Code quality guidelines and best practices for writing clean, maintainable code. Covers naming, structure, error handling, testing, and code review standards. Use when writing code, reviewing, refactoring, or asking "how should I name this", "best practice for", "clean code".

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.

LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.

Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior.

Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.

跨会话检索聊天原文（记忆不足时再用）。在回答任何关于历史事件、日期、人物、过去对话的问题时，如果记忆中没有相关信息或不足以回答，则需要使用跨会话检索聊天原文。用 mcp_exec_command 执行 scripts/search_history.py，读 ~/.jiuwenclaw/agent/sessions/*/history.json。支持 channel、session_id、关键词、时间窗。如果搜索结果不足，尝试用不同的关键词再次搜索。

在用户提供 Chrome/Perfetto trace.json、或排查 Ascend 上 MoE/FusedDeepMoe 等算子性能时使用。按 phase、category、core group、tid 统计耗时、overlap、bubble，输出 CSV、Markdown 报告与确定性诊断；可选外部 LLM 扩写分析。默认 phase 映射面向 UMDK FusedDeepMoe，其它 trace 需替换或扩展 config/phase_map.yaml。

Diffusion-based molecular docking. Predict protein-ligand binding poses from PDB/SMILES, confidence scores, virtual screening, for structure-based drug design. Not for affinity prediction.

Comprehensive patient stratification for precision medicine by integrating genomic, clinical, and therapeutic data. Given a disease/condition, genomic data (germline variants, somatic mutations, expression), and optional clinical parameters, performs multi-phase analysis across 9 phases covering disease disambiguation, genetic risk assessment, disease-specific molecular stratification, pharmacogenomic profiling, comorbidity/DDI risk, pathway analysis, clinical evidence and guideline mapping, clinical trial matching, and integrated outcome prediction. Generates a quantitative Precision Medicine Risk Score (0-100) with risk tier assignment (Low/Intermediate/High/Very High), treatment algorithm (1st/2nd/3rd line), pharmacogenomic guidance, clinical trial matches, and monitoring plan. Use when clinicians ask about patient risk stratification, treatment selection, prognosis prediction, or personalized therapeutic strategy across cancer, metabolic, cardiovascular, neurological, or rare diseases.

Search ClinicalTrials.gov with natural language queries. Find clinical trials, enrollment, and outcomes using Valyu semantic search.

NGS analysis toolkit. BAM to bigWig conversion, QC (correlation, PCA, fingerprints), heatmaps/profiles (TSS, peaks), for ChIP-seq, RNA-seq, ATAC-seq visualization.

Look up Twitter/X profiles - get bio, followers, tweets, and engagement

Find Key Opinion Leaders (KOLs) in a given domain by combining web research with LinkedIn post search. Given a company/idea and target domain, generates authority keywords, searches LinkedIn posts to find prolific authors with high engagement, and merges with web-researched influencers. Use when someone wants to "find influencers in X space" or "who are the KOLs for Y industry."

Track product champions for job changes and qualify their new companies against ICP. Takes a CSV of known champions (with LinkedIn URLs), creates a baseline snapshot via Apify enrichment, then detects when champions move to new companies. Scores new companies on a 0-4 ICP fit scale. Outputs a downloadable CSV of movers with qualification verdicts.

Subscribe to and scan industry newsletters for buying signals, competitor mentions, ICP pain-point language, and market shifts. Parses incoming newsletter emails via AgentMail, matches against keyword campaigns, and delivers a weekly digest of actionable signals. Use when a marketing team wants to turn newsletter subscriptions into an ongoing intelligence feed without manual reading.

Extract visual branding (colors, typography, layout patterns) from a client's website and generate a style preset compatible with the HTML slides skill and a brand config JSON for the content asset creator. Uses WebFetch to read pages and analyzes CSS/HTML to identify the color palette, font pairings, and aesthetic patterns.

Generate a personal voice guide for X (Twitter) and/or LinkedIn by scanning a user's past posts and iteratively refining with sample-and-feedback loops. Produces a structured markdown voice guide that sibling skills (create-x-content, create-linkedin-content) consume to draft in-voice posts. Different from brand-voice-extractor, which analyses company blogs/landing pages — this skill is for personal social voice.

Monitor Twitter/X, Reddit, LinkedIn, and Hacker News for trending narratives, viral posts, and hot-button topics in your space. Maps trends to ad hook opportunities with timing urgency scores. Tells you what to run ads about right now while the topic is hot.

Diagnose Meta Ads campaign performance and account gaps using Meta's actual system mechanics — including customer-journey coverage, Breakdown Effect, Learning Phase, Auction Overlap, Pacing, and Creative Fatigue. Use for performance diagnosis, account audits, full-funnel or TOF/MOF/BOF gap analysis, deciding what to test or create next, and producing novice-friendly recommendations without forcing every campaign or ad into a funnel stage.