Refactoring AI Prompts

Explains architectural decisions in terms of the trade-offs they buy and the costs they defer, rather than presenting patterns as rules. Useful for engineers starting to make system-level calls who need to justify a choice rather than just name it.

Runs a lighter pass than a full review, catching the mechanical issues — naming, dead code, missing null handling — before a human reviewer spends attention on them. Best used as the step between finishing a branch and opening the pull request.

Turns a diff into a clear commit message that explains intent rather than restating the changed lines. Useful for keeping a history that is worth reading during an incident, when the question is always why a change was made and not what it touched.

Frames the model as an experienced Ethereum developer writing Solidity contracts with the access rules and state handling spelled out. Useful for scaffolding a contract and understanding the trade-offs, though any output needs a real audit before deployment.

Converts a written description of a system or process into diagram source you can render. Useful for documenting an architecture, a request flow, or a state machine without opening a drawing tool, and the output stays diffable in version control.

Asks the model to produce SVG markup for a described graphic, returning code you can paste directly into a page. Handy for simple icons, diagrams, and placeholder artwork when you want scalable output with no binary asset or build step.

Positions the model as a solution architect who maps requirements onto systems, integrations, and deployment topology. Use it when you need to reason about how a new component fits an existing estate before committing to a technical direction.

Produces commit messages that follow the Conventional Commits specification, with the right type prefix and a scoped subject line. Keeps a shared history readable and makes automated changelog and semantic-version tooling work properly across a team.

Puts the model in reviewer mode so it reads a diff and reports problems, risks, and improvement suggestions instead of rewriting everything. Use it as a first pass before human review to catch obvious defects, naming issues, and missing error handling early.

Converts a written description of a matching rule into a regular expression, returning the pattern without a lecture attached. Best used for validation rules, log parsing, and find-and-replace work where you know the intent but not the syntax. Always test the output against edge cases.

A refactoring prompt that puts characterisation tests before any code movement and outlaws behaviour changes — suspected bugs get flagged, not silently fixed. The output is a safe, reviewable migration rather than a rewrite gamble.