Claude Code Skills · page 11

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.

Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention

Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.

Expert guidance for Fully Sharded Data Parallel training with PyTorch FSDP - parameter sharding, mixed precision, CPU offloading, FSDP2

High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from laptop to supercomputer with same code. Use when you want clean training loops with built-in best practices.

Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.

Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, transferring GPT-4 capabilities to open-source models, or reducing inference costs. Covers temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.

Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+ tokens), extending pre-trained models beyond original context limits, or implementing efficient positional encodings. Covers rotary embeddings, attention biases, interpolation methods, and extrapolation strategies for LLMs.

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

Reduce LLM size and accelerate inference using pruning techniques like Wanda and SparseGPT. Use when compressing models without retraining, achieving 50% sparsity with minimal accuracy loss, or enabling faster inference on hardware accelerators. Covers unstructured pruning, structured pruning, N:M sparsity, magnitude pruning, and one-shot methods.

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.

Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6× speedup), reducing latency for real-time applications, or deploying models with limited compute. Covers draft models, tree-based attention, Jacobi iteration, parallel token generation, and production deployment strategies.

Evaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benchmarks with pass@k metrics. Use when benchmarking code models, comparing coding abilities, testing multi-language support, or measuring code generation quality. Industry standard from BigCode Project used by HuggingFace leaderboards.

Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.

Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.

Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support

Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support

Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

Guides the usage of the Gemini API on Agent Platform with the Google Gen AI SDK for enterprise AI applications. Covers SDK usage (Python, JS/TS, Go, Java, C#), capabilities like Live API, tools, multimedia generation, caching, and batch prediction.

Use when the user asks to run Gemini CLI for code review, plan review, or big context (>200k) processing. Ideal for comprehensive analysis requiring large context windows. Uses Gemini 3 Pro by default for state-of-the-art reasoning and coding.

Creates detailed, sectionized implementation plans through research, stakeholder interviews, and multi-LLM review. Use when planning features that need thorough pre-implementation analysis.

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.

Validate a Composio CLI beta release and promote it to a stable release by dispatching the CLI binary workflow with an existing beta tag.

Write end-to-end tests for CLI commands using the Docker-based test framework in ts/e2e-tests/cli/.

CLI design guidelines — arguments, flags, subcommands, help, output, errors, interactivity, config precedence. Apply when designing new commands or reviewing CLI UX.

Implement new CLI commands in ts/packages/cli/ using Effect.ts patterns, service wiring, and @effect/cli declarations.

Release the CLI — diff features since last release, run /full-cli-test on them, and create a changeset PR with patch bump and test results.

Trigger a CI binary build via workflow dispatch, monitor it, download the artifact, and test the CLI binary locally.

Add hosted API key support to a tool so Sim provides the key when users don't bring their own. Use when adding hosted keys, BYOK support, hideWhenHosted, or hosted key pricing to a tool or block.

Review PRs and diffs for unbounded memory loading, concurrency explosions, oversized payload materialization, and missing pagination or byte caps. Use when reviewing cleanup jobs, background jobs, data imports/exports, file parsing, API fan-out, workflow execution payloads, large arrays/files, or any change that reads many rows, files, responses, logs, or external API pages into process memory.

Install, upgrade, and operate the Sim Helm chart on Kubernetes. Covers install path selection (inline / existingSecret / External Secrets Operator), required secret generation, the values.yaml mental model (env vs envDefaults vs Secret), and common failure triage. Invoke when a user asks about deploying Sim to a cluster, authoring a Sim values.yaml, debugging a Sim pod that won't start, upgrading a Sim release, or wiring Sim into a secret manager.

Author or review a Drizzle DB migration for zero-downtime safety — expand/contract phasing, backward-compatibility with the deployed app version, and writing the `-- migration-safe` acknowledgment the check:migrations lint requires. Use when adding/editing files under `packages/db/migrations/` or changing `packages/db/schema.ts`.

Expert prompt engineering skill that transforms Claude into "Alpha-Prompt" - a master prompt engineer who collaboratively crafts high-quality prompts through flexible dialogue. Activates when user asks to "optimize prompt", "improve system instruction", "enhance AI instruction", or mentions prompt engineering tasks.

