← ClaudeAtlas

codednalisted

Generate a "DNA profile" of a codebase's coding style (naming conventions, formatting, comment voice, structural habits, error-handling philosophy, and idioms), then use it so new or AI-written code is indistinguishable from the existing author's code. Use this skill whenever the user wants to match a codebase's existing style, make AI contributions "blend in", "mimic" or "fingerprint" a coding style, capture house conventions, build a style guide from existing code, onboard to an unfamiliar repo's conventions, or check whether code "looks AI-generated" / has tells. Triggers on phrases like "match my coding style", "make it look like I wrote it", "codedna", "capture the conventions", "style guide from the code", "why does this read as AI-generated", "genetic map of the code", "mimic the codebase", and on any request to contribute significant code to an existing repo where matching house style matters.
aihxp/codedna · ★ 1 · Code & Development · score 77
Install: claude install-skill aihxp/codedna
# CodeDNA Version: 1.0.1 ## What this is, and why it matters AI-written code is usually correct but recognizably foreign. It carries tells: a comment on every line, variable names two words longer than the human would pick, a try/catch the author never would have wrapped, a helper extracted where the author inlines, cheerful comment prose where the author is terse. Individually small; together they read as "not written by the person whose name is on the rest of the repo." CodeDNA closes that gap. It studies a codebase the way you would study a writer's voice, writes the conventions down in a profile, and uses that profile so future contributions match. The goal is not "good code in the abstract"; it is code that reads as if the original author wrote it. One idea runs through everything: **most surface formatting is already enforced by tooling, so it is not where the fingerprint lives.** Indentation, quotes, and semicolons get rewritten the moment someone runs Prettier, Black, gofmt, or rustfmt. The fingerprint that distinguishes a person is in the choices no formatter touches: what they name things, how much they comment and in what voice, how big their functions get, early returns versus nested conditionals, how defensively they handle errors, and the small idioms they reuse. Spend your attention there. ## Three modes Read the user's intent and pick one. - **Map** (default): scan a codebase and produce `CODEDNA.md`. Triggered by "capture the style", "build the codedn