72 curated skills + the whole ecosystem.

One-command installer for the full agent memory stack: cognee 1.0.x knowledge graph (pghybrid), LiteLLM proxy with rotation, cognee-api watchdog cron, nightly ingest pipeline, CouchDB 3 single-node document store, and a filesystem markdown vault wired to Obsidian. Install order is encoded in the 11 numbered scripts; the top-level install.sh --detect-first orchestrator runs only what is missing. Use when provisioning a fresh Linux box or replicating a production-grade agent memory stack anywhere.

Cost-tuned wiring of Understand-Anything for Hermes (V's agent CLI). Pins each of the 9 multi-agent pipeline workers to the cheapest model that meets its quality gate (Haiku for mechanical extraction, Sonnet for synthesis), enforces a per-run USD budget cap via Hermes-headless invocation, and documents the integration…

Convert any video or audio file to colored ASCII art MP4/GIF animations. Supports frame extraction, palette mapping, and batch rendering for terminal-art creative projects.

Generate images, video, and audio with ComfyUI — install, launch, manage nodes/models, run workflows with parameter injection. Uses the official comfy-cli for lifecycle and direct REST/WebSocket API for execution.

Cognee is a self-hosted knowledge-graph retrieval stack (Postgres 17 + pgvector + graph store) that turns your documents and conversation history into a queryable episodic memory for AI agents. Boots from docker-compose in under 60 seconds.

Official Scrapling 0.4.6 skill, ported from D4Vinci/Scrapling/agent-skill. Use when scraping needs anti-bot bypass (Cloudflare Turnstile), stealth browsing, spider crawls, or adaptive element relocation. Complements the legacy `scrapling` skill which documents X/Twitter syndication tricks specific to our fleet.

Pick the right Ollama model for a low-VRAM GPU (≤4 GB) when offloading LLM work from a paid API (Z.AI, OpenAI) to local. Avoids the 'I'll just use the latest Gemma' trap — the newest models often DON'T fit on small consumer GPUs even at Q4. Validated 2026-04-27 on a GTX 1650 Ti (4 GB) for Cognee LLM offload. Use when migrating cognee/embeddings/agent inference to local hardware, or when picking an Ollama model on any small GPU before pulling 9 GB of weights you can't run.

Local text-to-speech using Kokoro 82M. Generates natural female speech at 2.6x realtime on GPU, uses only ~1GB VRAM. Best local TTS option for constrained GPU hardware (4GB VRAM). Proven winner over Qwen3-TTS, Voxtral, Bark, Piper, WhisperSpeech, and eSpeak-NG.

Create 3Blue1Brown-style math and algorithm explanation videos with Manim Community Edition. Render LaTeX equations, geometric proofs, and animated code walkthroughs programmatically.

Run quantized GGUF models locally using llama.cpp for CPU/GPU inference. Includes HuggingFace Hub model discovery, context-length tuning, and OpenAI-compatible server mode.

Keep an AI agent's skills EVERGREEN. Install this once and your cookbook's skills stay current automatically — with atomic apply + auto-rollback so a bad skill version can NEVER leave your agent broken. This is the thin reconcile client for Recipes (recipes.wisechef.ai): it rides your host's existing scheduler (Hermes cron / Codex / Claude), pulls only the skills that changed (content-addressed delta, CDN-fronted), and atomically swaps them into your skills dir. On any health regression it auto-reverts to the last-known-good snapshot. Invoke when asked to 'keep my skills up to date', 'stop skill rot', 'set up cookbook auto-sync', or 'install the Recipes reconcile daemon'. Pull-only (uses your own API key, no inbound auth); 304-cheap polling; self-updating (ships as a skill in the cookbook it manages, so it updates itself).

Solo-operator AI agent that runs your business while you sleep — morning briefing, marketing, code, tickets, weekly review.

One-command bootstrap that installs the Maestro solo-operator agent framework plus 4 core dependencies: morning-brief, marketing-engine, code-dispatch, and health-watchdog.

Bidirectional sync between a headless VPS filesystem and an Obsidian LiveSync CouchDB vault. Uses filesystem-livesync daemon — no Obsidian desktop app needed on the server.

Automate your one-person company: daily morning brief, marketing pipeline, code task delegation, ticket triage, and weekly MRR retro — all from a single chef command.

Summarize any content via the summarize CLI — URLs, PDFs, images, audio files, YouTube videos. Multi-provider LLM support.

