skills

Karpathy's LLM Wiki: build/query interlinked markdown KB.

Use when parsing novel chapters into a structured knowledge base with chapter summaries, entity updates, change logs, memory points, and strict multi-step extraction/query/update discipline.

MiniMax API 调用指南，包含 M2.5/M2.7 模型选择、Python 调用示例、错误处理与性能对比。触发词：MiniMax API、调用 MiniMax、minimax-api

MiniMax M2.5/M2.7 模型上下文测试套件，支持 Recall、筛选、数学、代码、推理多维度测试。触发词：MiniMax 测试、上下文测试、minimax-context-testing

Use mmx to generate or analyze images via MiniMax (image-01 for generation, vision for understanding). Use when the user wants to generate images from text, analyze/describe images, extract text from images, check image content, or do character-consistent image generation. Triggers on "生成图片", "图片生成", "看图", "图片识别", "image generate", "describe image", "图片理解", "从图生图", "角色一致性".

一站式 MiniMax 图片生成封装脚本 mmx-image 的技能入口。 适用场景：（1）文生图（t2i）；（2）人物转换风格（i2i，保持同一人物特征）；（3）水印自动检测与裁剪；（4）参考图含 UI 元素时自动预清洗。 当用户说"生成图片"、"画一张图"、"把照片转成古风"、"人物换装"、或直接提供参考图要求 AI 基于该人物生成新图时使用此技能。

使用 mmx CLI 通过 MiniMax AI 生成音乐和歌曲翻唱。当用户想要创作音乐、生成歌曲、制作背景音乐（BGM）、创作器乐曲、翻唱歌曲，或提到"生成音乐"、"作曲"、"BGM"、"music generate"、"music cover"时使用。

Use when creating, updating, merging, or replacing project-level AI instruction files (CLAUDE.md / AGENTS.md / Agents.md) for Claude Code, OpenCode, Copilot CLI, Hermes Agent, or Gemini CLI.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Automatically detect and fix plugin project structure issues. Validates and corrects Skills, Agents, Commands, Hooks, Scripts, Templates, References against Claude Code plugin specifications.

Convert non-standard Claude Code plugin to standard format. Detects misplaced SKILL.md, skill.json, scripts and restructures to .claude-plugin/plugin.json + skills/name/SKILL.md format. Auto-registers to marketplace.json.

同步 `~/.agents` 目录到 Git 仓库。触发词：同步技能、sync agents、推送 skills、拉取已发布的技能更新。 自动使用 settings.json 中的环境变量（GIT_SSH_KEY、GIT_BRANCH 等）。 支持 --force 强制推送、--status 查看状态、--pull 仅拉取。 **破坏性操作（提交/拉取/推送/冲突解决）前必须先预览变更，用户确认后才执行。**

自动发现并整理本地 skill 目录为符号链接。先预览计划，用户确认后再执行。

Use mmx search to research coding solutions, find library docs, look up API references, investigate errors, or gather technical context before writing code. Triggers on "搜索解决方案", "查文档", "找库", "查 API", "搜错误", "research before coding", "look up how to", "find examples", "查怎么实现".

Use mmx vision to analyze code-related images — screenshots, architecture diagrams, UI mockups, error screenshots, database schemas, flowcharts — and generate coding plans or solutions. Triggers on "看这张截图", "分析这张图", "根据图片写代码", "看这个UI", "截图里的错误", "架构图", "看图写代码", "analyze screenshot", "code from image".

使用 MiniMax lyrics_generation API 直接生成歌词。当用户需要快速生成完整歌词、已有歌词需要修改润色时使用。触发词："生成歌词"、"写歌词"、"lyrics generation"、"歌词润色"、"修改歌词"。适合快速歌词生成场景，比 mmx-lyrics 更轻量。

Use mmx to generate song lyrics via MiniMax text model or music lyrics-optimizer. Use when the user wants to write song lyrics, generate verse/chorus structures, create lyrics for a specific style/theme, or needs structured lyrics with section tags for music generation. Triggers on "写歌词", "生成歌词", "lyrics", "创作歌词", "填词".

使用 MiniMax API 提取音频特征和歌词预处理。当用户想要分析音频文件、提取歌曲特征、预处理翻唱素材，或提到"音频预处理"、"提取特征"、"music_cover_preprocess"、"音频分析"时使用。

Use mmx to search the web via MiniMax search API. Use when the user wants to search the web, look up current information, find recent news, research a topic online, or verify facts. Triggers on "搜索", "查一下", "搜一搜", "最新消息", "查询", "网络搜索", "search", "look up", "find information about".

Use mmx to synthesize speech via MiniMax TTS (speech-2.8-hd). Use when the user wants text-to-speech, voice narration, audiobook, podcast scripts, subtitle generation, or high-quality audio from text. Triggers on "TTS", "语音合成", "朗读", "配音", "旁白", "生成音频", "文字转语音".

Use mmx to generate videos via MiniMax Hailuo models. Use when the user wants to generate video clips, text-to-video, image-to-video, subject reference video, or start-end frame interpolation. Triggers on "生成视频", "文生视频", "图生视频", "Hailuo", "视频片段", "T2V", "I2V", "video generate".

通用记忆库管理 — 结构性知识、任务计划、项目文档、问题解决方案