originalitylisted

Prior-art pass before writing something new — find what already does this, in the repo and outside it, then say plainly whether to reuse, wrap, or build and why. Auto-triggers on: "has this been done", "is there a library for", "should I build this", "prior art", "am I reinventing", "write a X from scratch"
Univers42/claude-deal-with-the-devil · ★ 3 · AI & Automation · score 69

Install: claude install-skill Univers42/claude-deal-with-the-devil

# Originality Original does not mean unprecedented; it means *knowing what exists and being deliberate about differing from it*. The expensive mistake is the fourth in-house implementation of something the stdlib has had since 2019 — written confidently, because nobody looked. This is `library-first` extended past the repo boundary. Ten minutes here routinely deletes a week. ## 1. Search inward first — the cheapest win ```sh .claude/tools/codemap.sh # where a symbol already lives .claude/tools/dupes.sh # blocks already repeated .claude/tools/facts.sh # what the project already scripts .claude/tools/scripts.sh list # the vetted external script registry rg -i '<the concept, and its synonyms>' ``` Search **behaviour, not your chosen name**. The existing implementation is called `normalize`, you were going to call it `sanitize`, and that is why you did not find it. Try three names. Check the dependency manifest too. The project very likely already pulls in a date library, a validator, a fetch wrapper, a retry helper — rung 4 of `rules/minimalism-ladder.md`. ## 2. Search outward — the ladder, in order 1. **The standard library.** Check it properly. `itertools`, `slices`, `Intl`, `<algorithm>`, `functools`, coreutils. Most "utility" code is a stdlib call with a different name. 2. **The platform.** The OS, the database, the HTTP layer, the runtime. A unique index beats an application-level duplicate check. `cron` beats a scheduler.