Menu
Skills in this repository
SkillsMP has collected 1,566 skills from yanacuti1121/Yana-AI. Open a skill to review its source and details.
yanacuti1121/Yana-AIShowing 40 of 1,566 collected skills.
llama.cpp GGUF/GGML quantization patterns for local LLM inference. Quantization levels (Q4_K_M, Q8_0, F16), CPU/GPU offloading, context window sizing, and embedding extraction via llama.cpp HTTP server. Sources: ggerganov/llama.cpp (MIT).
LLM.int8() and bitsandbytes 4-bit quantization (NF4/FP4) for memory-efficient LLM loading. Mixed-precision decomposition, QLoRA fine-tuning, double quantization, and GPU OOM prevention patterns. Sources: TimDettmers/bitsandbytes (MIT).
LLM inference load testing for throughput and concurrency limits. Token/s benchmarks, concurrent request sweeps, latency-vs-throughput curves, and breaking-point identification. Sources: vllm-project/vllm benchmarks (Apache-2.0).
LLM output validation patterns — structured output schemas, hallucination detection, retry-with-backoff, streaming output safety, tool-call validation, and confidence scoring. Sources: instructor-ai/instructor, colinhacks/zod, vercel/ai,…
Source string mutation with automatic source map tracking. In-place overwrite, insertion, removal, and prepend/append operations that preserve accurate sourcemap positions for patch loops and codemods. Sources: unjs/magic-string (MIT).
Linear algebra matrix operations and dimensionality reduction for embedding compression. PCA, SVD basics, matrix multiply, transpose, and vector projection — pure JS for agent memory optimization. Sources: scijs/matrix, numeric.js patterns.
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:…
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.
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.
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.
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,…
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.
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…
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:…
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.
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).
Generate many candidate options across different styles (concise, simple, creative, professional), then run pairwise elimination rounds to select the top 3 with reasons. Use when asked for 'phương án', 'cho mấy phương án', 'generate options and pick best',…
Source text: Vietnamese
Prioritize a task list by direct pairwise comparison instead of isolated scoring — compare every pair on importance, urgency, and impact, then produce a ranked execution order. Use when asked 'so từng cặp', 'sắp xếp ưu tiên', 'prioritize these tasks', 'việc…
Source text: Vietnamese
Nanosecond-speed process existence check by PID. sindresorhus/process-exists patterns, cross-platform PID validation, zombie detection, and watchdog loop for monitoring long-running agent sandboxes. Sources: sindresorhus/process-exists.
System process list monitoring and PID management for agent sandboxes. Cross-platform process enumeration, CPU/memory per-process metrics, filtering by name/PID, and detecting unexpected child spawns. Sources: sindresorhus/ps-list.
Code profiling and benchmarking laws from 5 repos. Statistical benchmark harnesses, micro-second delta timing, string-processing performance comparison, IOPS throttle monitoring, and bundle size analysis. Sources: bestiejs/benchmark.js,…
Facebook Pyre static type analysis patterns for Python. Incremental type checking, taint tracking for data-flow security analysis, pysa rules, and integrating Pyre into CI pipelines. Sources: facebook/pyre-check (MIT).
Raft distributed consensus algorithm patterns. Leader election, log replication, safety invariants, snapshot/compaction, and membership changes for replicated agent state machines. Sources: hashicorp/raft (MPL-2.0).
Node.js-native Raft consensus implementation for embedding distributed consensus directly into agent processes. In-process leader election, log replication via TCP, and state machine callbacks without external services. Sources: skiff-project/skiff (MIT).
RAM-backed virtual disk patterns for zero-latency sandbox scratch space. Linux tmpfs mounts, Node.js ramdisk creation, size-capped temp volumes, auto-cleanup on process exit, and race-condition-free temp directory isolation. Sources: mafintosh/ramdisk, Linux…
Raw IP packet analysis for agent network diagnostics. ICMP ping, TCP port probing, packet crafting, and network path tracing using raw sockets — for Sandbox network isolation validation. Sources: indutny/raw-socket.
ReAct (Reason+Act) agent loop — the canonical Observe/Think/Act cycle. Tool registry, stop conditions, turn budget, observation formatting. Every 2026 agent framework runs this loop under the hood. Sources: rohitg00/ai-engineering-from-scratch (Apache-2.0).
Prevent ReDoS (Regex Denial of Service) by escaping user-supplied regex special characters. escape-string-regexp patterns, catastrophic backtracking detection, safe dynamic regex construction, and timeout guards. Sources: sindresorhus/escape-string-regexp.
REPL-driven dynamic code execution patterns for compiled languages. Incremental eval, state persistence between evaluations, error recovery, and sandboxed expression execution. Sources: evcxr/evcxr (Apache-2.0).
ReWOO decoupled planning — Planner/Worker/Solver split. 5x fewer tokens than ReAct on HotpotQA, +4% accuracy. Plan-and-Execute generalization, planner distillation to 7B. When to use plan-first vs interleaved. Sources: rohitg00/ai-engineering-from-scratch…
Runtime operational metrics with meters, timers, histograms, and moving averages. node-measured patterns for tracking request rates, execution durations, error rates, and EWMAs for agent performance observability. Sources: caustik/node-measured.
Low-level OCI container runtime patterns using Linux namespaces and cgroups. Namespace isolation (pid/net/mnt/uts/ipc), cgroup resource caps, rootless execution, OCI bundle structure, and runc lifecycle (create/start/kill/delete). Sources: opencontainers/runc.
Strict HTML cleanup with self-closing tag repair, attribute allowlists, URL protocol validation, and CSS sanitization. apostrophecms/sanitize-html patterns for cleaning LLM-generated and web-scraped HTML. Sources: apostrophecms/sanitize-html.
Self-Refine iterative improvement (generate/feedback/refine loop, +20 avg across 7 tasks) and CRITIC external verification (tool-grounded critique for factual tasks). Anthropic evaluator-optimizer pattern. Sources: rohitg00/ai-engineering-from-scratch…
Mine session history for recurring mistakes and distill them into short CLAUDE.md rules. Use when asked to 'rút rule', 'tự rút rule', 'extract rules from sessions', 'lỗi lặp đi lặp lại', 'review past sessions for mistakes', 'optimize CLAUDE.md from history',…
Source text: Vietnamese
sGLang RadixAttention — KV cache stored in a radix tree, reused across requests sharing common prefixes. Cache-aware scheduling (depth-first, LRU at branch level). 29% throughput edge vs vLLM on ShareGPT, 6.4x on RAG workloads, 86% hit rate on voice. Deployed…
Smart layout patterns and aesthetic engineering from 20 production repos. Shadcn command patterns, floating UI positioning, virtual lists, drag-and-drop, resizable panels, CSS container queries, masonry, aesthetic micro-details (shadows, borders, glass), and…