Claude Skills by yanacuti1121

LLM observability with Langfuse — tracing, evals, prompt management, cost tracking

Use when building stateful multi-step agents, agent graphs, or workflows with LLMs. Triggers on: 'langgraph', 'state graph', 'stateful agent', 'agent workflow', 'agent loop', 'multi-step agent', 'persistent agent', 'human-in-the-loop agent', 'agent with memory', 'graph-based agent'.

Use for latency-sensitive systems such as realtime dashboards, market data, streaming agents, execution gateways, queues, caches, or HFT-like infrastructure where freshness and p95 latency matter.

Scan codebase and git history for leaked secrets — API keys, tokens, passwords, private keys. Uses whispers-style regex patterns + gitleaks. Run as /leak-check before any git push. Alerts on Claude API keys, GitHub tokens, AWS credentials, and generic high-entropy strings.

Anti-slop frontend framework cho AI agents — layout, typography, motion, spacing chuẩn thay vì boilerplate UI nhàm. Dùng khi muốn AI build UI có aesthetic thực sự.

Linkerd2 lightweight service mesh for automatic mTLS, traffic metrics, and certificate rotation. Zero-config mTLS, per-route metrics, traffic splits, retries, and timeouts for inter-agent communication. Sources: linkerd/linkerd2 (Apache-2.0).

lint-staged pre-commit hook for running linters only on Git-staged files. Pattern-matched task runners, auto-fix before commit, integration with husky, and staged-only analysis to minimize CI overhead. Sources: okonet/lint-staged (MIT).

iOS 26 Liquid Glass design system — dynamic glass material with blur, reflection, and interactive morphing for SwiftUI, UIKit, and WidgetKit.

Call 100+ LLMs through a single OpenAI-compatible interface with LiteLLM — use completion/acompletion/embedding with any provider (Anthropic, OpenAI, Google, Groq, Ollama, etc.), run a proxy server for team rate-limiting and cost tracking, load-balance across providers.

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).

Use when fine-tuning, training, or doing LoRA/QLoRA on LLMs. Triggers on: 'fine-tune LLM', 'LoRA', 'QLoRA', 'llamafactory', 'llama factory', 'train model', 'SFT', 'RLHF', 'DPO', 'fine tuning', 'custom model', 'instruction tuning', 'finetuning'.

LLM evaluation harness for accuracy benchmarking. MMLU/HumanEval/MATH eval runners, model-graded scoring, prompt regression testing, and per-skill accuracy tracking. Sources: openai/simple-evals (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, anthropics/anthropic-sdk-python, outlines-dev/outlines, guardrails-ai/guardrails.

Design UI for LLM-powered features — streaming response display, hallucination states, RAG source attribution, generative UI patterns, and AI trust calibration. Use when asked to "show streaming output", "display AI response", "cite sources", "handle AI errors", "show confidence", "build a chatbot UI", or "design for generative AI". Do NOT use for: backend LLM API integration or RAG pipeline design — those are separate concerns.

Design and run load tests — tool selection (k6/Locust), test type strategy (smoke/load/stress/spike/soak), ramp-up curves, p50/p95/p99 targets, and threshold policy. Use when asked to "write a load test", "k6 script", "performance test", "check how many concurrent users", "find breaking point", "set SLO thresholds", or "load test this API". Do NOT use for: frontend Core Web Vitals (see web-performance skill). Do NOT use for: unit/integration test suites. Do NOT use for: security fuzzing.

Production logging — structlog, structured JSON logs, correlation IDs, log levels, OpenTelemetry traces, no print()

Grafana Loki label-based log aggregation. LogQL queries, Promtail scraping, structured log indexing, stream selectors, and log-to-metric extraction for agent audit trail analysis. Sources: grafana/loki (AGPL-3.0).

Use when searching for where a function/class/variable is defined or referenced, before using grep on the codebase. Triggers on: 'where is X defined', 'find all uses of Y', 'what calls Z', 'go to definition', 'find references', searching for a symbol name across files.

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).

'This skill covers implementing Okta as a centralized identity provider

'Manages the end-to-end cyber threat intelligence lifecycle from planning

'Maps observed adversary behaviors, security alerts, and detection rules

End-to-end marketing campaign — audience research, positioning, landing page copy, email sequences, social posts, ad copy, video scripts, content calendars. Use with marketing-agent.

Convert mọi file (PDF, Word, Excel, PowerPoint, Image, Audio, HTML, CSV, YouTube URL) thành Markdown cho LLM. Tích hợp vào YAMTAM vault để import tài liệu.

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.

Self-hosted open-source collaboration platform — chat, bots, webhooks, plugins, AI integration, Slack/Teams alternative with full data control. Triggers on: 'mattermost', 'self-hosted slack alternative', 'open source teams alternative', 'mattermost bot', 'mattermost webhook', 'mattermost plugin', 'mattermost API', 'mattermost integration', 'AI agent mattermost', 'deploy mattermost', 'mattermost docker', 'mattermost kubernetes', 'enterprise chat self-hosted', 'mattermost AI features', 'matterm...

Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes — Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/PRD asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".

Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.

Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.

Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.

Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, branch -D, etc.) before they execute. Use when user wants to prevent destructive git operations, add git safety hooks, or block git push/reset in Claude Code.

A relentless interview to sharpen a plan or design.

A relentless interview to sharpen a plan or design, which also creates docs (ADR's and glossary) as we go.

Grill the user relentlessly about a plan or design. Use when the user wants to stress-test a plan before building, or uses any 'grill' trigger phrases.

Compact the current conversation into a handoff document for another agent to pick up.

Implement a piece of work based on a spec or set of tickets.

Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.

Migrate test files from `as` type assertions to @total-typescript/shoehorn. Use when user mentions shoehorn, wants to replace `as` in tests, or needs partial test data.

Build a throwaway prototype to answer a design question. Use when the user wants to sanity-check whether a state model or logic feels right, or explore what a UI should look like.

Investigate a question against high-trust primary sources and capture the findings as a Markdown file in the repo. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a background agent.

Use when you need to resolve an in-progress git merge/rebase conflict.

Create exercise directory structures with sections, problems, solutions, and explainers that pass linting. Use when user wants to scaffold exercises, create exercise stubs, or set up a new course section.

Set up Husky pre-commit hooks with lint-staged (Prettier), type checking, and tests in the current repo. Use when user wants to add pre-commit hooks, set up Husky, configure lint-staged, or add commit-time formatting/typechecking/testing.

Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.

Teach the user a new skill or concept, within this workspace.

Turn the current conversation into a spec and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.

Break a plan, spec, or the current conversation into a set of tracer-bullet tickets, each declaring its blocking edges, published to the configured tracker — edges as text in a local file, or native blocking links on a real tracker.

Move issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.