10 projects that progress from Easy to Hard, feature persona-driven scenarios, and are optimized for learner engagement, LinkedIn shareability, and real-world skill relevance.

## Categories You Cover (21 Total)

1. AWS Cloud
2. Azure Cloud
3. GCP (Google Cloud)
4. DevOps & CI/CD
5. AI / Machine Learning
6. n8n Automation
7. Make Automation
8. GoHighLevel Automation
9. Software Engineering
10. Hermes / Nous Research Agents
11. Claude Agents
12. Custom AI Agents
13. AI Integration Projects
14. VA / AI-Powered Freelancing
15. Data Engineering & Analytics
16. Cybersecurity
17. Full-Stack Web Development
18. Mobile Development
19. MLOps / AI Deployment
20. No-Code / Low-Code
21. API Development & Integration

What does this block do?

The heading and role paragraph tell Claude its identity and core output format: themed learnlists of 5-10 projects.

The numbered categories list constrains what Claude considers in-scope. When you ask for a roadmap later, Claude will reference this list to confirm which domain you want before generating.

💡 Why define a role in project instructions?

Claude performs best when project instructions provide general context, key guidelines, and a clear role. Without a role definition, Claude defaults to general-purpose assistant behaviour and produces inconsistent outputs across sessions.

Add the session workflow

A role definition tells Claude what it is. A session workflow tells it how to behave in every conversation. This structured flow ensures Claude always asks the right questions and uses web search to verify market data before generating.

In the same project instructions field, place your cursor at the end of the categories list.
Paste the following session workflow block directly below the categories:

## Session Workflow

Every new chat session follows this flow:

1. ASK which category the user wants to generate a roadmap for.
2. ASK if the user has existing draft ideas or project concepts for this topic. If yes, incorporate them into the roadmap while maintaining quality and structure standards.
3. RESEARCH current market demand using web search. Verify:
   - Top skills employers are requesting in this category (check job posting data)
   - Current tool versions and pricing (free tier availability)
   - Any new tools or frameworks that have emerged in the last 3 months
   - Trending project types that get engagement on professional platforms
4. GENERATE the complete learnlist output (metadata + project prompts).
5. VALIDATE feasibility of each project prompt.
6. PRESENT the complete roadmap for review.

What does this workflow do?

Steps 1-2 are the interaction protocol. Claude always asks which category you want and whether you have draft ideas before doing anything else.

Steps 3-6 are the output constraints. Claude must research current market demand with web search, generate a complete learnlist, validate feasibility, and present the result for review. This guarantees your roadmaps are grounded in real data rather than stale training knowledge.

Click Save (or click outside the instructions field) to confirm your project instructions are stored.

Instructions field looks blank after saving?

Some browsers require a page refresh to display saved instructions. Click back into your project and re-open the instructions field to confirm your text is there.

If the text is genuinely missing, check that you pasted into the project instructions field and not into a new chat message box. Help me find my project instructions field in Claude

Test your system prompt

Your project instructions are saved. Now you need to verify Claude actually picks them up in a new conversation.

Inside your Roadmap Factory project, click New chat (or the compose button) to start a fresh conversation.

Before you type anything, predict: what do you expect Claude to say when you ask it about its role? Will it describe a general assistant, or something more specific?

Type the following message and press Enter:

What is your role?

Claude should respond describing itself as a NextWork Roadmap & Project Prompt Factory. It should mention generating market-aligned roadmaps, the 21 categories it covers, and the structured session workflow (asking for your category first, then drafts, then researching market data).

What should I see?

Claude's response should reference specific details from your instructions: the 5-10 project range, Easy-to-Hard progression, persona-driven scenarios, and the step-by-step session flow. If it responds as a generic assistant with no mention of roadmaps or learnlists, the instructions did not save correctly.

🙋‍♀️ Claude responds as a generic assistant?

Make sure you opened the new chat inside the Roadmap Factory project, not from the main Claude sidebar. Chats started outside a project do not inherit that project's instructions.

If you are inside the project but still see generic responses, click back into the project settings and confirm your instructions text is visible in the field. Why is Claude not using my project instructions?

✔️ Awesome, I've got everything!

Great. Double check you have both sections (Your Role + Categories, and Session Workflow) visible in your project instructions field before moving on.

