my_claude_skills

LLM 微调专家。使用场景：微调 LLMs、训练自定义模型、适配基础模型进行特定任务。用于配置 LoRA/QLoRA 适配器、准备 JSONL 训练数据集、设置超参数、适配器训练、迁移学习、Hugging Face PEFT 微调、指令微调、RLHF、DPO 或量化部署微调模型。触发词：LoRA、QLoRA、PEFT、微调、适配器调优、LLM 训练、模型训练、自定义模型。

深度研究工具。仅在用户明确说'深度研究'、'深入调查'、'全面报告'时使用。比普通搜索慢 10-100 倍。对于普通研究/查找请求，使用 parallel-web-search。支持多轮：传递 --previous-interaction-id 以继续之前的上下文。

AI 模型提示工程指南：LLM、图像生成器、视频模型。技术：思维链、少样本、系统提示、负提示。模型：Claude、GPT-4、Gemini、FLUX、Veo、Stable Diffusion 提示。用于：更好的 AI 输出、一致的结果、复杂任务、优化。触发词：提示工程、如何提示、更好的提示、提示技巧、提示指南、LLM 提示、图像提示、AI 提示、提示优化、提示模板、提示结构、有效提示、提示技术。

RAG（检索增强生成）模式专家。用于构建 RAG 管道、嵌入文档、实现混合搜索、上下文检索、HyDE、Agentic RAG、多模态 RAG、查询分解、重排或 pgvector 搜索。触发词：RAG、检索、LLM、上下文、嵌入、HyDE、重排、pgvector、多模态。

Apache Spark 作业优化：分区策略、缓存、shuffle 优化和内存调优。用于提升 Spark 性能、调试慢作业或扩展数据处理管道。触发词：Spark、性能优化、分区、缓存、shuffle、数据倾斜。

Web 研究工具：搜索多个来源、综合发现并生成带引用的研究报告。用于用户要求在线研究主题、搜索网页、查找信息、获取最新信息、对比选项或生成研究报告。触发词：研究、网上搜索、查找信息、最新信息、对比选项、研究报告。

帮助用户发现和安装 agent skills。当用户询问"如何做 X"、"找一个做 X 的 skill"、"有没有能...的 skill"，或表达想扩展能力时触发此 skill。

Claude Skills 综合质量分析工具。从 GitHub、市场、ZIP 文件或本地目录获取 skills，从五个维度（结构、安全、UX、代码质量、集成）评估并给出评分。触发词："分析 skill 质量"、"审查这个 skill"、"评估 skill"、"skill 质量检查"。

Claude Code skills 综合安全分析工具。检测恶意代码、混淆、YAML 注入、typosquatting、时间炸弹、沙箱逃逸和 40+ 攻击模式。用于安装前分析 skills 或审计现有 skills。触发词："分析 skill 安全"、"检查 skill 是否安全"、"审计已安装 skills"、"扫描文件安装前"、"审查 skill 漏洞"。

通过 inference.sh CLI 进行 Web 搜索和内容提取，支持 Tavily 和 Exa。功能：AI 搜索、内容提取、直接回答、研究。用于：研究、RAG 管道、事实核查、内容聚合、agents。触发词：web search、tavily、exa、search api、content extraction、research、internet search、ai search、search assistant、web scraping、rag。

Feishu (Lark) API operations for documents, cloud storage, permissions, wiki, bitables, messaging, and group chat. Use when: (1) reading/writing Feishu documents, (2) managing cloud drive files, (3) managing document permissions, (4) navigating wiki/knowledge base, (5) working with bitable (multidimensional tables), (6) sending messages or managing groups, (7) searching documents/wiki. Triggers on 'feishu', 'lark', '飞书', 'cloud doc', 'wiki', 'bitable', 'im message', 'send message', 'group chat'. NOT for: calendar, audio/video calls, or when the task doesn't involve Feishu.

OMC agent 目录、可用工具、团队管道路由、提交协议和 skills 注册表。当委托 agent、使用 OMC 工具、协调团队、提交代码或调用 skills 时自动加载。

验证和适配 Claude Code skills 使其兼容 OpenCode。用于：（1）检查 skill 是否在 Claude Code 和 OpenCode 中都能工作；（2）在 OpenCode 中测试 skill；（3）检测平台特定问题。触发词：'opencode'、'opencode 兼容'、'skill 兼容'、'跨平台 skill'。NOT for：创建新 skill（用 skill-creator）、安装 skill（用 install-skill）。

通用的自我改进 agent，从所有 skill 经验中学习。使用多内存架构（语义 + 情景 + 工作）持续演进代码库。通过 hooks 在 skill 完成/错误时自动触发基于自检的自我修正。

Install a skill into Claude Code. Use when the user wants to add a new skill, install a skill from a local folder, or set up a custom skill. Triggers on "install skill", "add skill", "set up skill", "load skill".

