agentbox

Complete visual content system for Claude Code. Use when generating blog headers, infographics, editorial art, diagrams, or comics. Default aesthetic: light backgrounds, teal/burnt orange accents, hand-drawn sketch style.

Meta-skill for driving Blender via BlenderMCP: 3D modelling (box, hard-surface, boolean), digital sculpting, PBR material authoring, lighting, rendering, and scene assembly. Claude writes and runs bpy Python (execute_code), screenshots the viewport, inspects, and corrects — with GPU renders via blender-batch and interactive sessions via the gui-tools GPU sidecar. Use this whenever a task involves Blender, a .blend file, 3D modelling/sculpting/texturing, a Cycles or EEVEE render, or generating and processing 3D assets — even when the user does not name Blender explicitly (e.g. 'model a low-poly house', 'sculpt a creature head', 'make a PBR metal material', 'render this scene on the GPU', 'build a modular sci-fi corridor kit'). Not for 2D image editing (use imagemagick), text-to-image generation (use comfyui), or GIS map rendering (use qgis).

Real-time AI meeting copilot. Dual audio capture (system + mic), live transcription (faster-whisper local or Deepgram cloud), LLM coaching loop (Claude/GPT every ~35s), always-on-top overlay, session logging with markdown export. Use when the user says "meeting recap", "meeting copilot", "live transcription", "meeting insights", "coaching during meeting", or "record this call".

Expand and analyze URLs using Google Gemini 2.5 Flash URL Context API. Efficiently fetches, summarizes, and extracts information from up to 20 URLs per request with grounding metadata and source citations. Use when analysing or summarising specific URLs using Gemini's 1M-token context window.

Build world-class comprehensive reports with LaTeX, Python analytics, Wardley maps, diagrams-as-code (TikZ + Mermaid), Nano Banana AI infographics, multi-LLM research swarms, and automated quality control. Use when creating research reports, white papers, sector analyses, technical documentation, policy briefs, or any document requiring professional presentation with data-driven charts, citations, and strategic visualisations.

Plan UK ground-mounted (utility-scale) solar farms end-to-end: site suitability, optimal tilt and inter-row spacing (GCR), array capacity and annual yield, 3D layout with inter-row shadow review, generation forecasting, and DNO grid-connection feasibility. Composes the qgis (site/terrain/ALC/flood) and blender (3D array + shading) skills with PVGIS yield data, Open Climate Fix's quartz-solar-forecast, and OpenDSS grid modelling. Use whenever the task involves a UK solar farm / solar site / ground-mount PV array — assessing a field or parcel for solar, sizing capacity, choosing panel tilt/row spacing, estimating kWh/yield or land use, visualising an array layout, checking shading, forecasting generation, or DNO/grid-connection pre-checks (e.g. 'is this 30-acre field good for solar', 'what tilt and spacing for a solar farm near Peterborough', 'how many MW fit on 20 hectares', 'model the array layout and winter shadows'). UK-specific (latitude, diffuse light, ALC land grades, DNO). Not for rooftop domestic PV or

DeepSeek reasoning bridge. Use when the user says "ask deepseek", "consult with deepseek", "delegate reasoning to deepseek", or needs advanced multi-step reasoning. Invokes DeepSeek via MCP for complex problem-solving with structured Chain-of-Thought outputs.

Multi-repository coordination, synchronization, and architecture management with AI swarm orchestration. Use when coordinating across multiple repositories, syncing packages, or running org-wide automation.

Multi-source web research via the web-researcher-mcp Go server (tracks upstream v1.33.0) — you pick the search ENGINE and the trusted SOURCES, and every citation is a real, checkable link. ~26 MCP tools: web/image/news/academic/patent/structured search, search_and_scrape, sequential_search; domain search (clinical_search, legal_search/CourtListener, econ_search/World Bank+FRED, filing_search/SEC EDGAR); scrape_page (full PDF/DOCX/PPTX/YouTube/HN, not snippets); citation integrity (verify_citation, audit_bibliography, citation_graph, archive_source/Wayback, format_bibliography APA/MLA/BibTeX/RIS/CSL); a grounded `answer` that cites real sources; research session memory + export. Search LENSES restrict results to trusted domains (academic, clinical, legal, finance, government, journalism, devops, docs). Pluggable backends (Google PSE/Brave/Serper/SearXNG/SearchAPI/Exa) with failover. Use for reputation-attached research needing verifiable citations — the open, auditable, private counterpart to perplexity-resear

