coven-cave

Generate and edit presentation slides as PPTX files. Also create LinkedIn carousels and manage reusable slide layouts. TRIGGERS - Use this skill when user says: - "create slides for [brand]" / "generate presentation for [brand]" / "make slides for [brand]" - "create a carousel for [brand]" / "linkedin carousel" / "make a carousel about [topic]" - "edit this pptx" / "update the slides" / "modify this presentation" - "create a new layout" / "add a layout to the cookbook" / "make a [type] layout template" - "edit the [name] layout" / "update the cookbook" / "improve the [name] template" - Any request mentioning slides, presentations, carousels, PPTX, or layouts with a brand name Creates .pptx files compatible with PowerPoint, Google Slides, and Keynote. Creates PDF carousels for LinkedIn (square 1:1 format).

Deterministic pipeline for authoring, verifying, exporting, and deploying OpenCoven Slidev decks (the weekly Show'n Spells system and one-off branded decks). Use whenever asked to build, rebuild, update, export, or publish a Slidev slide deck for OpenCoven — weekly meeting decks, launch decks, or explainer decks.

Universal MCP client for connecting to any MCP server with progressive disclosure. Wraps MCP servers as skills to avoid context window bloat from tool definitions. Use when interacting with external MCP servers (Zapier, Sequential Thinking, GitHub, filesystem, etc.), listing available tools, or executing MCP tool calls. Triggers on requests like "connect to Zapier", "use MCP server", "list MCP tools", "call Zapier action", "use sequential thinking", or any MCP server interaction.

Use when asked to build, draft, assemble, or verify a Coven Cave Craft — a versioned bundle of skills, prompts, workflows, and runtime capabilities extracted from a familiar's Roles. Trigger on "build a craft", "draft a craft", "bundle my roles", "package this loadout", or a chat brief that carries the crafts drafts API contract.

Design the information architecture for a CovenWiki — a source-grounded wiki generated from one local repository. Takes a repo file inventory and README evidence as JSON; returns a strict JSON outline (pages, navigation, concepts, coverage notes). Used by the covenwiki-generate orchestrator (CovenWiki Route B Phase 1); not a general-purpose summarizer.

Write one source-cited CovenWiki page from an outline entry plus hydrated source evidence. Takes page spec, outline context, and source excerpts as JSON; returns strict JSON (markdown, citations, coverageNotes, relatedPages). Used by the covenwiki-generate orchestrator (CovenWiki Route B Phase 1); every claim must cite hydrated evidence.

Use when asked to build, author, draft, or save a Coven Cave skill (a SKILL.md familiars load), or when a chat brief carries the skills build API contract. Trigger on "build a skill", "write a SKILL.md", "author a skill for", "turn this procedure into a skill", or a brief naming POST /api/skills/build.

Use when asked to sew, finish, or draft a Grimoire stitch — distilling a thread of pinned sources into one durable Knowledge Vault entry — or when a chat brief carries the stitches sew API contract. Trigger on "sew this stitch", "finish the stitch", "sew a Grimoire stitch", or a pinned-sources digest asking for one durable reference entry.

Use for worldbuilding, story canon, a world encyclopedia, stubbing characters/settings/factions/lore, and flagging contradictions.

Launch the Coven Cave app (Next.js web build) and drive a real browser to any surface — Home, Chat, Board, Canvas (Triage/Sketch), Calendar, Library, etc. — taking screenshots. Use when asked to run/start/show/screenshot the app or verify a UI change in the real app. Browser-driven via Playwright; no daemon required (demo mode supplies data).

Guides researchers through structured ideation frameworks to discover high-impact research directions. Use when exploring new problem spaces, pivoting between projects, or seeking novel angles on existing work.

Applies cognitive science frameworks for creative thinking to CS and AI research ideation. Use when seeking genuinely novel research directions by leveraging combinatorial creativity, analogical reasoning, constraint manipulation, and other empirically grounded creative strategies.

Describe the HTTP request you want, review the generated cURL command, and run it with secrets injected at runtime via op run.

Use the bin/board CLI to add, claim, and complete tasks on BOARD.md rather than editing the file by hand.

Use when building or reviewing any component that renders async/remote data (lists, tables, dashboards, search results, detail pages, feeds) in React/Vue/Svelte. Enforces the "every data component has five states" discipline: loading, empty, partial, error, ideal. Covers loading-indicator selection (spinner vs skeleton vs progress vs shimmer) against Nielsen 0.1/1/10s limits and the spinner-flash rule; the four empty-state types (first-run, no-results, permission-denied, deleted) with distinct copy+CTA; the five-class error taxonomy (user/network/server/permission/not-found); retry patterns (backoff+jitter, TanStack retry:3/min(1000*2^n,30000), when NOT to retry); toast-vs-inline-vs-banner-vs-dialog placement; the 404/500/offline trio; optimistic UI + Suspense; and WCAG 2.2 SC 4.1.3 announcements (role=status, role=alert, aria-live, aria-busy) + focus. Flags happy-path-only components, blank loading, raw stack traces, blame-the-user copy, and critical errors in auto-dismissing toasts.

