skills

This repository bundles seven skill libraries that give AI coding agents concrete, checkable rules for different kinds of work. Each library is a standalone Claude Code plugin with its own README, changelog and licence, and can also be installed through a generic `npx skills@latest add` command. The stated design goals are scoped passes, one check per rule, context loaded on demand, and computed numbers (for example contrast ratios) rather than guessed ones. The libraries are: - kanso (code): anti-dilution rules for AI-written code, including audit, council (multi-agent consensus review), refactor, deep structural audit, task sharpening, commit splitting, PR descriptions, AGENTS.md/CLAUDE.md pruning, session handoff, and prompting guidance. - mimesis (prose): strips machine-made tells from human-facing writing and compiles dense instructions for machine readers, with skills for human-sounding text, concision, five named tones, and design-tell detection. - swarm (orchestration): fans work out to parallel agents, plans workstreams, runs planned work through cheaper models under a cost ceiling, best-of-N selection, and parallel critic passes. - flow (design): refines layout, typography, colour, controls, accessibility, copy and motion, with focused motion and accessibility skills. - delight (product feedback): refines feedback, celebrations, character reactions and progress, with focused character and habit skills. - yum (design research): finds, compares and adapts design references. - litmus (QA): plans, runs or repairs scoped web and mobile QA passes, plus diagnosis of intermittent failures without retry-until-green. Installation is offered both as a plugin marketplace for Claude Code and as a single npx command; the README notes that the standalone kanso, mimesis and swarm repositories remain live. The project is MIT licensed.