Headless-browser scraper with persistent cookie management via a per-skill systemd-user proxy daemon. Playwright-powered with Chromium; handles JS-heavy sites, login walls, and rotating sessions automatically.

Local, offline speech-to-text — 4-6x faster than OpenAI Whisper, 99+ languages, VAD included.

Pull GA4, GSC, Meta Ads, and TikTok Ads data for any client and generate a branded PDF in under a minute. White-label-ready.

Battle-tested protocol for coordinating 2+ AI agents on Discord without infinite loops, runaway costs, or silent failures. Includes marker protocol, n-turn anti-loop budget, domain routing rules, and a supervisor auto-action mandate.

Scan the local Hermes skill hub (~/.hermes/skills/) to discover installed skills before authoring a new one. Invoke when asked 'is there an existing skill for X?', 'check Hermes before authoring', or 'search local skills for Y'. The method walks ~/.hermes/skills/**/SKILL.md, parses YAML frontmatter (mirroring the parse_skill_md approach in scripts/harvest_cookbook.py), and fuzzy-matches the query against name, description, and tags via token overlap scoring. Output is canonical JSON: {hub, query, results[{name, description, version, install_command, source_url, license, match_score}], elapsed_ms, errors}. Pure filesystem scan — no network calls required. Complements hub-search-openclaw and hub-search-claude-code for exhaustive pre-authoring discovery.

Multi-platform content marketing agent (Douyin, Xiaohongshu, Kuaishou, Bilibili, WeChat Channels, TikTok, YouTube, Facebook, Instagram, Threads, X, Pinterest, LinkedIn). Thin MCP-first wrapper around the open-source AiToEarn project — publishes, engages, and tracks monetization (CPS / CPE / CPM payouts) across the channels above from one MCP server. Use when the team ships content across ≥3 platforms and wants one tool to publish/engage/monetize instead of N per-platform recipes. ICP: 1-person AI owners and 1-7 person builder teams running multi-platform content for clients.

Stress-test the team's plan, idea, or decision in attack mode — sort proposals into gold / trash / directionally-right-but-flawed and push each to bulletproof. Different from premortem (which imagines the plan already failed and works backward) — ruthless-mentor goes line-by-line through stated decisions and attacks each one specifically. MANDATORY TRIGGERS: 'be my ruthless mentor', 'be ruthless', 'stress test everything', 'if my ideas are trash say why', 'where am i full of shit', 'push back hard', 'don't be polite', 'attack mode', 'where am i wrong', 'bulletproof this'. STRONG TRIGGERS: 'poke holes', 'tear this apart', 'be honest', 'no sycophancy', 'push back where i'm wrong'. DO NOT trigger on: simple feedback requests, factual questions, premortem requests (load premortem instead). DO trigger when the user has expressed multiple opinions / made multiple decisions in one turn and asks for honest pushback.

Author a goal-execution plan-doc that is ready to paste into your goal/issue tracker as N discrete, falsifiable goals. Use when the conversation has produced a punch-list of decisions + scope and the next step is 'ship it as a tracked sprint with budget, phases, premortem, and acceptance gates.' Different from a generic markdown plan: plan-for-goal specifically closes with pre-formatted goal seeds and a single-paste 'go <codename>' trigger so the same plan-doc drives execution. Aliased as `golazo` (Polish/Spanish slang for spectacular goal — the plan IS the goal).

Search the OpenClaw plugin registry (via npm) to discover published plugins before authoring a new one. Invoke when asked 'is there an existing skill for X on OpenClaw?', 'check OpenClaw before authoring', or 'search the OpenClaw registry for Y'. The method shells out to `npm search --json --searchlimit=20 <query>` and filters results to packages whose name starts with `@openclaw/` or contains `openclaw-plugin`. An optional preview path via the gitlawb HTTP API (gitlawb.com/api/, alpha) is documented but not required. Output is canonical JSON: {hub, query, results[{name, description, version, install_command, source_url, license, match_score}], elapsed_ms, errors}. Complements hub-search-hermes for exhaustive pre-authoring discovery.

preview — Codex CLI plugin API not yet programmatically queryable; filesystem-only fallback ships today. Scanner skill for LLM agents that need to discover locally installed Codex skills before duplicating effort or authoring a new capability. Walks ~/.codex/skills/**/SKILL.md as the only reliable discovery path. If the codex CLI is on PATH and supports --list-plugins --json (probed at runtime; non-zero exit tolerated), its output augments the filesystem results. When the CLI API is absent or returns a non-zero exit, the gap is reported via the errors[] field rather than treated as a fatal failure. Emits the same standard JSON envelope used by the other three hub-search skills so agents can union results across all hubs.

