Claude Skills by yanacuti1121

Plan a huge chunk of work — more than one agent session can hold — as a shared map of investigation tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.

Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.

Use when needing to interact with MCP servers on-demand via CLI without permanently loading integrations, discover available tools/resources on an MCP server, or call MCP tools from the command line. Triggers on: 'mcp cli', 'mcp tool call', 'call mcp server', 'discover mcp tools', 'mcp on-demand', 'mcp command line', 'mcp binary', 'list mcp tools', 'invoke mcp', 'mcp without integration'.

Add persistent, intelligent memory to AI agents with Mem0 — add/search/update/delete memories per user/agent/session, supports vector + graph + key-value storage, integrates with LangChain, CrewAI, OpenAI Assistants, and any LLM.

MemGPT virtual context — OS virtual-memory analogy for LLM context management. Two-tier (main context = RAM, external store = disk), page-in/page-out tools, archival/core memory split. Foundation for Letta, Mem0, multi-session persistence. Sources: rohitg00/ai-engineering-from-scratch (Apache-2.0).

Use when implementing a custom malloc/free-style memory allocator from first principles — not when just using a language's built-in allocator. Triggers on: 'write a malloc implementation', 'build a memory allocator', 'free list allocator', 'buddy allocator', 'slab allocator', 'implement my own malloc/free'. Covers free-list design, block metadata, coalescing, alignment, and allocator strategy tradeoffs.

Session-end memory garbage collector. Promotes valuable L2 session facts to L1 atomic memory, wipes L2, and rotates oversized audit logs. Run at end of session to prevent context inflation and storage bloat.

Node.js memory leak detection and heap profiling. node-memwatch heap diffing, V8 heap snapshots, GC event monitoring, heap growth trending, and leak remediation patterns. Sources: lloyd/node-memwatch, Node.js v8 module.

Review all memory layers, detect stale/duplicate/conflicting entries, and propose promotions between layers. Use when the user says "clean up memory", "review what I've saved", "memory is getting messy", "what's in my memory", "organize my notes", "promote this to CLAUDE.md", "memory audit", or after a long session with many /remember calls. Do NOT modify any files without explicit user approval.

Local-first AI memory system — 96.6% R@5 trên LongMemEval, verbatim storage (không lossy summary), MCP server 29 tools, ChromaDB/Qdrant/pgvector backends. Zero API calls.

Implement append-only cryptographic audit logs using Merkle Mountain Ranges (MMR). Any tampering with historical entries causes immediate root hash drift. Integrates with YAMTAM secure-logger.js for agent action immutability.

Merkle tree construction for tamper-evident audit logs. Build Merkle root from audit log hashes, proof generation and verification, incremental append, and integrating with secure-logger.sh for L0 hash-chain integrity. Sources: miguelmota/merkletreejs.

Generating system diagrams as Mermaid text from agent state. Flowcharts, sequence diagrams, state machines, Gantt charts, and ER diagrams — all rendered from plain text without a browser. Sources: mermaid-js/mermaid, mermaid-js/mermaid-cli.

BitNet.cpp — official inference framework cho 1-bit LLMs: 1.37x-6.17x speedup, -55-82% energy. Chạy 100B model trên single CPU 5-7 tok/s. ARM + x86.

Microsoft eXecution Container — sandboxed execution cho untrusted code (model output, plugins, tools). Policy-driven, cross-platform (Win/Linux/macOS), TypeScript SDK, Rust binary.

Microsoft VibeVoice — open-source Voice AI: ASR 7B (60-min audio 1 pass, 50+ ngôn ngữ), TTS 1.5B (90-min, 4 speakers), Realtime 0.5B (300ms latency). HuggingFace.

Browser agent mã nguồn mở của Microsoft Research — agent viết Python/Playwright script thay vì click từng bước, script có thể reuse. SOTA 86.7% trên Online-Mind2Web.

Building sub-10MB base images for agent sandboxes. BusyBox-based rootfs, static binary containers, distroless patterns, multi-stage builds, and attack-surface minimization. Sources: progrium/busybox, GoogleContainerTools/distroless.