ⓧ I'd like to double check the full code

Here is the complete project instructions text you should have at this point:

# NextWork Roadmap & Project Prompt Factory

## Your Role

You are a NextWork Roadmap & Project Prompt Factory. Your job is to generate market-aligned learning roadmaps with ready-to-paste project prompts for NextWork's project generator. Each roadmap you produce is a themed learnlist of 5-10 projects that progress from Easy to Hard, feature persona-driven scenarios, and are optimized for learner engagement, LinkedIn shareability, and real-world skill relevance.

## Categories You Cover (21 Total)

1. AWS Cloud
2. Azure Cloud
3. GCP (Google Cloud)
4. DevOps & CI/CD
5. AI / Machine Learning
6. n8n Automation
7. Make Automation
8. GoHighLevel Automation
9. Software Engineering
10. Hermes / Nous Research Agents
11. Claude Agents
12. Custom AI Agents
13. AI Integration Projects
14. VA / AI-Powered Freelancing
15. Data Engineering & Analytics
16. Cybersecurity
17. Full-Stack Web Development
18. Mobile Development
19. MLOps / AI Deployment
20. No-Code / Low-Code
21. API Development & Integration

## Session Workflow

Every new chat session follows this flow:

1. ASK which category the user wants to generate a roadmap for.
2. ASK if the user has existing draft ideas or project concepts for this topic. If yes, incorporate them into the roadmap while maintaining quality and structure standards.
3. RESEARCH current market demand using web search. Verify:
   - Top skills employers are requesting in this category (check job posting data)
   - Current tool versions and pricing (free tier availability)
   - Any new tools or frameworks that have emerged in the last 3 months
   - Trending project types that get engagement on professional platforms
4. GENERATE the complete learnlist output (metadata + project prompts).
5. VALIDATE feasibility of each project prompt.
6. PRESENT the complete roadmap for review.

Your system prompt is working. Every new chat in this project now starts with Claude fully briefed on its role, categories, and workflow. Next up, you will design the locked template that controls exactly how each project prompt is structured.

Design the Project Prompt Template

Your Roadmap Factory now knows its role and workflow. When you open a chat, it asks the right questions and follows the right sequence. But there is a gap: when it generates project prompts, each one will have a different structure unless you lock it down.

Inconsistent prompts produce inconsistent results downstream. In this step, you will define the exact template every prompt must follow, write persona guidelines so each scenario feels like a real job, and add a feasibility validator that catches unrealistic projects before they reach learners.

In this step, get ready to:

Define the exact project prompt template structure with all required fields.
Write persona guidelines that place the learner in a real professional scenario.
Add a feasibility validator checklist that gates every generated prompt.

Define the prompt structure

Every project prompt your factory generates needs the same fields in the same order. This consistency means you can paste any output directly into the NextWork project generator without reformatting.

Return to your Roadmap Factory project in Claude Projects.
Click into the project instructions field where your existing instructions live.
Scroll to the bottom of your current instructions (after the Session Workflow section) and paste the following template block:

## Project Prompt Template

Every project prompt you generate MUST follow this exact structure:

---

**[PROJECT NUMBER]. [LINKEDIN-WORTHY TITLE]**

**Persona:** [2-3 sentences. The learner IS the professional. Use: "You just got hired as...", "Your first client just asked you to...", "Your team lead just assigned you to...", "You are a [role] at [type of company] and today you need to..." Never invent character names. The scenario highlights the real-world skill being practiced.]

**Difficulty:** [Easy / Medium / Hard]

**Tools & Services:** [Specific tool names with version if relevant. Note free tier status for each.]

**What You Will Build:** [1-2 sentences describing the demonstrable, visual, or functional end product. Must be something you can screenshot, demo, or show someone.]

**Time Estimate:** [60-90 minutes]

**Key Concepts:** [3-5 specific technical concepts the learner will practice]

**Why This Matters:** [1 sentence connecting to market demand data.]

What does this template do?

Each field has a specific purpose in the final project output:

The Persona field places the learner in a real-world job scenario. No fictional characters.
The Tools & Services field names specific tools with their free-tier status. This prevents projects that require paid resources.
The What You Will Build field ensures every project produces something demonstrable. No invisible backend-only outcomes.
The Why This Matters field ties each project to current market demand data.

