writing-agent-harness

Generate and edit images with ByteDance Seedream 5.0 Pro/Lite image models: text-to-image, text-rich graphics, reference-image style transfer, no-mask colored-box local edits, outpainting, and batch generation. Use for production visual work, especially product/e-commerce imagery, editorial infographics, or replacing titles, objects, and materials in existing images.

Render Markdown into polished WeChat Official Account (微信公众号) HTML preview using this repo's style presets. Use when the user asks to 排版/美化/生成/预览微信公众号文章, wants a free md2wechat alternative, or needs a Markdown article converted into WeChat-ready inline HTML before 草稿箱/publish workflow.

把一篇本地 markdown 文档转写成飞书云文档(docx),排版精美、block 结构完整。 Mermaid 默认保留为 ```mermaid 代码块(源码可复制);--mermaid-mode whiteboard 时渲染成飞书画板。 触发:用户说"把 markdown 发到飞书"/"转写到飞书"/"这篇要发飞书";或给了本地 .md 路径,可附带 docx URL(不提供则创建新文档)。 不触发:微信公众号(走 wechat-publish-workflow)、Notion、Drive 原生 .md(走 lark-markdown)、docx 内部编辑(走 lark-doc)。

通过官方 `ntn` CLI 操作 Notion（创建/读取页面、上传文件/图片、设 cover/properties/icon、列出 blocks、追加内容）。 封装了已踩过的坑（空括号 hang、emoji 内联失败、YAML frontmatter 输入方向不解析、block type 不能 PATCH 改、paragraph.icon:null 拒绝等）， 提供 `uv run scripts/ntn_cli.py <cmd>` 统一入口，默认用 `ntn login` OAuth（无需 integration token 和 share connection）。 当其他 skill 需要写/读 Notion 页面，或 agent 自己要操作 Notion 时，优先用本 skill 而不是手写 `ntn api` 或直接调 REST API。

Generate article illustrations in multiple styles — watercolor art, flat illustrations, technical diagrams, infographics, and more. Use when agent needs covers, inset illustrations, atmospheric dividers, blog banners, or technical diagrams for articles. Supports text-only generation, reference-image-plus-text style guidance, and prompt-only dry-run review.

Generate, edit, and batch-generate raster images with OpenAI's gpt-image-2 model via the Image API. Use when the user wants to create or modify a picture file — a photo, poster, product shot, character, infographic, illustration, collage, banner, or slide. Handles background swaps, object removal, text replacement, style transfer, and up to 16 reference images. Batch mode for many prompts or JSONL files. gpt-image-2 is SOTA for in-image text rendering (2026/06) — ideal when the picture needs readable words. Outputs PNG, JPEG, or WebP. Do NOT use for SVG/vector icons, matplotlib charts, Mermaid/Excalidraw diagrams, HTML/CSS mockups, or other code-native output.

Publish finalized articles from Erik Lee's writing-agent-harness content/origin directory to Erik Lee's personal Astro blog repo eriklee-blog. Use only for Erik's local blog workflow: syncing Markdown/MDX, assets, taxonomy metadata, build checks, git commits/branches, and Cloudflare Pages deployment handoff.

Generate speech audio from text using Volcano Engine's (火山引擎) Doubao Speech Synthesis Model 2.0 (豆包语音合成大模型2.0, seed-tts-2.0). Use this skill whenever the user needs text-to-speech, voiceover, narration, dubbing, 配音, 旁白, 朗读, TTS, speech synthesis, or audio generation from text. Also use when the user mentions 火山引擎 TTS, 豆包语音, volcengine speech, or needs AI-generated voice audio for videos, podcasts, or presentations. Supports single-sentence and batch synthesis with Chinese, English, Japanese, Spanish, Indonesian, Portuguese, Korean, and dialects. This skill should ALWAYS be used for any TTS/语音合成 task — never attempt to call the Volcano Engine TTS API directly without it.

火山引擎 Seedance 2.0 视频生成。用户说”生成视频””做短视频””把这张图做成视频””图生视频””文生视频””产品广告视频””视频镜头””优化 Seedance 提示词”等时使用；支持文生视频、首帧/首尾帧图生视频、多模态参考（图+视+音组合）、批量并行 shots、任务创建/轮询/下载、任务列表/取消/删除、提示词优化。

用 Playwright 把微信公众号文章（文章流程）同步到草稿箱。输入 origin index.md（content/origin/YYYY-MM-DD-&lt;slug&gt;/，frontmatter 元数据权威源）+ wechat-article-renderer 产出的 HTML preview，自动登录态复用、注入正文、上传正文图片到微信 CDN、写标题/作者/摘要、保存草稿并报告 appmsgid。当用户要"发布公众号草稿""同步到草稿箱""publish wechat draft"时使用。仅创建草稿，不点发布/群发。

微信公众号文章发布 workflow。Use when the user wants to publish/sync/create 草稿箱 draft from a Markdown article: generate WeChat preview, verify HTML/images/封面, remove external href, use wechat-article-publisher, save 草稿箱, and hand off final publish review.