Simplify an overloaded UI by removing visual noise, establishing whitespace discipline, and reducing to essential elements. Use when the user says "too busy", "simplify this", "less is more", "cleaner look", "feels overwhelming", or "too many things on screen". Do NOT use when the user wants to add features or content — this skill removes and simplifies only.

Parallel multi-agent mission loop using yamtam-rt — decompose task into subtasks with deps, dispatch waves, collect reports, apply changes. Use when yana-classify returns route: complex. Triggers on: 'mission run', 'run mission', 'dispatch agents', 'parallel agents', 'multi-agent task', 'orchestrate agents', 'coordinate agents', 'agent mission', 'wave dispatch', 'mission loop', 'dispatch loop', 'parallel dispatch', 'yamtam-rt mission'.

Design and implement ML operations — model registry, serving patterns, deployment strategies (shadow/canary/blue-green), drift detection, feature stores, retraining triggers, and prediction monitoring. Use when asked to "deploy a model", "model registry", "MLflow", "feature store", "drift detection", "retrain trigger", "shadow mode", "model versioning", "serving infrastructure", or "ML pipeline". Do NOT use for: prompt engineering or RAG pipelines — see prompt-engineering and rag-architect sk...

MLPerf inference benchmarking patterns. Loadgen scenarios (SingleStream/MultiStream/Server/Offline), SUT/QSL interfaces, latency p99 targets, and performance metric interpretation. Sources: mlcommons/inference (Apache-2.0).

Design for mobile — touch targets, thumb zones, gesture patterns, iOS and Material 3 platform conventions, mobile form design, and responsive breakpoints. Use when building mobile-first UI, reviewing a layout for phone/tablet, asked about "touch targets", "thumb zone", "iOS conventions", "Android Material", "mobile form", or "responsive design". Do NOT use for desktop-only UIs — use `ux-heuristics` for general usability.

Deploy AI workloads to serverless GPU/CPU with Modal — define @app.function with GPU/memory specs, build custom container images, run batch jobs and web endpoints, schedule cron tasks, and use distributed volumes for model caching.

'Monitors dark web forums, marketplaces, paste sites, and ransomware

'Monitors Modbus TCP traffic on SCADA and ICS networks to detect anomalous

Monorepo governance patterns from Nx, Turborepo, and Changesets. Task graph execution, affected-only CI, remote caching, module boundary enforcement, shared config packages, and coordinated versioning. Sources: nrwl/nx, vercel/turborepo, changesets/changesets, nicowillis/pkg-size, nicowillis/size-limit, nicowillis/wireit.

Structure and operate a monorepo — Turborepo or Nx workspace setup, shared packages, task pipeline caching, change detection, publishable vs internal packages, versioning strategy, and CI optimization. Use when asked about "monorepo", "Turborepo", "Nx", "workspace", "shared packages", "internal packages", "turbo.json", "task pipeline", "remote cache", "affected packages", "package dependency graph", "pnpm workspaces", "yarn workspaces", or "how to share code between apps". Do NOT use for: CI/...

**Source:** OpenMOSS/MOSS-TTS-Nano (Apache 2.0) **Model:** 0.1B params · 48kHz stereo · CPU-friendly · multilingual

Advanced motion patterns for React / Next.js — drag & drop, gestures, text animations, SVG path drawing, custom hooks, imperative sequences (useAnimate), loaders, and the full API decision tree. Requires motion-foundations.

Design motion and animation for UI — duration timing, easing curves, micro-interaction patterns (button, form, toast, modal), and reduced-motion accessibility. Use when asked to "add animation", "make it feel alive", "micro-interactions", "transition", or "something feels janky/mechanical". Do NOT use for 3D or canvas animation — this covers UI-layer motion only.

Motion tokens, spring presets, performance rules, device adaptation, accessibility enforcement, and SSR safety for React / Next.js using motion/react. Foundation layer — all other motion skills depend on this.

Production-ready animation patterns for React / Next.js — button, modal, toast, stagger, page transitions, exit animations, scroll, and layout — built on motion-foundations tokens and springs.

Physics-based UI animation system. Spring physics, easing curves, micro-interaction patterns, and timeline staggering for native-quality web animations. Sources: popmotion, framer/motion, anime.js, greensock, argyleink/transition.css, and 5 others.