Save your project instructions.
Open a new chat inside your project and type: "Generate one sample project prompt for the Cloud Computing category."
You should see Claude output a prompt that includes each of the fields you defined (Persona, Difficulty, Tools & Services, What You Will Build, Time Estimate, Key Concepts, Why This Matters).

Claude not following the template?

Make sure you pasted the template below your existing instructions, not inside them. It should start with its own ## Project Prompt Template heading.
Confirm you saved the project instructions before opening the new chat. Unsaved changes do not apply to new sessions.
Open a fresh chat. Edits to project instructions only take effect in new chat sessions, not existing ones.

Help me debug why Claude isn't following my project prompt template.

Add persona guidelines

The template tells Claude what fields to output, but it does not tell Claude how to write them. Persona quality is the difference between a project that feels like homework and one that feels like a real day on the job.

In your project instructions, scroll below the Project Prompt Template section you just added.
Paste the following Persona Guidelines block:

## Persona Guidelines

- The learner IS the professional. Never say "imagine you are" - say "You are" or "You just..."
- No invented character names. The persona is the learner stepping into the role.
- Scenarios must reflect actual tasks that real professionals do in that role.
- The challenge in the scenario should be the exact skill the project teaches.
- Examples:
  - Cloud: "You just got hired as a junior cloud engineer. Your team lead assigned you to set up the staging environment for a new microservice before Friday."
  - VA/Freelancing: "Your first automation client just sent you a message: they need their lead intake form connected to their CRM with AI-powered scoring by next week."
  - DevOps: "You are a DevOps engineer and the deployment pipeline broke at 2am. Your job is to rebuild it with proper monitoring so this never wakes anyone up again."
  - AI Agents: "Your company just approved budget for an AI assistant that handles tier-1 support tickets. You have been tasked with building the proof of concept."

Why 'You are' instead of 'Imagine you are'?

"Imagine you are" creates psychological distance. The learner is observing a hypothetical. "You just got hired as" puts them directly in the seat. The scenario becomes their reality for the next 60 minutes.

The examples cover four different persona framings (just hired, first client, team assignment, company approval) so Claude has multiple patterns to draw from across categories.

Save your project instructions.

Add the feasibility validator

A project that requires a paid API with a waitlist or needs 4 hours to complete will never get finished by learners. The feasibility validator catches these problems at generation time.

In your project instructions, scroll back up to the Project Prompt Template section.
Find the line that says:

**Why This Matters:** [1 sentence connecting to market demand data.]

Directly below that line, paste the following feasibility check block and closing separator:

**Feasibility Check:**
- [ ] All tools have free tiers or are open source
- [ ] APIs are accessible without waitlists or approval periods
- [ ] Completable in 60-90 minutes by beginner-to-intermediate learner
- [ ] Produces a visible, demonstrable outcome
- [ ] No paid resources required (or cost clearly stated if unavoidable)

---

What does the feasibility check do?

This checklist acts as a gate. Claude must verify all five criteria for each prompt it generates. If a project fails any criterion (like requiring a tool with a waitlist), Claude flags it and adjusts before presenting the roadmap.

The closing --- separator marks the end of each project prompt block, keeping the template visually distinct from the next prompt in the output.

Save your project instructions.

Before you test this, what do you expect Claude's output to look like when you ask it for a sample prompt? Think about which fields should appear and in what order.

Open a new chat inside your project.
Type: "Generate a single sample project prompt for the DevOps category."

Claude should output a prompt containing all required fields in the exact structure you defined: a LinkedIn-worthy title, a persona scenario using "You are" or "You just..." framing (no invented names), difficulty level, tools with free-tier notes, a demonstrable outcome, time estimate, key concepts, a market relevance sentence, and a completed feasibility checklist.

Missing the feasibility checklist?

Check that the feasibility block is inside the template (between the opening --- and the closing ---), not after it. It is part of the template structure.
Confirm you opened a fresh chat after saving. Project instruction changes do not apply to existing conversations.

Help me figure out why Claude is not including the feasibility check in its generated prompts.

✔️ Awesome, I've got everything!

Great. Confirm you have saved your project instructions with all three additions: the Project Prompt Template, the Persona Guidelines, and the Feasibility Check block inside the template.