写作任务发布后 closeout。Use after WeChat/blog draft creation, publishing, or final handoff when the user says 任务收尾/归档/复盘/回填链接/清理素材/整理 memory/git/task; archive final article state, move image/video binaries to .local-archive/YYYY-MM-DD-<slug>/, preserve prompts/metadata/manifests/notes, update publish status, decide memory/skill improvements, and prepare git/task handoff without publishing.

将任意网页文章（微信公众号、技术博客、论文页面等）抓取、清洗并转写到用户指定的 Notion page 或 database row。 当用户给出一个或多个文章 URL 和一个 Notion 目标链接，并说"读到 Notion / 收藏到 Notion / 剪藏到 Notion / 整理进笔记 / 整理进 Notion"时，必须使用本 skill。 支持微信公众号 Playwright 抓取（含 data-src 懒加载图片）、通用站点 firecrawl/tavily 抓取、图片本地上传到 Notion（解决微信防盗链）、表格/代码块适配 Notion 排版、database property 启发式填充、cover 图片自动设置。 依赖 notion-cli skill 的 ntn CLI 封装；认证走 ntn login OAuth，无需手动创建 integration 或 share connection。

使用火山引擎 BigMusic (Seed-Music) 生成 AIGC 视频/文章配乐（无人声纯音乐）。 触发：BGM provider=bigmusic 时由 video-composer / final compose 调用， 或用户说"生成 BGM / 背景音乐 / 配乐 / 视频音乐"。 根据中文风格描述 + 时长生成乐器纯音乐（无人声），走按时长计费入口。

Fetch Volcano Engine (火山引擎) official documentation pages as clean Markdown. Use this skill whenever the user provides a volcengine.com/docs/ URL, wants to look up a Volcano Engine API reference, needs to check API parameters, models, or product documentation from 火山引擎, or mentions 火山引擎文档/火山引擎API/火山引擎接口文档. Also use when the user says "查一下火山引擎的...", "帮我看一下豆包/方舟/Seedance 的 API 文档", or pastes any volcengine.com URL. Supports single and multi-URL concurrent fetching with Playwright (default, free) and Firecrawl (fallback). This skill should ALWAYS be used for fetching Volcano Engine doc pages — never use WebFetch, firecrawl-scrape, or raw curl directly for volcengine.com URLs.

Create a lightly packaged article-ready video clip from a video-material-ingest package using ffmpeg and HyperFrames.

Help a human choose article-relevant highlight moments from a local video-material-ingest package. Use this whenever the user wants to find/select/review video highlights, create a contact sheet, choose clips for an article, or prepare start/end/title/caption candidates before running article-video-clip.

Ingest known video URLs with yt-dlp into traceable local asset folders for writing, visuals, short-video production, and HyperFrames.

提取微信公众号文章（mp.weixin.qq.com）为结构化 Markdown + assets，用于 AI 写作素材收集。Use when the user provides a WeChat article URL and wants to extract its content for research, reference, or writing material.

火山引擎联网搜索 API — 网页搜索和图片搜索。Use when the user wants to search the web, find articles, look up Chinese-language information, get recent news, or says 搜一下/查一下/搜索/找找/网上有没有/搜图片/中文搜索/国内信息. 基于火山引擎 Doubao Search API，中文内容覆盖和时效性优于通用搜索引擎。Trigger on any web search intent, especially Chinese-language or China-focused queries.

把模糊写作灵感打磨成清晰 writing brief 和 outline。Use whenever the user says 我想写一篇/有个想法/帮我理一下/脑暴/选题/文章思路/outline, or provides rough notes before drafting. This skill should run before research/draft/polish when the article angle, thesis, target reader, or structure is not yet clear.

发布前文章 readiness 检查。Use when the user says ready/发布前检查/最后看一遍/能不能进排版/能不能建草稿, or asks whether a Markdown/MDX article is ready for WeChat/blog rendering, 草稿箱, channel packaging, or final human review; verify editorial readiness, facts, Markdown/MDX hygiene, frontmatter, image/video references, and channel handoff blockers without doing post-publish cleanup.

润色和打磨文章写作。Use when the user asks to polish/精修/润色/压缩/扩写/restructure a blog post, essay, industry analysis, technical article, Agent/AI article, or Chinese/bilingual draft; improve logic, remove repetition, strengthen voice, raise professional depth, and adapt writing style to the genre while preserving the author's perspective.