Scanner skill for LLM agents that need to discover locally-installed Claude Code skills and community npm plugins before installing a duplicate or authoring a new skill from scratch. Walks ~/.claude/skills/**/SKILL.md (default skills directory) and the XDG variant at ~/.config/claude/skills/, then queries the npm registry for packages in the claude-plugin-* namespace. If the Hermes CLI is on PATH, augments results with hermes plugins search <q>. Emits a standard JSON envelope compatible with hub-search-hermes and hub-search-openclaw so agents can union results across all four hubs. Missing skills directories are treated as empty results, not errors.

Run customer interviews that extract real signal — ask about past behavior not your idea, build commitment ladders, and avoid leading questions that produce false positives.

Turn any codebase or knowledge base into an interactive knowledge graph using Understand-Anything (Claude Code plugin, MIT-licensed). Produces a JSON graph, interactive dashboard with guided tours, diff impact analysis, and an auto-generated onboarding doc.

Run pre-launch checks before going live: verify DNS resolves to your tunnel, origin is current commit, Alembic head matches, and smoke tests pass on the public URL.

Diagnose Railway services returning 404 before you file an outage ticket — distinguishes the ~3-second cold-start wake window from actual deployment failures.

Build production-quality MCP servers in Python (FastMCP/Pydantic) or TypeScript (MCP SDK/Zod) using a 4-phase process: Research, Implement, Review, Evaluate.

Build a knowledge graph from any codebase, docs, or paper — generates interactive HTML, queryable JSON, and an audit report exposing god nodes, communities, and surprising connections.

Build and query an interlinked markdown knowledge base using Karpathy's LLM Wiki — auto-link pages and surface connections across your notes and docs.

Debug failing GitHub Actions checks on any PR — inspect logs with gh, summarize the root cause, draft a fix plan, and implement only after explicit approval.

Analyze any draft or idea against the Minto Pyramid Principle and deliver actionable restructuring recommendations. Extracts the pyramid (one-sentence answer, 2-4 MECE arguments, evidence per argument), diagnoses gaps, and delivers the fix as a visual HTML artifact: the pyramid rendered as a hierarchy, the exact opener to use, and a numbered restructuring plan. Use before any drafting, editing, or rewriting skill so the structure is sound before voice work begins. MANDATORY TRIGGERS: 'minto this', 'run minto', 'apply minto', 'minto-ify', 'pyramid this', 'build the pyramid', 'pyramid principle this', 'answer-first this', 'structure this as a pyramid'. STRONG TRIGGERS: 'what is the takeaway', 'is this MECE', 'does this hold up logically', 'pressure-test this', 'give me the one-sentence answer'. Do NOT trigger on requests to draft content, edit voice, or compress to a platform.

Automate TikTok slideshow marketing for any app or product. Researches competitors, generates AI images, adds text overlays, posts via Postiz, tracks analytics, and iterates on what works. Use when setting up TikTok marketing automation, creating slideshow posts, analyzing post performance, optimizing app marketing funnels, or when a user mentions TikTok growth, slideshow ads, or social media marketing for their app. Covers competitor research (browser-based), image generation, text overlays, TikTok posting (Postiz API), cross-posting to Instagram/YouTube/Threads, analytics tracking, hook testing, CTA optimization, conversion tracking with RevenueCat, and a full feedback loop that adjusts hooks and CTAs based on views vs conversions.

Analyze any codebase with GitNexus — index dependencies, call chains, clusters, and execution flows to accelerate bug hunting and architecture reviews.

Ship new agent skills from scratch, optimize existing ones, and benchmark triggering accuracy with variance-based evals — all in one workflow.

Diagnose and restore OpenClaw system failures through 7 structured phases: Triage → Evidence → 5 Whys → Restore → Prevent → Monitor → Document.

Iterate on SVG icons and logos using vision-AI feedback in a test harness before touching production — handles score variance and gates on human eyeball review.

Search the web and get clean, AI-ready results via Tavily API — deep research mode, news search, and agent-native output format built for pipelines.