Background cron scheduler for recurring agent tasks. Start with /hermes-scheduler. Manages scheduled jobs that invoke Claude Code on cron/interval/one-shot schedules. Jobs persist across restarts. Inspired by NousResearch/hermes-agent scheduler. Use when scheduling recurring agent tasks on cron/interval/one-shot schedules.

Source-grounded knowledge corpus for the RuvNet ecosystem (21+ repos), living in the shared ruvector-postgres sidecar under namespace `ruvnet-kb` — same embedding space (Xinference bge-small-en-v1.5, 384-dim, ADR-015) and same memory_entries table as all other agent memory. Query via the search_ruvnet MCP tool or memory_search scoped to the namespace. Prevents hallucination about ruflo, ruvector, safla, agentdb, agentic-flow, and other RuvNet tools by grounding every assertion in indexed source code.

Answer questions about the owner's private personal email archive via the local Private Email MCP Gateway. TRIGGER when the user asks anything that requires looking in their personal email — "did I get an invoice from X", "when did Y email me about Z", "find the thread about my flight", "what's my account number with the bank", "summarize what HMRC sent me", "search for anyone called <name> in my email". The gateway runs locally and reasons over the mail archive with a local model. It exposes TWO access tiers: `ask_email` returns privacy-sanitized, schema-abstracted results (default, safe); and the break-glass pair `fetch_email_raw` / `fetch_email_by_ref` return RAW verbatim mail (real senders, subjects, dates, links, full text) and are gated by an owner-allow-listed Nostr pubkey capability token. SKIP for sending email, calendar, or non-personal/work mailboxes.

Cryptographically sign PDFs (invoices, contracts, letters, forms) with open-source tooling — pyHanko applying a PAdES/eIDAS-aligned digital signature plus an optional visible signature panel. TRIGGER when the user asks to "digitally sign", "e-sign", "add a digital signature to", "sign this invoice/contract/PDF", "make a signed PDF", "certify a PDF", or needs a signed document because a counterparty requires signed invoices/paperwork. Also covers generating a self-signed signing identity (X.509 key + cert), embedding an RFC-3161 trusted timestamp, verifying a signature (pdfsig), and the trust/upgrade path from self-signed to a fully-trusted (green-tick / eIDAS QES) certificate. SKIP for merely stamping an image of a handwritten signature with no cryptography (use imagemagick/pdftk), for filling form fields (pdftk), or for editing PDF content (pikepdf/pymupdf).

Ceramic Search API client for high-recall keyword web search with rich page extracts (up to 8k chars per result). Returns 10 structured results per query with title, URL, and long-form description. Exact-match keyword engine — best for LLM-augmented retrieval where you need dense source context, not a synthesized answer. Complements perplexity-research (authoritative primary sources) and web-researcher (verifiable citations + lenses).

Premium UI/UX design audit and refinement skill. Conducts systematic visual audits of existing apps and produces phased, implementation-ready design plans. Use this skill whenever the user asks to audit a UI, improve an app's visual design, make an interface feel more polished or premium, review design consistency, fix visual hierarchy, or refine spacing/typography/color. Also trigger when the user says "design review", "make it look better", "UI polish", "visual refinement", "design pass", "audit the design", or references making an app feel more professional. This skill is purely visual — it does not touch functionality, logic, or features. It elevates what exists.