Patent application drafting based on codebase analysis. Use when: (1) writing patent applications from code, (2) drafting invention disclosures, (3) preparing patent claims. Triggers on 'patent', '专利', 'write patent', 'patent application', 'invention disclosure', '撰写专利'. Requires: user must provide a DOCX template file. NOT for: legal advice, patent search (freedom-to-operate), or when no codebase is involved.

Turn Claude Code into a persistent agent with memory, identity, and self-learning. Use when: (1) starting a new project that needs a persistent AI assistant, (2) setting up memory/identity files for Claude Code, (3) user wants Claude Code to remember context across sessions. NOT for: one-off tasks, installing skills, or when session persistence is not needed.

Create, edit, improve, or audit Claude Code skills. Use when creating a new skill from scratch, improving an existing skill, reviewing a SKILL.md file, or restructuring a skill directory. Triggers on "create a skill", "author a skill", "tidy up a skill", "improve this skill", "review the skill", "clean up the skill", "audit the skill". NOT for: installing skills (use install-skill), using skills, or general file editing.

Translate OpenClaw skills into Claude Code skills. Use when: (1) user wants to adapt an OpenClaw skill for Claude Code, (2) migrating skills between platforms, (3) converting SKILL.md with OpenClaw metadata to Claude Code format. Triggers on 'translate skill', 'convert skill', 'adapt skill', 'openclaw skill to claude code'. NOT for: creating new skills from scratch (use skill-creator), installing skills (use install-skill).

Essential SSH commands for secure remote access, key management, tunneling, and file transfers. Use when the user needs help with SSH connections, key generation, port forwarding, file transfers (scp/sftp/rsync), or SSH configuration. NOT for: general networking questions, VPN setup, or when SSH is not involved.

Web search using Tavily's LLM-optimized API. Use when the user asks to search the web, look up information online, find recent news, or research a topic. NOT for: searching local files, code lookup, or when the answer is purely based on existing knowledge.

Protect against context compaction hallucination. Use when: (1) performing multi-step tasks that span many turns, (2) context compaction may have occurred and work state is unclear, (3) user mentions compact/compression/context loss. Triggers on 'compact', '上下文压缩', '你忘了', '你之前在做什么', '确认进度', 'checkpoint'. Also activate proactively for any task expected to take more than 10 turns.

Structured prompt tuning workflow with logic-first discipline. Use when: (1) modifying system prompts, (2) fixing prompt behavior based on test cases, (3) iterating on prompt templates, (4) debugging prompt outputs. Triggers on 'tune prompt', 'fix prompt', 'prompt 调优', 'prompt 不对', '改 prompt', '调整提示词'. NOT for: writing prompts from scratch (use general coding), non-prompt code changes, or when no prompt file is involved.

Headless browser automation CLI optimized for AI agents with accessibility tree snapshots and ref-based element selection. Use when automating multi-step web workflows, scraping structured data, or testing SPAs. NOT for: simple one-off page fetches (use curl/fetch), screenshot analysis, or when the user just needs to read a webpage.

Create, inspect, and edit Microsoft Excel workbooks and XLSX files with reliable formulas, dates, types, formatting, recalculation, and template preservation. Use when (1) the task is about Excel, `.xlsx`, `.xlsm`, `.xls`, `.csv`, or `.tsv`; (2) formulas, formatting, workbook structure, or compatibility matter; (3) the file must stay reliable after edits. NOT for: plain text data dumps, CSV-only workflows, or when pandas alone suffices.

Edit PDFs with natural-language instructions using the nano-pdf CLI. Use when the user wants to modify a specific page in a PDF using natural language. NOT for: creating PDFs from scratch, reading/extracting PDF content, or converting between formats.

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude realizes its knowledge is outdated or incorrect, (6) A better approach is discovered for a recurring task. Also review learnings before major tasks.

Create, inspect, and edit Microsoft Word documents and DOCX files with reliable styles, numbering, tracked changes, tables, sections, and compatibility checks. Use when (1) the task is about Word or `.docx`; (2) the file includes tracked changes, comments, fields, tables, templates, or page layout constraints; (3) the document must survive round-trip editing without formatting drift. NOT for: plain text documents, PDF editing, or when formatting doesn't matter.