prompt-vault

graph TD
    A[Try random prompt] --> B{Does it work?}
    B -->|No| C[Throw it away]
    B -->|Yes| D[Put in vault]
    D --> E[Use it everywhere]
    E --> F[Find edge case where it fails]
    F --> G[Tweak prompt]
    G --> B
    C --> H[Try another random prompt]
    H --> B

Battle-tested prompts that actually work. This is where prompts graduate from “I tried this once” to “I’ve used this 50 times and it still works.”

Contents

good/ - Examples of effective prompts and patterns

• example1.md - Direct instructions get correct action
• example2.md - Decomposed tasks with clear constraints
• example3.md - Confirmation before execution

bad/ - Examples of ineffective prompts and anti-patterns

• example1.md - Vague instructions lead to wrong action

repeated-prompts.md - Prompts you find yourself using repeatedly

What goes here

• Prompts that have proven themselves in real scenarios
• Templates with clear placeholders
• Patterns that work across different LLMs
• Examples of inputs and outputs
• Real conversation examples (good and bad) with analysis

What doesn’t go here

• Prompts you just thought of
• Theoretical perfect prompts
• Prompts that worked once under ideal conditions