Adds two community skills to the list.
Problem it solves: by default an LLM returns the statistical average of its training data — the obvious, top-of-StackOverflow answer. This skill forces Claude past that: it names and bans the default solution, expresses the problem as a TRIZ contradiction, reframes it, and collides the problem with a distant domain to surface one genuinely non-obvious solution — delivered with its insight and an honestly named trade-off.
Who uses it: developers, architects, and PMs who are stuck on the "obvious" approach and want a better one (API performance, distributed-systems design, onboarding, naming, strategy).
Example: /creativity our API is too slow → instead of "add caching + CDN", it proposes eliminating the request entirely (pre-generate the bounded response space at build time), with the trade-off named.
Problem it solves: AI-generated Russian text has a recognizable robotic "smell" — bureaucratese, monotone rhythm, no authorial voice. This skill rewrites it into natural, living Russian. It's grounded in two classics of Russian editorial style (Nora Gal's The Word Living and Dead, Maxim Ilyahov's Write, Cut) plus the metrics AI-detectors actually use (perplexity, burstiness). The goal is good writing; passing detectors is a side effect. Who uses it: Russian-speaking copywriters, marketers, and writers producing content with AI. Has rewrite / write / audit modes, plus tone and format options.
Both skills are MIT-licensed, documented (SKILL.md + reference files), and tested in Claude Code. They provide standalone value and don't require any external paid service.