Use when building or reviewing web forms (sign-up, checkout, settings, listing/create flows) in React with react-hook-form + zod (or Vue/Svelte equivalents) — covers validation timing (onBlur/onChange/onSubmit/onTouched), field-level error UX and wording, required-vs-optional indicators, progressive disclosure (multi-step, accordions, show-more), input-type/keyboard selection (inputmode, autocomplete tokens), password UX (reveal toggle, strength meter, allow paste), save-state/autosave indicators, server-error surfacing back to fields, and WCAG 2.2 form accessibility (aria-invalid, aria-describedby, role="alert", fieldset/legend, SC 3.3.1/3.3.3/3.3.7/3.3.8). Flags common anti-patterns like reset-next-to-submit, disabled-submit-hiding-errors, validate-on-every-keystroke, and blocking paste in password fields.

Run both the skill-structure gate and the CodeQL SDK security scan, then report a compact ready-to-submit status block.

Use X/Twitter safely: draft posts, look up public context, prepare approved-post workflows, and keep social publishing actions explicit.

Use when measuring or laying out multiline text in a web UI WITHOUT triggering DOM reflow (getBoundingClientRect / offsetHeight / measuring a hidden clone) — the @chenglou/pretext library (pure JS/TS, works across all languages, renders to DOM/Canvas/SVG/WebGL and soon server-side). Covers the prepare()/layout() split (one-time analysis + canvas measurement vs the cheap pure-arithmetic hot path; never re-run prepare() on resize), height + lineCount measurement for list virtualization/occlusion, layout-shift prevention when text loads late, dev/CI-time button/label overflow checks (browser-free), and userland masonry/JS-flexbox layout; manual line layout via prepareWithSegments + layoutWithLines / walkLineRanges / measureLineStats / measureNaturalWidth / layoutNextLine(Range) / materializeLineRange (multiline shrink-wrap, balanced text, text flowed around a float, canvas/SVG/WebGL rendering); the @chenglou/pretext/rich-inline helper for inline chips/mentions/code-spans with break:'never' + caller-owned extraWi

Use when building or reviewing any chart, graph, KPI card, dashboard, or analytics view — to pick the right chart type (bar/line/area/scatter/histogram/heatmap/sparkline/small-multiples), lay out KPIs (overview-then-detail, F-pattern), colour data safely (sequential/diverging/categorical, Okabe-Ito & Viridis, never colour-only), design hover/brush/drill-down/cross-filter interactions, handle time-series (zero baseline, time zones, aggregation, missing data), cover loading/empty/error + real-time refresh states, make charts accessible (alt text, data tables, keyboard series nav), export to PNG/PDF/CSV, and choose a React charting library (Recharts vs Nivo vs Victory vs visx vs D3 vs Observable Plot). Reach for it on any request mentioning "chart", "graph", "dashboard", "data viz", "KPI", "metric", "plot", "which chart", "colour palette", or when a PR adds a visualization.

Use when choosing, comparing, or borrowing from an existing design system rather than building components from scratch: a map + decision guide covering Material 3, Fluent 2, Apple HIG, GitHub Primer, IBM Carbon, Ant Design, Chakra UI v3, Mantine, HeroUI (ex-NextUI), Radix + shadcn/ui, and Adobe Spectrum / React Aria (plus SLDS, Atlassian, Polaris, Base Web). Classifies each into four tiers (platform language you conform to / product DS / installable community kit / headless own-code substrate) and states its license, framework support, token architecture, and a11y posture, with a "reach for which DS when" flow and live adoption signals (npm/stars; Polaris repo archived; NextUI to HeroUI; Ant agent docs). Encodes the CovenCave matrix: borrow semantic-first tokens (Primer), a11y discipline (Carbon), and the ownership model (Radix/shadcn); avoid Material dynamic color, Ant density, light-mode-first defaults, and vendor skins. Hands-on building belongs to sibling skills.

Use when implementing UI motion in React with Motion (formerly Framer Motion; npm `motion`, import from `motion/react`): declarative animate/initial/exit, spring-vs-tween transitions, layout & shared-element transitions (layout / layoutId / LayoutGroup), AnimatePresence enter/exit, staggering and imperative sequences (useAnimate / useAnimationControls), cross-device keyboard-aware gestures (hover/tap/focus/pan/drag/Reorder), scroll-linked & scroll-triggered motion (useScroll / useInView / useTransform), motion values as a no-re-render reactive primitive (useMotionValue / useTransform), and production concerns: prefers-reduced-motion a11y (MotionConfig reducedMotion / useReducedMotion), transform-only hardware-accelerated performance, and choosing Motion vs CSS vs React Spring vs GSAP vs the Web Animations API. Includes recipes: modal, drawer/sheet, toast stack, list reorder, page transitions, shared-element hero image, spinner, scroll progress, parallax, stagger-in lists.