ⓧ I'd like to double check the full code

Your project instructions should contain these sections in order. Scroll through your instructions and confirm each is present:

## Your Role section (from Step 2).
## Categories You Cover (21 Total) section (from Step 2).
## Session Workflow section (from Step 2).
## Project Prompt Template section with all fields from **[PROJECT NUMBER]** through **Feasibility Check:** and its five checklist items, ending with a closing ---.
## Persona Guidelines section with four bullet rules and four examples (Cloud, VA/Freelancing, DevOps, AI Agents).

Your factory now has a locked template. Every prompt it generates will follow the same structure, making them ready to paste straight into the NextWork generator. Next up, you will feed it real market data so those prompts are backed by current demand.

Add Market Research Baseline & Refresh Mechanics

Your project instructions now define Claude's role, workflow, and prompt template. Every new chat session knows exactly what to produce and how to structure it.

There is a problem though. When Claude uses web search alone to gather market data, the results are slow and inconsistent. One session might find great stats, while another surfaces irrelevant blog posts. Baking verified market data directly into your instructions gives Claude a reliable foundation that it can then update with fresh searches.

In this step, get ready to:

Embed current market research data into your project instructions as a knowledge baseline.
Add a refresh protocol that tells Claude when and how to verify its data with web search.
Optionally upload supporting documents to your project's knowledge base.

Embed the market research baseline

Your market research baseline gives Claude a starting foundation of verified data so it does not need to search from scratch every time. You will place this new section between the Session Workflow and the Project Prompt Template in your project instructions.

Return to your Roadmap Factory project in Claude Projects.
Click into the project instructions field (where your existing instructions live).
Find the end of your ## Session Workflow section. Place your cursor after the last workflow item.
Paste the following Cloud/DevOps and AI/ML market data below the Session Workflow section:

## Market Research Baseline (Last Updated: August 2026)

Use this as a starting foundation. Always verify with web search before generating.

### Cloud / DevOps / SRE (from 360k+ job postings, Dec 2025 - Jun 2026)
- Terraform: 43.7% of DevOps postings
- Kubernetes: 43.0%
- AWS: 41.4% (30% of all tech postings)
- Python: 38.8%
- CI/CD: 32.4%
- Azure: 30.6% (24% of all tech postings)
- Docker: 13% of all tech postings
- GCP: 14% of all tech postings

### AI / ML (fastest growing category)
- AI is the #1 in-demand hard skill: 19.8% of all 360k+ postings
- AI/ML/data science postings up 163% YoY
- Specific AI subskills: LLMs (4.2%), Generative AI (3.3%), LangChain/LangGraph (2.2%), Prompt Engineering (1.6%), RAG (1.6%)
- Remote postings 3.2x likelier to require LLM experience

What does this data do?

Each category section gives Claude specific percentages from real job posting analyses. When Claude generates a "Why This Matters" line for a project prompt, it can cite these exact figures instead of making vague claims.

The "Last Updated: August 2026" date stamp tells Claude when this data was captured. This becomes critical for the refresh protocol you will add next.

Directly below the AI / ML section you just pasted, add the Automation, Software Engineering, and Key Market Signals data:

### Automation / Freelancing
- AI automation specialist: fastest-growing job category Q1 2026 (+340% postings)
- Average hourly rate for AI automation: $127/hour on Upwork
- n8n: Open-source, self-hosted, LangChain integration, unlimited executions.
- Make: Balanced power-to-cost ratio, visual workflow design.
- GoHighLevel: CRM-centric with Voice AI, Conversation AI, Agent Studio.
- Global workflow automation market: $23.77B (2025) projected to $37.45B by 2030
- 76% of businesses want to automate but lack in-house skills

### Software Engineering (stable high demand)
- Full Stack Development, React, JavaScript/TypeScript remain top
- AWS appears in 30% of all postings
- Software engineering roles: 38% of all tech posting volume

### Key Market Signals
- Unemployment for tech roles at historic lows
- Employers prioritize end-to-end execution over deep single-tool expertise
- Projects demonstrating real-world workflows are highly valued by hiring managers

Why include Automation and Freelancing data?

The automation category is the fastest-growing job segment, with a 340% increase in postings. Including this data means Claude will weight automation projects appropriately when generating roadmaps for that category.

The Key Market Signals section gives Claude cross-category context. It steers project design toward end-to-end workflows and demonstrable outcomes regardless of which specific category you request.

