@zhaobod1

extracts visual brand specifications from existing websites and produces a machine-readable brand-spec.md via a 5-step process (ask, search, download, verify, codify). output integrates directly with huo15-openclaw-frontend-design for consistent design application.

huo15-wvi is a structured hr interview tool grounded in super's work values inventory, guiding users through five stages (17-item scoring, aspirations, non-negotiables, core three, satisfaction gap) to surface retention risk and career alignment.

huo15-influencer-video-skill generates first-person ecommerce videos via seedance 2.0, adding script-driven voiceover (edge-tts or volcano tts), auto bgm mixing, and subtitle burn. includes 8 persona templates with automatic selection by product category.

huo15-openclaw-design-critique provides structured evaluation of web design across 6 dimensions, aesthetic tier classification (AI-Slop to Master), visual era signal detection, and Master-level benchmark comparison. output includes radar chart, keep/fix/quick wins triage, and upgrade paths.

huo15-openclaw-asr converts audio, video, and terminal recordings to structured text via whisper or manual extraction, generates meeting notes from the transcription, then delivers both to openclaw for type classification and routing. handles quality checks and optional local export to markdown or word.

huo15-mind-map converts markdown outlines, json, opml, and xmind files into multi-format mind maps (xmind, png, pdf, svg, opml, freemind) with 13 configurable styles ranging from modern business to editorial magazine and swiss design aesthetics.

huo15-odoo19-module-dev covers odoo 19 module structure, model definitions, view design, and business logic through code templates and reference examples, but lacks explicit procedural steps and failure recovery patterns.

huo15-openclaw-plan-form normalizes heterogeneous excel production/planning tables into one of three standard templates, performs cell-by-cell diffing against prior versions, and flags changes with color coding and comments. requires manual validation of classification confidence and column mapping.

huo15-openclaw-openai-knowledge-base implements a karpathy-style llm librarian for managing structured knowledge across agent and shared scopes, with obsidian sync, typed relations, confidence scoring, and daily note bidirectional integration.

encodes karpathy's four llm programming guidelines (think before coding, simplicity first, surgical changes, goal-driven execution) as principles with examples and antipatterns, but lacks actionable procedures for agent invocation.

huo15-autoresearch-loop implements karpathy's autonomous research cycle (modify, verify, keep/discard, repeat) via git-backed iteration with configurable convergence criteria and state persistence.

huo15-comic-lipsync syncs lip movement to dialogue in video clips using kling 2.5 api, skipping silent shots and padding short audio to minimum duration.

huo15-comic-dub generates wav files from comic dialogue scripts using volcano ark seed-tts, assigns character voices automatically or from metadata, outputs named by scene-character-index.

huo15-skills provides enterprise word document generation supporting rule-based and template modes, with detailed formatting standards and document type templates. however, the actual skill implementation is missing, leaving unclear how content flows to output or handles failures.

huo15-openclaw-office-doc generates word and pdf documents across 39 enterprise templates (contracts, reports, manuals, proposals). covers trigger phrases and format presets extensively but lacks actionable procedures, error handling, and clear input/output contracts for end-to-end workflows.

generates three-image character card sets (full body, closeup, chibi) via seedream 4.0 api for comic storyboard consistency, reading character metadata from script json and outputting png assets to a manifest.

huo15-openclaw-design-director generates three contrasting design direction briefs from 24 design philosophies mapped to hex palettes, typeface pairs, and contemporary anchors, with audit tier escalation (junior/senior/master) and 2026 trend anchoring. procedure is aspirational rather than executable; lacks explicit failure recovery.

huo15-comic-script generates structured screenplay json from one-sentence themes in chinese fantasy genres (xianxia, palace intrigue, wuxia). output includes characters, scenes, dialogue, camera angles, and mood. designed as upstream dependency for downstream skills.

huo15-comic-bgm generates a single background track matched to script mood and duration using suno v5, with fallback to local chinese instrumental templates when unavailable.

huo15-openclaw-frontend-design documents a multi-genre UI design system with design tokens, mobile platform support, and 2026 trend patches, but the actual executable procedures are absent - mostly aspirational framework without step-by-step implementation paths.

Use this skill whenever the user wants to manage their work — to-dos / tasks ("记个待办" "加个任务" "我的待办" "做个待办事项" "标记完成"), projects and project tasks, timesheets /...

把 desktop-control MCP（火一五桌控，36 个 tool）用对的心法。 AX 树优先而非像素点击；act tool 必带 expect_after；session 状态摘要 [state ...] 要读。 触发：用户说"火一五桌控/桌控/帮我点/帮我截屏/控制电脑/自动化 macOS/打开 X...

麻省理工学院48小时学习法技能（青岛火一五信息科技有限公司）。完整还原 Ihtesham Ali 原始三问框架 + 反馈循环 + 完整 48h 三阶段时间线，叠加网上最佳实践（synthesis / contradictions / gaps / Feynman teach-back / weakness ana...

2026 现代美学的流程图、泳道图、系统架构图、C4 架构图、时序图、状态图、ER 图、甘特图生成。Linear/Vercel/Radix 配色 + 软阴影 + 圆角 + 判断色强调 + 容器分层。输入 YAML/JSON 规格或 Mermaid/PlantUML/DOT 源码，输出 SVG/PNG/PDF/dr...

