Your Claude
setup.

• Claude chat: Ask questions, analyze files, create drafts, and explore an idea. A Free plan exists, with limits.
• Skills: Reusable instructions and resources that help Claude perform a recurring kind of task. Anthropic says Skills are available on Free and paid plans when code execution is enabled.
• Claude Code: A coding assistant that works with a project’s files, runs commands, and checks changes. It requires an eligible paid plan or a separately billed API route.
• Cowork: Claude’s agentic workflow for non-coding tasks such as research, file organization, and document creation. It is available on eligible paid plans. The interface is changing: some accounts show a Cowork selector, while a newer experience handles tasks in a unified chat.

You do not need all four for every project. Use chat to investigate, a Skill for a repeatable method, Claude Code to build software, and Cowork for larger knowledge-work tasks.

Sign in at claude.ai. For a recurring app idea, create a Project named for the problem you are studying. Add a short project instruction: target customer, what evidence counts, budget, and what Claude must not assume. Upload only material you have permission to use. Keep a separate notes file of actual customer conversations and public sources.

Project instruction starter

Anthropic’s built-in Skills can help with files such as PDFs, spreadsheets, and slides. In current Claude settings, ensure code execution and file creation are enabled, then look under Customize → Skills and turn on the relevant Skill. You can also create a personal Skill for a repeated process, such as “summarize five customer interviews in a fixed format.” Test it on one task, inspect the result, and revise the instructions. Use only Skills from sources you trust, especially if they include code.

Example Skill brief

Open an App Store category and study one established app. Record its public positioning, feature list, and recent reviews. Use screenshots only as research inputs. Apple describes ratings and reviews as customer feedback, but a few reviews can be unrepresentative. Look for repeated complaints and verify them with direct conversations. Do not reuse the app’s branding or reproduce its interface; Apple’s rules warn against copycat apps.

Research prompt

Before building, speak to five target users. Ask about their current workflow and what they already pay for. Choose one problem to solve. Create a one-page product brief: customer, painful task, smallest useful feature, success metric, and what is out of scope.

Install Claude Code using Anthropic’s current quickstart or use the Code tab in the Claude desktop app. Open a new project folder and ask Claude to describe its plan before editing files. Keep the first version small: one core workflow, sample data, no production payments or sensitive information. Review every permission request, inspect the changes, and test the result yourself.

Build prompt

If you have never coded, ask Claude to explain where the app runs, which files it changed, and how to start and stop the preview. Save a copy before major changes. Do not publish a prototype with real customer data until privacy and security have been reviewed.

Cowork is useful for tasks such as organizing interview notes, comparing competitor claims, creating a pitch deck, or preparing a research report. On accounts that show it, select Cowork and describe the output. On accounts with the newer unified Claude experience, describe the task in a normal conversation. Review its plan and outputs, especially if it accesses local files or connected services.

Cowork task prompt

1. Days 1-2: Choose a narrow buyer and study public alternatives.
2. Days 3-5: Interview five people; record exact problems and workarounds.
3. Day 6: Pick one problem and write the one-page brief.
4. Days 7-10: Build the smallest prototype in Claude Code.
5. Days 11-12: Let two target users try it without coaching.
6. Days 13-14: Fix the biggest obstacle and decide whether to continue, change, or stop.

Your goal at day 14 is learning, not a valuation. Repeat only if people use the prototype, request it again, or show credible willingness to pay.