Before you test, do you think Claude will already reference this specific data if you ask about DevOps skills? Or will it fall back on its general knowledge?

Save your project instructions by clicking outside the text field or pressing the save control.
Open a new chat inside the project.
Type: "What are the top skills for DevOps roles based on your data?"

Claude should respond with specific percentages from your baseline: Terraform at 43.7%, Kubernetes at 43.0%, AWS at 41.4%. If it gives vague answers without numbers, double-check that your Market Research Baseline section is properly saved in the instructions.

Claude not citing your specific data?

Make sure you pasted the Market Research Baseline section between the Session Workflow and the Project Prompt Template. If it is at the very bottom, Claude may not weight it as heavily.

Also confirm you opened a new chat after saving. Existing chats do not always pick up instruction changes mid-conversation.

Still not working? Help me troubleshoot why Claude isn't using my project instructions data.

Add the refresh protocol

Your baseline data has a timestamp. The refresh protocol tells Claude what to do when that data ages past 3 months: search the web for updates before generating a roadmap.

In your project instructions, place your cursor directly below the Key Market Signals section you just added.
Paste the following refresh protocol:

### Refresh Protocol
If any baseline data is older than 3 months from the current date, use web search to verify and update before generating. Always search for new tool releases, pricing/free-tier changes, emerging tools gaining traction, and shifts in job posting data.

Why 3 months as the threshold?

The tech job market shifts quarterly. Tool releases, pricing changes, and new frameworks tend to gain visibility within a 3-month cycle. Setting this threshold means Claude will always verify before generating, but it will not waste time re-searching data you just updated last week.

While you are editing, also streamline your Session Workflow section. Remove the sub-bullet details under items 2 and 3 since that verification detail is now captured in the Refresh Protocol. Your Session Workflow items should be concise single lines:

## Session Workflow

Every new chat session follows this flow:

1. ASK which category the user wants to generate a roadmap for.
2. ASK if the user has existing draft ideas or project concepts for this topic.
3. RESEARCH current market demand using web search.
4. GENERATE the complete learnlist output (metadata + project prompts).
5. VALIDATE feasibility of each project prompt.
6. PRESENT the complete roadmap for review.

Why condense the workflow?

Claude performs best when project instructions are concise and focused on essential information. Reserve task-specific detail for the relevant sections. Your Refresh Protocol now handles the "what to verify" instructions, so the Session Workflow only needs to state the high-level flow.

Also condense your Project Prompt Template field descriptions and Persona Guidelines to their essential content. Remove the extended examples from Persona Guidelines and the verbose field descriptions from the template. Use the double-check tab below as a reference for the condensed versions.
Save your project instructions.

Keep instructions tight

Anthropic's official guidance is to keep project instructions focused on general context and key guidelines. Detailed task-specific instructions belong in the chat itself. Since you are adding substantial market data, trimming verbose sections elsewhere keeps your total instruction length manageable.

Upload supporting documents (optional)

If you have existing draft project ideas, research notes, or reference materials for specific categories, you can upload them to your project's knowledge base. Claude will draw on these when you mention having drafts.

In your Roadmap Factory project, look for the knowledge base area (where you can add files or documents).
Upload any draft learnlist ideas, market research PDFs, or notes you want Claude to reference.
If you do not have any drafts ready, skip this. You can always add documents later.

What is RAG in Claude Projects?

When your uploaded knowledge approaches the context window limits, RAG (Retrieval-Augmented Generation) activates automatically. It expands Claude's effective capacity by up to 10x, pulling in only the relevant portions of your documents for each chat.

RAG is available on all Claude plans and works alongside web search and extended thinking.

Now for the real test. Before you run it, what do you expect Claude to do differently now that it has both baseline data AND a refresh protocol?

Open a new chat inside your project.
Type: "What are the current top skills for DevOps roles?"

Claude should respond with specific data from your baseline (Terraform 43.7%, Kubernetes 43.0%, AWS 41.4%) and indicate it verified or attempted to verify with web search. You may see it reference the job posting analysis or note the data source period (Dec 2025 to Jun 2026).

Claude not mentioning web search verification?

Check that web search is still enabled in your account settings (Settings > Search and tools > Web search toggled on).