当用户需要弃用一个工作流节点（保留向后兼容、隐藏出模板面板）时触发该 skill。FastGPT 工作流节点的弃用流程标准化封装，覆盖模板、Dispatcher、UI 引用等所有需要改动的位置。

将 FastGPT 文档从中文翻译为面向北美用户的英文。当用户提到翻译文档、i18n、国际化、translate docs、新增/修改了中文文档需要同步英文版时，使用此 skill。也适用于用户要求检查文档翻译缺失、批量翻译、或对比中英文文档差异的场景。

为 FastGPT 新资源接入权限管理。当用户需要为新资源（如 AgentSkill、Plugin 等）添加权限支持时触发。

FastGPT API 开发规范。重点强调使用 zod schema 定义入参和出参,在 API 文档中声明路由信息,编写对应的 OpenAPI 文档,以及在 API 路由中使用 schema.parse 进行验证。

仅当用户明确手动指定使用 pr-review skill 时触发；不要因为用户传入 PR 链接、要求 review 或要求代码审查而自动触发。

当用户需要编写一个单元测试时，触发该 skill，编写单元测试。

手动触发的 FastGPT PR 或本地分支变更梳理技能。仅当用户显式调用 $pr-change-analysis 时使用；用于 reviewer 分析一个 GitHub PR 或当前本地分支相对 upstream/main 的需求变更、影响范围、代码质量与代码风格，不用于自动审查触发。

Professional finance research toolkit — backtesting (7 engines + benchmark comparison panel), factor analysis, Alpha Zoo (452 pre-built alphas across qlib158/alpha101/gtja191/academic), options pricing, 79 finance skills, 29 multi-agent swarm teams, Trade Journal analyzer, and Shadow Account (extract → backtest → render) across 18 market-data sources (tushare, yfinance, okx, akshare, baostock, tencent, mootdx, ccxt, futu, local, eastmoney, sina, stooq, yahoo, plus optional-key finnhub/alphavantage/tiingo/fmp).

ADR/H-share/A-share cross-listing premium analysis — track pricing gaps between US-listed ADRs, HK-listed H-shares, and A-shares for arbitrage signals, dual-listing valuation, and delisting risk assessment.

AKShare financial data aggregator (18k+ stars). Free, no API key. Covers A-shares, US, HK, futures, macro, forex. Primary fallback for tushare and yfinance.

Browse and bench the bundled alpha zoos — prebuilt cross-sectional factor libraries (Kakushadze 101, GTJA 191, Qlib 158, Fama-French / Carhart). Use when the user asks "which alphas exist", wants metadata on a named alpha, or wants to run IC/IR on a whole zoo over a universe.

A 股 ST/*ST 风险预测框架 — 基于最新中报/三季报或业绩预告/快报，预测下一财年是否会因营收、利润、净资产、分红不达标而被风险警示，并将新浪监管处罚记录作为独立证据面纳入风险等级。仅适用于 A 股，不预测财务造假。

Asset allocation theory and optimizer usage — MPT / Black-Litterman / risk budgeting / all-weather strategy, including guides for 4 optimizers and rebalancing rules.

Diagnose failed or underperforming backtests, locate the root cause, and fix the issue

Behavioral finance applications: theories of overreaction and underreaction, behavioral explanations for momentum and reversal, investor sentiment cycles, cognitive-bias checklists, and debiasing quantitative strategies.

Candlestick pattern recognition engine, pure pandas vectorized implementation of 15 classic candlestick patterns (5 single-candle + 5 double-candle + 4 triple-candle + 1 trend confirmation), generating a composite signal from bullish/bearish pattern scores.

CCXT unified crypto exchange library (100+ exchanges). Free public market data. Fallback when OKX is unavailable.

基于缠论（缠中说禅）的形态识别引擎，使用czsc库自动检测K线分型、笔、中枢，并生成一买/一卖/二买/二卖/三买/三卖等买卖点信号。支持多周期分析和形态分类（3/5/7/9/11笔形态）。

Commodity analysis (oil supply-demand balance / gold pricing / copper as an economic predictor / inventory cycles / futures premium-discount structure / seasonality), generating directional commodity signals.

A股可转债分析——转股/纯债/期权三维估值、下修/强赎/回售博弈、双低策略与转债轮动选债框架

公司事件驱动分析：并购套利价差计算、大股东增减持信号、股权激励解读、定增配股影响评估、A股ST/退市预警

Correlation and cointegration analysis — co-movement discovery, deep return-correlation analysis, sector clustering, realized correlation, Engle-Granger / Johansen cointegration, half-life, Kalman dynamic hedge ratio, cross-market linkage analysis, and pair-trading signal generation

固收与信用分析：信用债评级、利差分析、违约风险评估、城投债研究、可转债定价与策略。

Write signal_engine.py for portfolios spanning multiple markets (A-shares + crypto, equity + forex, etc.)

