LeeorNahum

Create, update, apply, and validate a repository's DESIGN.md, the token-and-prose file that is a repository's living source of truth for its visual identity. Use when building or restyling UI, components, layouts, colors, typography, spacing, or themes in a repository that has, or should have, a DESIGN.md, including using its tokens and identity to guide external visual assets such as a logo or marketing material that need to stay on-brand. Use when starting a design system for a repository, when defining or changing design tokens that DESIGN.md should record, or when reading DESIGN.md and keeping it in sync with the implemented UI.

Opinionated structure for a Context directory: the durable home for a project's plans, research, notes, schedules, program info, and shared memory as markup. Use when creating, saving, organizing, or reorganizing any durable documentation or planning material, when the user says to save, note, remember, capture, or document project material for later (even vaguely), never for personal reminders or in-chat preferences, when adding plans, meeting notes, research, or reference material to a repository, when reading or maintaining an existing Context directory and its index, or whenever a Temp-Context directory exists and needs draining into Context. Not for code documentation, READMEs, or comments that live with the source.

Subagent Spawn Book (SSB) is used before creating any subagent, whether the user asks for one, the agent decides to delegate, or a plan includes delegation or fan-out. Select a default model when none is named, name the session, then read the selected model's spawn reference.

Opinionated defaults and standards for building TypeScript web products and web-technology repos, covering monorepo layout, rendering, surfaces, routing, IDs, key generation, domains, origins, env contracts, branches, deployment, DNS, auth, data ownership, media, jobs, AI model routing, billing, admin, responsive layout, fonts, product UX, and release operating philosophy. Load this skill early and keep it loaded for any web work: whenever planning, deciding, designing, scaffolding, auditing, naming, reviewing, or implementing anything in a web app, marketing site, signed-in app, SaaS, extension, desktop, or mobile surface built with HTML, CSS, JS, or TS. This includes backend-only web product pieces, such as an API route, a reactive backend function, or a webhook handler, since env contracts, auth, data ownership, and deployment are opinions this skill holds regardless of whether a given file renders UI. When unsure whether it applies, load it. Almost every decision in a web repo touches an opinion it holds.

Create, design, refine, and validate Agent Skills that follow the open SKILL.md standard. Use when the user wants to create or author a new Agent Skill, improve an Agent Skill's triggering description, organize an Agent Skill's references, scripts, or assets directories, preserve key wording and meaning during Agent Skill edits, decide an Agent Skill's automatic-versus-manual invocation style, or asks about Agent Skill structure, naming conventions, or SKILL.md format.

Opinionated standards for building Paper Minecraft plugin repositories: repo shape and single-source versioning, code architecture, state and config ownership, packet and presentation policy, cross-play GUI constraints, release contracts, and idempotency rules. Use when creating a new Paper plugin repo, scaffolding its Maven build or package layout, deciding where a piece of state or config belongs, adding packet-based presentation or display entities, designing a GUI or command that must work on both Java and cross-play clients, or cutting a plugin release and version bump.

Inspect branch state and safely promote code between deployment branches. Use when moving changes between development, staging, and production branches, handling branch drift or divergence, or chaining promotion stages.

Review and improve visual, spatial, textual, and user-facing artifacts for coherent-but-wrong failures that violate human normalcy, including nonsensical wording, unsupported claims, wrong scale, impossible adjacency, unreadable hierarchy, duplicated patterns, meta leakage, and broken viewer flow. Always use this skill when creating or critiquing UI, decks, diagrams, docs, PDFs, signage, booths, copy, or other artifacts where local plausibility can hide global incoherence.

Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

Ask the user more useful questions when clarification, confirmation, unblocking, or sharper direction would help. Use whenever requirements are unclear, multiple paths remain, confidence is low, a real blocker appears, or the user implicitly or explicitly wants questions back. If the user mentions this skill, asks you to ask questions, or asks for a more interactive back-and-forth, prioritize loading and using this skill in that turn and keep its behavior active while it remains relevant. If the harness has a dedicated question tool or question UI, prefer using it so the user can answer inline during the same flow. Default to a single question per turn and never ask more than two at once. Prefer iterative followups where each answer shapes the next question, and make each question meaningfully useful.

Build or audit a body of work to rigorous, present-day correctness: ground in the live code, verify every external assumption at the source, and eliminate or never introduce anything broken, obsolete, duplicated, incoherent, or worked-around. Runs whole-repo when invoked bare, or appended to a specific task or focus (a feature, fix, subsystem, surface, or file set), where it holds that task and everything in its blast radius to the same standard so the work is done right in one pass rather than patched and audited later. This skill is strictly manual and must only be invoked by name. When invoked, load this skill and follow it for deep, whole-product correctness, cleanliness, and coherence work across backend, data, APIs, protocols, frontend, UI, UX, copy, configuration, environment contracts, integration points, and product flows, or for self-critical building and rewrites validated against up-to-date, authoritative references.

Keep Claude Code in sync with cross-tool Agent Skills and AGENTS.md by regenerating a managed block in CLAUDE.md. Run it whenever anything under .agents changes, such as a skill being added, removed, renamed, or having its name or description edited, and whenever a repository has an AGENTS.md or .agents/skills but no up-to-date CLAUDE.md, because Claude Code natively reads only CLAUDE.md and .claude/skills. It lists each skill's name, path, and description so Claude can read matching skills directly.