Claude Prompt Techniques

Define the goal, relevant context and output structure before choosing stylistic language.

#Make the task observable

Instead of “write a program,” specify a Python function taking a list of strings, retaining items longer than five characters and sorting them alphabetically, with type hints and a docstring.

Explain the situation, source materials, audience and constraints. Request JSON, a table, bullets, code or a plan-before-implementation explicitly.

#Separate complex inputs

XML tags or clearly named sections can distinguish background, data, constraints and output format. A relevant role narrows the task:

textCopy
You are a frontend reviewer.
Prioritize bugs, regressions and accessibility issues.

Examples are useful for fixed output formats, labels and rewriting patterns.

#Engineering requests

Provide environment/framework details, say whether the task is explanation/review/implementation, list assumptions and request a plan for complex work.

textCopy
First explain the current state.
Then propose a plan.
List the files you expect to touch.
Only after that, write the code.

#Problem-solving brief

textCopy
Problem: [observed behavior]
Environment: [relevant versions]
Already tried: [attempts and results]
Please identify likely causes, propose a bounded solution
and explain how to verify it.

#Learning brief

textCopy
Explain [concept] for someone with [background].
Include a one-sentence definition, the problem it solves,
a basic example and important limitations.

#Common failures

Too many unrelated goals, missing constraints, abstract acceptance criteria and unspecified format weaken the result. Do not assume earlier conversations are available: provide the relevant code and the requested next change. State edge cases and performance requirements where they matter.

#References and next steps

• Prompt engineering↗
• XML tags↗
• Coding
• Projects
• Artifacts