Use when building, reviewing, or debugging any mobile/responsive web UI (or a PWA / web view meant to feel native) — sizing touch targets, fixing the 100vh/notch/on-screen-keyboard bugs, making layouts adapt with container queries instead of device-pixel breakpoints, choosing bottom-nav vs hamburger / sheet vs modal / action-sheet vs dropdown, wiring the right mobile keyboard (inputmode/enterkeyhint/autocomplete), handling touch gestures with Pointer Events + touch-action, honoring hover-vs-touch, or hitting mobile Core Web Vitals (LCP/INP/CLS) with responsive images and lazy-loading. Reach for it on any cue like "mobile", "touch", "responsive", "breakpoint", "safe area / notch / Dynamic Island", "svh/dvh", "container query", "tap target", "hamburger vs tab bar", "PWA/offline", "keyboard covers input", "swipe/drag", or a PR that ships a phone layout. Framework-agnostic; pairs with wcag-a11y-audit and framer-motion-patterns.

Use when building or editing React UI with shadcn/ui, Radix Primitives, and Tailwind — the "open code" pattern where components are copied into your repo (components/ui/*) via the shadcn CLI and owned/edited by you, not installed as a runtime package. Covers the cva + tailwind-merge + clsx (cn) styling stack, the Radix headless a11y substrate (focus management, ARIA, keyboard nav, portals) exposed through asChild/Slot and data-slot/data-state, and concrete recipes for Button, Dialog, AlertDialog, DropdownMenu, Popover, Tooltip, Combobox (Popover+Command), Select, DataTable (TanStack Table), Sheet, Drawer (Vaul), and Form (react-hook-form + zod). Also decides when to prefer shadcn vs React Aria Components, HeroUI, Base UI, Chakra v3, Mantine, MUI, or Ant. Do NOT use for non-React frameworks (use shadcn-vue/svelte ports), deep design-token theming (tailwind-design-tokens), rich motion (framer-motion-patterns), or form UX depth (form-ux-patterns).

Use when building or theming a Tailwind CSS v4 (or migrating a v3) UI with design tokens, dark mode, or multi-theme support — covers the CSS-first `@theme` model, OKLCH color scales, the two-tier primitive/semantic token architecture (shadcn `background`/`foreground` pairs bridged with `@theme inline`), class vs media vs `data-theme` dark mode with FOUC-safe toggling, arbitrary values, `@source inline` safelisting, and v3→v4 migration; framework-agnostic (React/Vue/Svelte/Astro/HTML) with shadcn-specific glue called out.

Use when auditing or building any web/desktop UI for accessibility, wiring up ARIA/keyboard/focus, choosing color-contrast and target-size values, fixing screen-reader or focus-management bugs, honoring reduced-motion, or setting up automated a11y tests (axe-core, Playwright, jest-axe, eslint-plugin-jsx-a11y). This is the framework-agnostic WCAG 2.2 Level AA checklist and manual-testing playbook — reach for it whenever someone says "accessible", "a11y", "WCAG", "screen reader", "keyboard nav", "focus ring", "contrast", "ARIA", "compliance", or when a PR adds interactive components (modals, menus, tabs, drag-and-drop, forms, toasts) that must work without a mouse and with assistive tech.

Retrieve and inject secrets with op, keeping only op:// references on disk and never displaying secret values.

Audit likely large cache and build directories, then delete reproducible junk with exact paths after making blast radius explicit.

Drive a PR through triage, review, prepare, merge, and post-merge modes, answering the checkpoint questions before each transition.

Inspect a coding-agent session's history before acting, then send one bounded, state-aware instruction and require evidence before declaring completion.

Isolate each agent in its own worktree, claim the branch with cv-claim, and rely on pre-commit/pre-push hooks to block unsafe writes.

Create a daily.dev feed, follow goal-relevant tags, and run scheduled learning loops that fetch, note, and share insights autonomously.

Check archive freshness with status/doctor, run bounded search and channel slices, and use read-only SQL for exact counts before ever mutating the DB.

Diagnose disk hotspots, check per-repo activity, then remove only safe regenerable artifacts and report the space freed.

Run ebook-convert to translate between EPUB and PDF, then validate output size and structure before saving to research/sources/.

Route all HeyGen usage through MCP, the heygen CLI, or the OpenClaw plugin's v3 pipeline — never raw api.heygen.com calls.

Pick a sensible default model, pass media inputs straight to flags, and submit with --wait so the CLI blocks and prints the result URL in one shot.

Discover exact team/project/label/state IDs and search for duplicates before executing one focused GraphQL mutation.

Inspect existing components and conventions first, define the UX contract, then implement idiomatic Lit with native semantics, polished states, and verified accessibility.

Compose piped Lobster commands or .lobster workflow files to run deterministic multi-step automations with approval gates and persistent state.