Custom Prompts: Shaping Claude's Behaviour for Your Context
Advanced: Custom Prompts: Shaping Claude's Behaviour for Your Context
Custom Prompts: Shaping Claude’s Behaviour for Your Context
Series: Claude Learning Journey · Advanced Usage
A system prompt is an instruction you give Claude at the start of every session. The default is generic. Custom prompts are how you make Claude know about your specific context: your project, your preferences, your conventions, your standards. Used well, custom prompts are the difference between a generic Claude and a Claude that understands how you work.
The key insight: a good custom prompt is not a long prompt. It is a specific one. You are not describing your entire project — you are telling Claude the things that will most affect how it works with you.
What Goes in a Custom Prompt
The things that most affect how Claude works with you:
Your role and perspective: “I am a backend engineer working primarily in Python and Go. I prefer minimal code that is easy to read over clever code that is hard to follow.”
Your conventions: “We use type hints in Python everywhere. We follow Black formatting. We avoid premature abstraction.”
Your constraints: “We do not use external libraries unless necessary. We prefer standard library solutions.”
Your preferences: “When you suggest changes, show the before and after. When you write tests, use pytest. Keep functions under 40 lines.”
What Does Not Go in a Custom Prompt
A custom prompt is not a project README. Do not paste the entire project description in. Do not include the full context of what you are working on right now — that goes in the conversation, not the system prompt.
A custom prompt is also not a list of things Claude cannot do. Claude already knows its own limitations. Negative instructions (“never do X”) are less effective than positive ones (“prefer Y”).
Keeping Custom Prompts Updated
Custom prompts age badly if they are not maintained. The prompt that made sense three months ago might not reflect how you work now. Review and update them periodically.
The useful test: in a fresh session with your custom prompt, ask Claude to describe how you work. If it gets it wrong or incomplete, update the prompt.
What You’ll Learn
- What belongs in a custom prompt versus project context
- How to write instructions that shape behaviour without over-specifying
- The difference between negative and positive instructions
- Keeping custom prompts current
Try It Yourself
Write a custom prompt for a project you work on regularly. Keep it under 200 words. Focus on how you work, not what the project is. Start a fresh session and see how differently Claude performs with the context versus without it.
What’s Next
Custom prompts shape Claude’s behaviour in the conversation. But behaviour has security implications — the next post is about thinking clearly about what Claude can and cannot be trusted with.
Part of the Claude Learning Journey series · Next: Security: What Claude Can and Cannot Be Trusted With