Fix Stripe webhook handlers crashing after a Python SDK 15.x upgrade — converts dict-style .get() calls to attribute access so payments stop silently failing.

Rotate Stripe prices on a live product without breaking in-flight checkouts — creates the new Price, swaps the default, and deactivates the old one safely.

Research any startup idea and generate a full kit: market analysis, competitor map, product specs, and a high-fidelity landing page — ready to pitch or build.

Generate architecture diagrams and hero images from any codebase using GitNexus + D2 + Playwright — ship visual READMEs, blog assets, and architecture overviews.

Map your competitive alternatives, unique attributes, and best-fit customers to the right market category — ship a defensible positioning canvas in one session.

Design MVPs and validated learning experiments using Build-Measure-Learn — define actionable metrics, run pivot-or-persevere decisions, and cut vanity metrics.

Render marketing reels, explainers, and animated carousels as deterministic MP4 from HTML + CSS + GSAP — headless Chrome + FFmpeg pipeline, no React, no bundler.

Generate vertical animated invitation cards (wedding, birthday, event) as 1080x1920 H.264 MP4 under 4MB — plays inline on WhatsApp, Instagram Stories, and email.

Craft irresistible offers using the Value Equation, bonus stacking, risk-reversing guarantees, and the MAGIC naming formula — includes starving-crowd targeting and ethical scarcity.

Read and edit DNS records on Hostinger domains via their Developer API — no hpanel login needed. Validated on live CNAME swaps for Railway custom-domain updates.

Ship React/Next.js/Tailwind UIs that pass accessibility checks — mobile-first layout, 70-20-10 color theory, performance targets, and memorable design patterns.

Model software around bounded contexts, aggregates, and ubiquitous language — map domain events, define context boundaries, and ship code that mirrors your business.

Generate niche pain profiles, early-adopter personas, and a competitor map with real pricing — includes vitamin/painkiller verdict, cobbled-solution audit, and moat assessment.

Scrutinize code for security holes, race conditions, and bad patterns — structured output with Blocking, Required, and Suggestions severity tiers for every finding.

Generate project ideas through creative constraints — works for code, hardware, art, writing, and tools. Use when you have skills but no direction.

Review code with a security-first lens — catch SQL injection, race conditions, and quality anti-patterns using structured guidelines for thorough PR reviews.

Write readable, maintainable code using disciplined naming, small functions, and clean error handling — covers SRP, comment discipline, formatting, and unit testing.

Structure software so dependencies point inward from frameworks → use cases → entities — covers hexagonal architecture, SOLID, ports and adapters, and component principles.

Fix Caddy serving the homepage on every route for static multi-page sites (Astro, Next, Hugo, Eleventy) — diagnoses SPA fallback traps, port conflicts, and permission issues.

Ship a stack-agnostic brand meta-repo with design tokens, logos, and Tailwind preset — auto-opens bump PRs in every consumer repo (Astro, React, Vite) on tag release.

Explore requirements and design before implementation — run structured brainstorming for features, components, or behavior changes when scope is ambiguous.

Run a full SEO audit on any URL — Lighthouse scores, crawl for broken links and missing meta/alt text, schema.org check, and optional competitor diff. Outputs PDF or markdown.

Spin up a branded client dashboard with live GA4, GSC, Meta Ads, and TikTok metrics in under 60 seconds. White-label, password-protected — replace Whatagraph today.

Generate personalized cold emails with MX deliverability checks — feed a CSV of prospects and ship outreach at scale with optional SMTP send.

Convert a discovery call transcript into a branded SOW PDF with scope, pricing tiers, and case studies — Claude-powered extraction, weasyprint-rendered output.

Build ETL pipelines with schema pre-flight checks, cursor-based pagination recovery, and idempotent inserts — handles 50+ connector patterns across SQL and REST sources.

Generate images from text prompts using Stable Diffusion, FLUX, or OpenAI gpt-image-1 — picks the right backend based on cost, style, and resolution. Returns URL or file path.

Generate structured PR descriptions from git diffs using Conventional Commits format. Analyzes branch diff, commits, and status to produce ready-to-paste PR descriptions.

Edit PDFs with natural-language instructions using the nano-pdf CLI. Modify text, fix typos, update titles, and make content changes to specific pages without manual editing.

Search 100k+ indexed AI skills, tools, agents, and apps by natural language task description. "Google for AI capabilities.