Setting Up Projects for Organized Workflows
This lesson walks you through creating a Claude Project from scratch, writing effective custom instructions, and uploading knowledge files that make Claude instantly useful for your specific work.
Creating a Project
Step 1: Open the Projects Panel
On claude.ai, click Projects in the left sidebar. Then click Create Project.
Step 2: Name Your Project
Choose a clear, specific name:
| Good Names | Why They Work |
|---|---|
| "Q1 Marketing Campaign" | Specific timeframe and scope |
| "React Dashboard App" | Clear technology and deliverable |
| "PhD Thesis — Chapter 3" | Organized by section |
| "Client: Acme Corp" | Grouped by client |
Avoid generic names like "Work Stuff" or "AI Help" — they become useless once you have multiple projects.
Step 3: Add a Description
Write a one-sentence summary of the project's purpose. This helps you (and team members) quickly identify what the project covers.
Writing Effective Custom Instructions
Custom instructions are the most impactful part of any Project. They tell Claude how to behave in every conversation within that project.
The Five-Part Framework
Structure your instructions around these five areas:
1. Role — Who Claude should act as
You are a senior content strategist helping me write
blog posts for a B2B SaaS startup.
2. Context — Background information
Our product is a project management tool for remote teams.
Our audience is engineering managers at companies with 50-500 employees.
We compete with Asana, Monday.com, and Linear.
3. Guidelines — Rules and preferences
- Use active voice and short sentences
- Include data or examples to support every claim
- Avoid buzzwords like "synergy" and "leverage"
- Write at a 10th-grade reading level
4. Format — How to structure responses
- Blog posts: 1000-1500 words with H2 headers every 200 words
- Always include a TL;DR at the top
- End with 3 actionable takeaways
5. Constraints — What to avoid
- Never mention competitors by name in published content
- Do not make claims about uptime or performance without data
- Avoid first-person plural ("we believe") in technical posts
Real Example: Developer Project
Here is a complete set of instructions for a software development project:
Role: You are a senior full-stack developer working on my Next.js application.
Tech stack:
- Next.js 15 with App Router and Server Components
- TypeScript in strict mode
- Tailwind CSS for styling
- Supabase for database and authentication
- Jest for testing
Coding standards:
- Prefer server components unless client interactivity is needed
- Use "use client" directive only when necessary
- Handle errors in all async functions with try/catch
- Write TypeScript types for all function parameters and returns
- Follow existing patterns in the codebase
Response format:
- Show complete files, not snippets, when making changes
- Explain WHY you made architectural decisions
- Flag potential performance issues proactively
- Suggest tests for any new functionality
Uploading Knowledge Files
Knowledge files give Claude reference material it can access in every conversation within the project.
What to Upload
| File Type | Examples | Best For |
|---|---|---|
| Documents | Brand guides, SOPs, policies | Consistent voice and rules |
| Code files | Config files, key modules, schemas | Technical context |
| Data | CSVs, product catalogs, pricing | Reference lookups |
| Research | Papers, reports, analysis | Informed responses |
| Transcripts | Meeting notes, interviews | Context and decisions |
Upload Best Practices
Do:
- Name files descriptively:
api-schema-v2.jsonnotdata.json - Upload the most critical documents first (there is a total size limit)
- Use plain text or Markdown when possible — they are indexed more effectively than heavily formatted PDFs
- Remove outdated files when you upload updated versions
Avoid:
- Uploading entire codebases — focus on key files (configs, schemas, core modules)
- Uploading images unless Claude needs to analyze them (they consume quota quickly)
- Duplicating content across files — consolidate instead
How Claude Uses Knowledge Files
When you ask a question in a project conversation, Claude automatically searches your knowledge files for relevant information. You do not need to tell it to "look at the uploaded file" — it does this by default.
You can also reference files explicitly:
Based on the brand guidelines I uploaded, review this email draft
for tone and style consistency.
Organizing Multiple Projects
As you create more projects, keep them organized:
By Purpose
- "Blog Content" — writing instructions + brand guide
- "Backend API" — technical context + code files
- "Client Proposals" — proposal templates + pricing data
By Client or Stakeholder
- "Client: Acme Corp" — their brand guide, past deliverables, preferences
- "Client: Globex" — different voice, different requirements
By Timeframe
- "Q1 2026 Campaign" — seasonal context and goals
- "Product Launch — March" — launch-specific materials
Key Takeaways
- Name projects specifically — generic names become useless as your project list grows
- Structure custom instructions with role, context, guidelines, format, and constraints
- Upload knowledge files that Claude can reference across all conversations in the project
- Use descriptive file names and prefer plain text or Markdown for better indexing
- Organize projects by purpose, client, or timeframe to keep your workspace manageable