Opinionated playbook and current reference for building full-stack web apps in Leptos (the Rust fine-grained-reactive framework). Use whenever the task involves Leptos: writing or editing components and the view! macro, signals/memos/effects, server functions (#[server]), SSR / hydration / islands, leptos_router, Resources/Actions/Suspense, or scaffolding with cargo-leptos / Trunk. Trigger on "Leptos", "view! macro", "RwSignal", "server function", "hydration mismatch", "signal disposed panic", "cargo leptos", "Thaw / leptos-use / Tailwind in Rust", choosing a Leptos component / styling / chart / icon library, reactive-state design, feature-flag (ssr/hydrate/csr) build breaks, WASM bundle-size tuning, or 0.6→0.7→0.8 migration. Covers tips, best practice, the ecosystem "eye candy", and the common traps. Pairs with open-design / design-audit (which own the visual layer; this skill owns the Rust + reactive layer).

Design production pipeline with structured intake, brand-system enforcement, and 5-dimensional quality gates. Use when generating web prototypes, landing pages, dashboards, mobile mockups, presentations, or any visual artifact. Combines structured questioning, design-system token enforcement, P0/P1/P2 checklists, and anti-slop critique. Supports 129 brand specifications and 64 surface types.

Three-API Perplexity client: Search API (/search) for structured web results with domain/date filters, Agent API (/v1/agent) for multi-step deep research with reasoning, and Chat Completions (/chat/completions) for quick sonar queries. Use when you need live web research, academic citation discovery, UK ecology/policy lookups, or deep multi-step investigations with provenance.

Audit and fix AI writing tells in prose, markdown, presentation content, and fiction. Removes Claude/LLM fingerprints at three levels: lexical (banned vocabulary, hedge words), structural (em-dash density, "The X" headings, negative parallelism, throat-clearing, transition overuse), and narrative (thematic over-explanation, embodied-emotion overuse, single-track plots, tidy resolutions, vague allusions, no fourth-wall breaks). Also provides generative principles for writing fresh prose that does not need sanitising. Enforces UK English spelling. Use when writing or editing public-facing content, presentations, documentation, articles, blog posts, tutorials, or short fiction. Invoke with: "sanitise this", "clean up the writing", "remove AI tells", "de-slop", "make this read human", "write a blog post", "draft technical content".

Unified dispatcher for 88+ skills. Use when you don't know which skill to invoke — describe your task and get routed to the optimal skill, agent composition, or MCP tool.

Ground agent reasoning in DreamLab's formal knowledge graph (5,975 OWL classes, Oxigraph/Whelk) via the pervasive ontology binding (PRD-020/ADR-112). Use when you want to ground or augment thinking in the ontology, check what the knowledge graph says about a concept, retrieve a budget-bounded provenance-scoped subgraph (ontology_ask), run read-only SPARQL, find class neighbours or shortest paths, or propose a governed enrichment. Read-pervasive, write-governed; budget-bounded and fail-open so it never bloats the context window or blocks a turn.

402-aware outbound payment pipeline

AI image and video generation using ComfyUI node-based workflow system with Stable Diffusion, FLUX, and other diffusion models via local or distributed GPU compute (Salad Cloud). Use when generating AI images or video with Stable Diffusion, FLUX, or other diffusion models.

Debug host web servers from inside Docker containers. Use when bridging HTTPS to HTTP, taking screenshots with Playwright, analysing with Chrome DevTools, or solving cross-origin security issues when accessing host development servers.

Unified Claude Code V3 + Agentic QE meta-skill with 114+ agents, EDD loop, and quality gates. Use when implementing features with TDD, debugging hard bugs (feedback-loop-first protocol), stress-testing designs, or running quality gate pipelines. Supersedes agentic-qe, reasoningbank-*, and pair-programming.

Create new Claude Code Skills with proper YAML frontmatter, progressive disclosure structure, and complete directory organisation. Use when you need to build custom skills for specific workflows, generate skill templates, or understand the Claude Skills specification.

Empirically optimize any existing Claude skill against a measurable reward signal using a closed SkillOpt loop (rollout → reflect → aggregate → select → evaluate) plus a noise-robust held-out A/B comparator. Use when you want a skill's prose tuned by evidence rather than eyeballed — e.g. raising an agent's success rate on a bounded, scoreable task — or to compare a single-optimizer arm against a mesh arm. Complements skill-builder (which authors skills); this one tunes them.