火一五文生图提示词 v3.2 — AI 创作生态中枢，14 件套 + ⭐v3.2 审美锚点系统：88 预设新增 aesthetic_anchor 字段（每个预设映射到具体导演 / 摄影师 / 艺术家 / 电影场景，赛博朋克→Roger Deakins 银翼杀手 2049 / 汉服写真→张叔平一代宗师 / ghib...

从想法到论文的全自主研究管道，6阶段，含HITL，输出到Obsidian

火一五文生图提示词 v2.3 — 文生图&视频六件套：(1) enhance_prompt.py 文生图：88 风格预设 + 混合预设 (-p A+B --mix 0.6) + 五锁一致性 + 角色设定图 + 系列批量 + basic/pro/master 三档；(2) enhance_video.py 视频提示...

【青岛火一五信息科技有限公司】基于 Karpathy LLM Knowledge Base 三层架构（Data Ingest → Compilation → Active Maintenance）的知识捕获与管理技能。将知识点写入 memory/ 目录并同步到公司 Odoo 知识库。

一键生成 3-5 分钟国风 AI 漫剧。主题→分幕剧本→角色卡→分镜关键帧→图生视频→TTS 配音→对口型→BGM→字幕→拼接成片。自带三级成本熔断（硬限/预警/降级建议），支持失败续跑。触发词：AI 漫剧、生成漫剧、国风漫剧、仙侠短剧、comic drama、做漫剧。

读 script.json + 角色卡，每个镜头生成 1 张关键帧（Seedream 4.0 图生图，传入该镜出场角色的三联卡做多图参考保持角色一致性）。触发词：分镜关键帧、storyboard、漫剧分镜图。

分镜关键帧 → 短视频（Seedance 2.0 图生视频，关键帧做 first_frame，5s/镜，最多 3 并发，开启 return_last_frame 用于下镜衔接）。触发词：图生视频、分镜视频、漫剧视频化。

对最近修改的代码做"复用/质量/效率"三维审查，产出可执行的清理清单；然后实际修复命中的问题。用于刚写完功能、PR 提交前、重构前的自检。触发词：简化代码、清理冗余、重构建议、simplify、code cleanup、代码体检、能不能更简洁。

对当前分支的 pending changes 做安全评审，命中密钥泄露 / SQL 注入 / XSS / SSRF / 权限绕过 / 危险依赖 六大类漏洞后出分级报告，再在用户批准下修复。用于 PR 合并前、对外开源前、线上事故后复盘。触发词：安全评审、security review、漏洞检查、密钥扫描、我的代码...

对 GitHub / cnb.cool PR 做综合代码评审（设计 / 实现 / 测试 / 安全 / 可维护五维），借 gh CLI 拉 diff，产出可粘贴到 PR 的评论 markdown。触发词：评审 PR、code review、审一下这个 PR、帮我 review、看看这个合并请求。

FFmpeg 把 lipsync 视频按顺序 concat + 叠 BGM + 烧字幕 + 0.3s 淡入淡出，输出 final.mp4。触发词：视频拼接、成片合成、FFmpeg 拼接。

火一五排版发布技能 / 火一五 markdown 排版 / 火一五 PDF 导出 / 火一五出版 / 火一五发布 / huo15-markdown-export — 【青岛火一五】markdown 一键导出 PDF / Word / HTML / 长图 / 公众号 inline。**12 主题**默认 **app...

自然语言操作 辉火云企业套件 — 实施经理助手，覆盖任务/CRM/项目/工单/财务/销售/HR/库存/生产/客服/知识/Studio 元编程 13 业务域 204 个工具。v1.23.0 默认权限模型反转：odoo_connect 默认按 sender_id 隔离（每个企微/钉钉/飞书成员配自己的 Odoo 账号...

【青岛火一五信息科技有限公司】企业级 Word 文档生成技能，支持两种模式：规则模式（默认）和模板模式。触发词：写word、写文档、生成word、生成文档、创建文档、.docx、Word文档、写合同、写方案、写报告、写会议纪要、按模板生成。

Use when the user wants to write, analyze, or improve Xiaohongshu (小红书) content — drafting notes, coaching writing skills, diagnosing AI-speak or Jarvis-trap...

绿火 — 火一五旗下 CRM 销售员智能体（拓客 / 获客 / 跟进 / 客户画像 / 销售话术 / 信号识别）。**必须先装 @huo15/huo15-huihuoyun-odoo** 才能落地操作辉火云 Odoo19 企业版 CRM。v0.2 起接入企查查 OpenAPI 工商底座（需配 qichacha_a...

火一五·克劳德·龙虾增强插件 v6.7.13 — large-file-bridge 兜底死循环修 + 用户上传后流程引导：实测用户『我已经上传好了』后 AI 没调 enhance_upload_check 反被兜底又塞了一份上传链接死循环。根因：before_agent_reply 每轮都触发,injected...

火一五·企业微信（WeCom）OpenClaw 插件 v2.8.25 — 默认走 Bot WebSocket（响应快、配置简单），自带加密媒体解密 / Agent 主动发消息 / 微信客服三通道接入 / 多账号切换。v2.8.25 重点：GUIDANCE 优先级翻转回 MEDIA: 直发——v2.8.22 当时担...

OpenClaw 微信服务号（公众号）渠道插件 v2.3.5 hotfix —— **修 errcode 45002 content size out of limit**：truncateForWechatText 改字节截断（微信限 2048 字节不是字符，中文 1 字 = 3 字节），默认 1900 字节留...

OWL (Odoo Web Library) 开发技能 — Odoo 前端组件开发框架，基于 QWeb 模板、响应式状态管理、OWL 组件生命周期