Crypto-derivatives strategies — perpetual funding-rate arbitrage, futures term-structure contango/backwardation trading, and option volatility-smile / Greeks analysis.

The single ROUTER for every data need. Load this skill BEFORE any backtest, data-fetch, or research task to pick the best available source/tool, honour auth (env) requirements, and avoid ban-risk providers.

DeFi yield analysis and optimization — lending rates, LP yields, staking returns, yield farming strategies, risk-adjusted yield comparison, and protocol-level sustainability assessment.

Dividend stock analysis for income, dividend-growth, and shareholder-return strategies, including yield quality, payout sustainability, ex-dividend mechanics, and yield-trap checks.

Read any common document/data file — PDF, Word (.docx), Excel (.xlsx/.xls), PowerPoint (.pptx), images (OCR), CSV/TSV, plain text, JSON/YAML/TOML, HTML/XML, and most source-code files. Use the `read_document` tool.

盈利预测与一致预期分析（自上而下/自下而上预测法/SUE/PEAD/分析师预期修正），捕捉业绩超预期交易机会。

Earnings estimate revisions, guidance analysis, and post-earnings drift (PEAD) — track analyst consensus changes, earnings surprise patterns, and management guidance shifts for US/HK equities.

SEC EDGAR filing analysis — 10-K, 10-Q, 8-K, proxy statements, insider Form 4. Extract key financials, risk factors, management discussion, and generate investment signals from US public company filings.

Elliott Wave Theory signal engine. Detects swing points through Zigzag, matches 5-wave impulse and 3-wave corrective structures, validates them with Fibonacci wave relationships, and generates trend-top / correction-complete signals. Pure in-house pandas implementation.

ETF分析：产品筛选、费率对比、跟踪误差、流动性评估、策略应用与中国市场ETF量化配置框架。

Event-driven strategy based on sentiment-scored signals from news, announcements, and macro events. The LLM acts as the NLP engine, and event data follows a CSV schema.

Trade execution modeling (backtest only) — slippage formulas (linear / square-root impact), VWAP/TWAP execution logic, market-impact cost estimation, and execution-assumption configuration.

Factor research framework with IC/IR analysis, quantile backtesting, and factor combination. Suitable for cross-sectional factor evaluation across multiple instruments.

财报三表深度解读——三表勾稽关系、盈利质量(应计vs现金流)分析、杜邦分解、10+财务造假红旗指标

基金分析与筛选：晨星评级/夏普比率/信息比率、Sharpe风格箱分析、风格漂移检测、基金经理评价、FOF组合构建、ETF选择

Fundamental factor screening — filter stocks by PE/PB/ROE, financial statement fields, and other metrics for value or growth selection. Supports A-shares (via tushare extra_fields or fundamental_fields) and HK/US stocks (via yfinance Ticker info).

Geopolitical risk analysis: quantify crisis signals, identify precursors, and build event-driven strategies for war, sanctions, and supply disruption scenarios.

Global macro analysis framework (central bank policy transmission / FX forecasting / geopolitical risk / capital flows), used to build macro factor signals that drive cross-asset allocation.

Harmonic Patterns signal engine. Identifies XABCD five-point structures such as Gartley/Bat/Butterfly/Crab based on Fibonacci geometry, and generates trading signals in the PRZ (Potential Reversal Zone).

Hedging strategy design (beta hedge / option protection / tail risk / cross-asset hedging), including hedge-ratio calculation and cost evaluation.

Stock Connect (Shanghai/Shenzhen-Hong Kong) fund flow analysis — Northbound (foreign into A-shares), Southbound (mainland into HK), sector allocation tracking, and cross-border arbitrage signals.

Ichimoku Kinko Hyo five-line system signal engine. A standalone Japanese technical-analysis school that generates trading signals from Tenkan/Kijun crossovers, cloud position, and Chikou confirmation. Pure pandas implementation.

Liquidation level analysis and heatmap interpretation — identify leveraged position concentration, liquidation cascades, stop-hunt zones, and use liquidation data as support/resistance signals.

Macroeconomic cycle positioning and central-bank policy interpretation, including GDP/CPI/PMI/rates/FX analysis, with output in the form of major-asset allocation tilts.

Market microstructure: bid-ask spread analysis, order-flow toxicity metrics (VPIN / Kyle lambda), liquidity measures (Amihud / Roll), price-impact models, limit-order-book analysis, and China A-share call auction / block trade mechanics.