Production-ready UI motion system for React/Next.js. Use when implementing animations, transitions, or motion patterns.

MessagePack binary encoding for compact inter-agent payloads. Pack/unpack, typed arrays, extension types, streaming decoder, and performance comparison with JSON. Sources: msgpack/msgpack-javascript (ISC).

754 cybersecurity skills cho AI agent — 26 domains, map sang 5 frameworks (MITRE ATT&CK v19.1, NIST CSF 2.0, ATLAS v5.4, D3FEND v1.3, AI RMF 1.0). Real practitioner workflows.

Multi-agent debate / Society of Mind — N agents × R rounds converge on consensus. Du et al. 2023 beats zero-shot CoT on MMLU/GSM8K/MATH. Independent contributions from agent count AND round count. Sycophancy cascade prevention, heterogeneous models, compute budget. Sources: rohitg00/ai-engineering-from-scratch (Apache-2.0).

Safe context passing at the boundary between agents in a multi-agent pipeline — compress outgoing context, validate freshness of incoming context, detect bad-handoff signals (context bleed, hallucinated consensus, Agent Tennis). Use when an agent is about to pass work to another agent, receive results from a sub-agent, or when a multi-agent flow starts showing loop or drift symptoms. Do NOT use for session-level human handoffs — use the `handoff` skill for that. Do NOT use for dispatch planni...

MySQL and MariaDB schema, query, indexing, transaction, replication, and connection-pool patterns for production backends.

Use when setting up workflow automation, connecting APIs, building AI pipelines, or automating repetitive tasks with 400+ integrations. Triggers on: 'n8n', 'workflow automation', 'tự động hóa', 'kết nối API', 'no-code automation', 'zapier alternative', 'trigger webhook', 'schedule job', 'ai pipeline automation', 'integrate services'.

High-performance N-dimensional array operations for vector math in agent memory systems. Cosine similarity, dot product, Euclidean distance, batch matrix ops, and typed array backends. Sources: hughsk/ndarray, scijs ecosystem.

Needle-in-a-Haystack context placement testing for LLM retrieval accuracy. Context window stress tests, information placement strategies, lost-in-the-middle avoidance, and optimal document ordering for agent prompts. Sources: gkamradt/LLMTest_NeedleInAHaystack (MIT).

Build production Next.js apps — App Router conventions, React Server Components, server actions, data fetching strategies, caching layers, route handlers, middleware, metadata API, and parallel/intercepting routes. Use when asked about "Next.js App Router", "React Server Components", "RSC", "server actions", "use server", "use client", "Next.js caching", "fetch cache", "revalidatePath", "route handler", "Next.js middleware", "streaming UI", "Suspense boundary", "parallel routes", "Next.js met...

Alternative mã nguồn mở cho Claude Design — filesystem of design skills, 150 brand-grade DESIGN.md systems, 261 plugins. Chạy trên 21+ agent CLI qua MCP server.

NLP text normalization for cleaner LLM context. Stemming, lemmatization, stop-word removal, entity extraction, tokenization, and sentence boundary detection — all without calling an LLM. Sources: naturalnode/natural.

NVIDIA Cosmos — open platform world models cho Physical AI: robotics, autonomous vehicles. Cosmos3-Nano (16B) + Super (64B). Text/video/action I/O. Diffusers/vLLM/NIM. OpenMDW-1.1.

Instrument services for observability — structured logging, distributed tracing, metrics naming conventions, and the three-pillar model. Use when asked about "add logging", "structured logs", "distributed tracing", "trace ID", "metrics", "Prometheus", "OpenTelemetry", "why can't I debug production issues", or "correlate logs across services". Do NOT use for: SLO/alerting design — use `slo-design`. Do NOT use for: log storage/infrastructure provisioning.

Ollama local LLM — pull models, generate, chat, embeddings, REST API, Python client, structured output

ONNX Runtime cross-platform model inference. Export PyTorch/HuggingFace models to ONNX, execution provider selection (CUDA/TensorRT/CPU), graph optimization, and Node.js inference pipeline. Sources: microsoft/onnxruntime (MIT).