If Claude cites the data but does not mention verifying, that is still acceptable. The refresh protocol activates most visibly when the baseline data ages past 3 months.

Need more help? Why isn't Claude using web search to verify my market data?

✔️ Awesome, I've got everything!

Great. Double check you have saved your project instructions with all the changes.

ⓧ I'd like to double check the full code

Your full project instructions should now contain these sections in order. Scroll through your instructions and confirm each is present:

## Your Role (unchanged from before).
## Categories You Cover (21 Total) (unchanged from before).
## Session Workflow (streamlined to 6 single-line items with no sub-bullets).
## Market Research Baseline (Last Updated: August 2026) (NEW: includes Cloud/DevOps/SRE, AI/ML, Automation/Freelancing, Software Engineering, Key Market Signals, and Refresh Protocol subsections).
## Project Prompt Template (condensed field descriptions).
## Persona Guidelines (condensed to core rules without extended examples).

If any section is missing or out of order, revisit the substeps above to add it.

Your roadmap factory now has a data-driven memory. Next, you will lock in the output format so every roadmap Claude produces is copy-paste ready with metadata, branding, and proper project progression.

Build the Learnlist Output Format & Cover Image Concepts

Your factory can now generate individual project prompts with personas and feasibility checks. But a collection of prompts without packaging is not a publishable learnlist.

In this step, you will add the output structure that turns raw prompts into a cohesive roadmap. Each generated learnlist will include metadata, visual branding, and a progression strategy that tells Claude how to order projects from Easy to Hard.

In this step, get ready to:

Define the learnlist metadata format that wraps every roadmap output.
Add fish-out-of-water cover image branding guidelines with domain-specific examples.
Configure progression logic so Claude orders projects appropriately per domain.

Condense existing instructions to make room

Your project instructions are growing. Claude performs best when project instructions provide concise general context rather than verbose detail.

In your Roadmap Factory project, click into the project instructions field.
Select your entire ## Market Research Baseline section (from the heading down through the Refresh Protocol).
Replace it with this condensed version:

## Market Research Baseline (Last Updated: August 2026)

### Cloud / DevOps / SRE
- Terraform: 43.7%, Kubernetes: 43.0%, AWS: 41.4%, Python: 38.8%, CI/CD: 32.4%, Azure: 30.6%, Docker: 13%, GCP: 14%

### AI / ML
- AI #1 in-demand hard skill: 19.8% of postings. AI/ML postings up 163% YoY. LLMs 4.2%, GenAI 3.3%, LangChain 2.2%, Prompt Engineering 1.6%, RAG 1.6%

### Automation / Freelancing
- AI automation specialist +340% postings Q1 2026. $127/hr avg on Upwork. n8n, Make, GoHighLevel top tools. Market $23.77B (2025).

### Software Engineering
- Full Stack, React, JS/TS top. AWS in 30% of postings. SE roles 38% of tech volume.

### Refresh Protocol
If baseline data is older than 3 months, use web search to verify before generating. Always verify tool versions, pricing changes, and new entrants.

Why condense?

Claude's official guidance is to keep project instructions focused on general context and key guidelines. The detailed numbers are still there for reference, just trimmed of redundant prose. Task-specific detail belongs in the chat itself.

Now select your ## Project Prompt Template and ## Persona Guidelines sections (everything from the Template heading through the end of your current instructions).
Replace them with this condensed version:

## Project Prompt Template

---

**[PROJECT NUMBER]. [LINKEDIN-WORTHY TITLE]**

**Persona:** [2-3 sentences. Learner IS the professional.]

**Difficulty:** [Easy / Medium / Hard]

**Tools & Services:** [Tool names with free tier notes.]

**What You Will Build:** [Demonstrable end product.]

**Time Estimate:** [60-90 minutes]

**Key Concepts:** [3-5 concepts]

**Why This Matters:** [Market demand connection.]

**Feasibility Check:**
- [ ] All tools have free tiers or are open source
- [ ] APIs accessible without waitlists
- [ ] Completable in 60-90 minutes
- [ ] Produces visible outcome
- [ ] No paid resources required

---

## Persona Guidelines

- Learner IS the professional. Use "You are" or "You just..."
- No invented character names.
- Scenarios reflect actual professional tasks.

What changed?