Write SEO-optimized blog posts, landing pages, and content improvements following Google's E-E-A-T and Helpful Content guidelines. Handles new content creation from a keyword or topic, and improving existing pages. Use when asked to "write a blog post", "create a landing page", "improve this page", "write content about X", "content for keyword X", "draft an article", "blog post about", "landing page for", "service page", "product page copy", "rewrite this page", "make this page rank better", "content brief", "how-to guide", "listicle", or any content creation or improvement task for a website.

AI-powered SEO/SEM automation suite with 6 specialised skills. Google Search Console integration, content writing (E-E-A-T), keyword research, meta tag optimisation, JSON-LD schema markup, and GEO (Generative Engine Optimisation) for AI citations. Use when doing SEO audits, keyword research, schema markup, or AI search citation optimisation.

Process and manipulate images with format conversion (PNG→JPG→WebP→GIF), resizing, cropping, filtering, batch operations, and image metadata extraction. Use for thumbnail generation, image optimisation, format conversion, watermarks, and bulk image processing tasks.

Foundation library for Logseq ontology manipulation with OWL2 DL TTL export. Use when manipulating Logseq ontology structures or exporting to OWL2 DL TTL format.

Validation, enrichment, and TTL generation for Logseq ontology with VisionClaw/WebVOWL compatibility. Use when validating, enriching, or generating TTL for Logseq ontology with VisionClaw/WebVOWL compatibility.

Browser automation, web scraping, visual testing, and WebGPU validation via the browsercontainer sidecar (chrome-devtools-mcp, 40+ tools). Use for navigating pages, clicking elements, filling forms, taking screenshots, executing JavaScript, accessibility snapshots, and WebGPU rendering validation. Visual access via VNC on port 5903. Hardware-accelerated Chrome Beta 149+ with NVIDIA Vulkan/ANGLE.

Comprehensive Wardley mapping toolkit that transforms any input (structured data, unstructured text, business descriptions, technical architectures, competitive landscapes, or abstract concepts) into strategic Wardley maps. Default render path is **Mermaid `wardley-beta`** (mmdc 11.15.0+) -- pure code, version-controllable, GitHub-native, no external service. Legacy OnlineWardleyMaps `.owm` and OWM->Mermaid conversion via tractorjuice/wardley-maps-mermaid tooling. Use when creating strategic Wardley maps showing component evolution and value chains for decision-making.

Summarize web content including YouTube videos with semantic topic links for Logseq and Obsidian. Uses Z.AI service (port 9600) for cost-effective summarization. Supports markdown, plain text, and note-taking formats. Use when summarising web pages, YouTube transcripts, or extracting key information from URLs.

AI-first development methodology with specification-driven artifact traceability. PRD → SPEC → ADR → TASK → CODE → REVIEW → DEPLOY pipeline with AI-native ADRs (model selection, prompt strategy, agent orchestration), LLM evaluation suites, guardrails specifications, and sprint-driven workflows. Optimised for Claude Code and Ruflo/RuVector. Use when you need SPEC/ADR/TASK artifact chain traceability, AI-native ADR categories, or sprint scaffolding.

Professional document preparation with LaTeX - compile papers, presentations, and technical documentation. Use when compiling LaTeX papers, Beamer presentations, or technical documentation.

Browser automation via the external browsercontainer sidecar (chrome-devtools-mcp). Use for navigating web pages, screenshots, accessibility snapshots, form-filling, JavaScript evaluation, and WebGPU rendering validation. Connects over MCP SSE.

End-to-end book publishing pipeline from markdown manuscript to arXiv/KDP/print-ready PDF. Orchestrates parallel agent swarms for LaTeX conversion, citation extraction, diagram generation, image upcycling, and visual verification. Use when publishing a book or preparing a manuscript for academic or commercial publication.

Convert markdown/HTML manuscripts to arXiv-compliant LaTeX with memoir class, biblatex citations, professional typesetting, and parallel agent swarm conversion. Use when converting a multi-chapter markdown book to LaTeX for academic publication, arXiv submission, or print-ready PDF.