Prompt library / coding

Code prompts should feel like operating procedures, not vibes.

The goal is not prettier code samples. The goal is to reduce ambiguity, force structure early and produce outputs that survive review, test and deployment.

Operator notes

Start with the required output structure before asking for implementation details.
If the task is risky, require assumptions, tests and failure modes in the same response.
Keep reusable prompt blocks copyable so editors and agents can work from the same baseline.

Validated patterns

42 ↑

Production-ready prompt blocks

Hallucination drop

-65% ↓

Against naive prompting

Token efficiency

+30% ↑

Shorter corrective loops

Security pass

100%

Schema-first outputs

Core coding prompt patterns

Pattern	Best use	Guardrail	Payload
Architecture wrapper	Greenfield services	Lock contracts first	Medium
Type guard generator	Parsing third-party APIs	Schema before implementation	Low
Unit test factory	Refactors and bugfixes	List edge cases explicitly	Medium
Legacy refactor engine	Old code modernization	Preserve behavior, then rewrite	High

Claude 3.5 Sonnet

Clean Architecture Wrapper

98%

Forces the model to split entities, use cases, adapters and interfaces before it writes code.

Act as a senior architect. Return the solution in Entity, UseCase, Repository and Controller layers. Use dependency inversion and explicit contracts.

GPT-4o

Type Guard Generator

95%

Turns loosely defined payloads into typed validators with runtime protection.

Generate a Zod schema and a matching TypeScript type guard for the object below. Include exhaustive error handling and test cases.

Llama 3.1

Unit Test Factory

92%

Expands a single function into meaningful happy-path, edge-case and integration tests.

Analyze the function and generate 10 Vitest cases, including null handling, race conditions and 2 timeout simulations.

Gemini 1.5 Pro

Legacy Refactor Engine

96%

Useful when migrating old code without losing behavior or test coverage.

Identify code smells, preserve behavior, propose a staged refactor and output the final modern implementation with regression notes.