The template structure is identical. You trimmed the placeholder descriptions to their essentials. Claude still knows what each field expects because the field names are self-explanatory.

Confirm your project instructions now end at the Persona Guidelines section with no trailing content below it.

Define the learnlist metadata format

Every roadmap needs a header that wraps the project prompts into a publishable learnlist. This metadata block tells Claude exactly what information to output before the prompts begin.

At the very bottom of your project instructions (after the Persona Guidelines section), add this new section:

## Learnlist Output Format

Every roadmap output must begin with this metadata block:

---

### LEARNLIST METADATA

**Title:** [Catchy, specific learnlist title]

**Description:** [2-3 sentences. Who is this for? What will they achieve?]

**Target Audience:** [Job Seekers | Career Switchers | Developers Upskilling | Freelancers Building Services]

**Estimated Total Time:** [Sum of all project times]

**Progression:** [Sequential | Independent | Mixed]

**Cover Image Concept:** [Fish-out-of-water scene description]

---

What does this metadata do?

Each field gives you copy-paste-ready learnlist packaging. The Title and Description become the learnlist card on the platform. Target Audience helps you verify the roadmap is focused. Progression tells Claude how to sequence the projects that follow.

Save your project instructions.
Open a new chat in your project and type: "Generate just the metadata header for a Cloud Computing learnlist."
You should see Claude output a metadata block with all six fields filled in (Title, Description, Target Audience, Estimated Total Time, Progression, Cover Image Concept).

Claude not following the metadata format?

Make sure you pasted the section below your Persona Guidelines, not inside it. The heading ## Learnlist Output Format must be at the same level as your other top-level sections.

Help me debug my learnlist metadata output format

Define the fish branding

Each learnlist gets a unique cover image concept featuring a fish placed in an unexpected, non-aquatic environment. This is the signature visual brand element that represents moving out of your comfort zone to learn something new.

Directly below the closing --- of your metadata block, add this section:

### Cover Image Concept Guidelines

Your signature brand element is a FISH placed in unexpected, non-aquatic environments.

Rules:
- NEVER place the fish in water, ocean, aquarium, or any aquatic environment
- Scene should be metaphorically connected to the roadmap's domain
- Make it humorous and unexpected
- Describe in enough detail for an AI image generator

Examples:
- Cloud Computing: "A tropical fish wearing a hard hat, standing on a cloud made of server racks, plugging ethernet cables into the sky"
- DevOps: "A pufferfish in a control room surrounded by monitors showing deployment pipelines, wearing headphones and drinking coffee at 3am"
- AI Agents: "A swordfish in a business suit commanding a room full of robot assistants from a whiteboard"
- Automation: "A clownfish in a factory setting, juggling gears and conveyor belts, with a clipboard checklist floating nearby"
- Cybersecurity: "A lionfish in a bank vault, wearing night-vision goggles, surrounded by lasers it is carefully navigating through"
- VA/Freelancing: "A goldfish at a standing desk with 6 monitors, each showing a different client chat, wearing reading glasses"
- Full-Stack Web Dev: "A beta fish building a house of cards made of HTML tags, balancing CSS on its fin"

Why these specific rules?

Without the "never aquatic" constraint, Claude defaults to putting fish in water. That defeats the metaphor. The examples give Claude a reference point for tone and detail level so it generates scenes that an AI image tool can actually produce.

Save your project instructions.
In the same chat (or a new one), ask Claude: "Generate a cover image concept for a Cybersecurity learnlist."
You should see a fish-out-of-water scene description that avoids any aquatic setting and connects metaphorically to cybersecurity.

Fish ended up in water?

Check that the "NEVER place the fish in water" rule appears in your instructions. If Claude still defaults to aquatic scenes, add a stronger constraint: "If the scene contains water in any form, reject it and generate a new concept."

Help me fix my cover image guidelines so the fish never appears in water

Add the progression logic

Different domains need different project ordering. A Cloud Engineering roadmap builds sequentially (project 2 uses infrastructure from project 1). An AI Tools roadmap works better as independent projects exploring different tools.

Directly below your Cover Image Concept Guidelines section, add this final section:

### Progression Logic

- Sequential: Each project builds on the previous. Use for skill-building domains like Cloud Engineering.
- Independent: Each project is standalone. Use for breadth domains like AI Tools.
- Mixed: Some sequential, some standalone. Use when appropriate.

What does this do?

This tells Claude how to structure dependencies between projects. When it picks "Sequential" in the metadata, later projects reference outputs from earlier ones. When it picks "Independent", each project stands alone.

Save your project instructions.

Now it is time to test the full factory end-to-end. Before you run this test, what do you expect the output to contain?

Open a new chat in your project.
When Claude asks which category you want, respond with a category of your choice (e.g. "Automation").
When Claude asks for draft ideas, respond with "No drafts, generate from scratch."
Claude should output a complete learnlist: the metadata header (with all six fields including a fish cover image concept and a Progression value), followed by 5-10 project prompts in your exact template format, progressing from Easy to Hard.

Output missing the metadata header?

Make sure the ## Learnlist Output Format section appears after your Persona Guidelines. Also check that your Session Workflow step 4 still says "GENERATE the complete learnlist output (metadata + project prompts)" so Claude knows to include both parts.

Help me get Claude to output the full learnlist with metadata and project prompts together

✔️ Awesome, I've got everything!

Great. Double check you have saved your project instructions.

ⓧ I'd like to double check the full code

Scroll through your project instructions and confirm these sections appear in order:

## Your Role (with 15 categories and session workflow).
## Market Research Baseline (condensed, with Cloud/DevOps, AI/ML, Automation, Software Engineering subsections and Refresh Protocol).
## Project Prompt Template (condensed template with all fields).
## Persona Guidelines (3 concise rules).
## Learnlist Output Format (metadata block with Title, Description, Target Audience, Estimated Total Time, Progression, Cover Image Concept).
### Cover Image Concept Guidelines (4 rules, 7 domain examples).
### Progression Logic (Sequential, Independent, Mixed).

Your factory now produces complete, packaged learnlists with branding and structure. Next up, you will add a self-audit mechanism that scores each roadmap before you publish it.

Secret mission

Add a Self-Audit Engagement Scorer

Generating roadmaps is only half the battle. Without a quality gate, some roadmaps will produce projects that learners abandon halfway through. Add a self-audit mechanism that scores each generated roadmap against engagement criteria, flagging low-performers before you publish.

Clean Up Your Resources

This project runs entirely within your existing Claude Pro or Max subscription. There are no additional ongoing costs beyond your subscription fee.

Decide whether to keep your project workspace available for future use or delete it entirely.

Resources you used:

1 Claude Project workspace ("Roadmap Factory") with project instructions configured.
Uploaded knowledge documents (if you added any during Step 4).

Keep everything running

No action needed. Choose this if you plan to keep generating roadmaps.

Your "Roadmap Factory" project persists indefinitely in your Claude account.
Open a new chat inside the project anytime to generate a fresh roadmap.
Your project instructions and uploaded documents remain available across sessions.

Pause - I'll come back to this later

There is nothing to pause. Claude Projects are always available when you return. Your instructions and knowledge documents stay exactly as you left them.

Projects do not consume resources when idle.
When you are ready to use it again, open the project from your Projects panel and start a new chat.

Delete - I don't want to use this again

Remove the project workspace and all its contents permanently. This deletes your project instructions and all chat history within the project.

Click **Projects** in the left sidebar of claude.ai.
Locate your "Roadmap Factory" project.
Click the three dots menu on the upper right corner of the project card.
Select **Delete** from the menu that appears.
Click **Yes, delete** in the confirmation pop-up.

What gets deleted?

Deleting the project removes your project instructions, all uploaded knowledge documents, and every chat conversation within the project. Your Claude account and subscription are not affected.

Nice Work!

You just built a fully configured Claude Projects workspace that generates market-aligned roadmaps with ready-to-paste project prompts on demand.

You've learned how to:

Write a system prompt that defines a role, interaction protocol, and output constraints so every chat session starts aligned to your goals.
Design a locked project prompt template with persona-driven scenarios, feasibility checks, and LinkedIn-worthy outcomes that feeds consistent results into a project generator.
Embed market research baseline data with a refresh protocol and a complete learnlist output format so your roadmaps stay current and are copy-paste ready with metadata and fish-out-of-water cover image branding.
Secret Mission: Added a self-audit engagement scorer that rates each roadmap against shareability, completion likelihood, and market relevance criteria.

Ready to quiz yourself?