# agent-skills
[](https://github.com/davidtoby/agent-skills/actions/workflows/validate-skills-repo.yml)
[](https://github.com/davidtoby/agent-skills/actions/workflows/release-skill-packages.yml)
[](https://github.com/davidtoby/agent-skills/releases)
[](./LICENSE)
**Build once. Ship to many agents.**
**一次沉淀，多个 Agent 复用。**
A bilingual repository of reusable skills for **OpenClaw**, **Hermes Agent**, **Claude Code**, **Codex**, and other agent runtimes.
This repo stores skills that survived real work: not toy prompts, but repeatable workflows, helper scripts, references, packaged artifacts, and field-tested delivery patterns.
这是一个面向 **OpenClaw**、**Hermes Agent**、**Claude Code**、**Codex** 以及其他 Agent 运行时的**双语 Skill 仓库**。
这里沉淀的是经真实任务验证后仍然值得复用的 Skill：不是一次性 prompt，而是可重复执行的工作流、脚本、参考资料、打包产物，以及踩坑之后留下来的可靠方法。
---
## Quick start / 快速开始
### English
If you only want to use a skill quickly:
1. Open the target skill under [`skills/`](./skills/)
2. Read `SKILL.md`
3. Copy the folder into your own skill workspace
4. If your runtime supports direct import, use the matching `.skill` file under [`packages/`](./packages/)
Example:
```bash
cp -R skills/consulting-pdf-from-youtube /path/to/your/skills/
```
### 中文
如果你只是想快速使用某个 Skill：
1. 打开 [`skills/`](./skills/) 下对应的目录
2. 先读 `SKILL.md`
3. 把整个目录复制到你自己的 skill workspace
4. 如果你的运行环境支持直接导入，可以直接使用 [`packages/`](./packages/) 下对应的 `.skill` 文件
示例：
```bash
cp -R skills/consulting-pdf-from-youtube /path/to/your/skills/
```
---
## Choose the right skill / 如何选 Skill
### English
Use this quick map if you are not sure where to start:
- **I have an academic paper PDF and want a Chinese insight handout**
→ [`academic-paper-to-chinese-insight-pdf`](./skills/academic-paper-to-chinese-insight-pdf/)
- **I already have Chinese report content, but the PDF typography/font rendering is bad**
→ [`chinese-pdf-report`](./skills/chinese-pdf-report/)
- **I have a YouTube link and want a polished report or consulting-style PDF**
→ [`consulting-pdf-from-youtube`](./skills/consulting-pdf-from-youtube/)
- **My video has no subtitles / I want a Chinese PDF report from any video**
→ [`chinese-video-transcribe-pdf`](./skills/chinese-video-transcribe-pdf/)
- **I need bilingual English/Chinese subtitles for a video**
→ [`video-bilingual-subtitle-delivery`](./skills/video-bilingual-subtitle-delivery/)
- **I have Chinese draft text that feels templated or AI-written and want a more natural rewrite**
→ [`humanizer-zh`](./skills/humanizer-zh/)
- **My GitHub push/auth flow is broken and I need a reliable recovery path**
→ [`github-auth-recovery`](./skills/github-auth-recovery/)
- **Feishu approval buttons fail, show 200340, or users must type `/approve always` manually**
→ [`feishu-approval-fallback`](./skills/feishu-approval-fallback/)
- **Feishu websocket mode reconnects too slowly or seems unstable**
→ [`feishu-websocket-stability`](./skills/feishu-websocket-stability/)
- **My OpenClaw Feishu bot stops replying, throws unexpected pairing codes, or two Feishu bot windows seem swapped**
→ [`openclaw-feishu-bot-troubleshooting`](./skills/openclaw-feishu-bot-troubleshooting/)
- **I need the full fresh-machine setup for a local LiteLLM + Vertex Gemini gateway on macOS, including LaunchAgent autostart plus Claude Code and OpenClaw integration**
→ [`litellm-vertex-gemini-local-gateway`](./skills/litellm-vertex-gemini-local-gateway/)
- **My local LiteLLM gateway already works and I only want OpenClaw to use that Gemini route safely, with a reversible config change**
→ [`openclaw-litellm-gemini-gateway`](./skills/openclaw-litellm-gemini-gateway/)
- **My local LiteLLM + Vertex AI proxy is down, stuck at startup, or `/ui/login/` says `Not connected to DB!`**
→ [`litellm-vertex-proxy-repair`](./skills/litellm-vertex-proxy-repair/)
- **I want Claude Code to use an existing LiteLLM gateway, especially for Gemini via Vertex AI, without breaking my current Claude/Hermes setup**
→ [`claude-code-litellm-gateway`](./skills/claude-code-litellm-gateway/)
- **I want native OpenCode to show a local LiteLLM-routed Gemini model in `/models`, without wrappers or changing the default model**
→ [`opencode-litellm-gemini-provider`](./skills/opencode-litellm-gemini-provider/)
### 中文
如果你不确定该选哪个 Skill，可以先看这个快速决策区：
- **我有一篇学术论文 PDF，想做成中文精读/洞察版讲义**
→ [`academic-paper-to-chinese-insight-pdf`](./skills/academic-paper-to-chinese-insight-pdf/)
- **我已经有中文报告内容，但 PDF 字体/排版很差**
→ [`chinese-pdf-report`](./skills/chinese-pdf-report/)
- **我有一个 YouTube 链接，想产出高质量报告或咨询风 PDF**
→ [`consulting-pdf-from-youtube`](./skills/consulting-pdf-from-youtube/)
- **视频没有字幕 / 想从任意视频生成中文 PDF 报告**
→ [`chinese-video-transcribe-pdf`](./skills/chinese-video-transcribe-pdf/)
- **我需要给视频做中英双语字幕**
→ [`video-bilingual-subtitle-delivery`](./skills/video-bilingual-subtitle-delivery/)
- **我有一段中文草稿，读起来模板化或 AI 味很重，想保留原意但改得更自然**
→ [`humanizer-zh`](./skills/humanizer-zh/)
- **我的 GitHub push / 认证流程坏了，需要一条可靠恢复路径**
→ [`github-auth-recovery`](./skills/github-auth-recovery/)
- **Feishu 审批按钮失效、报 200340，或只能手动输入 `/approve always`**
→ [`feishu-approval-fallback`](./skills/feishu-approval-fallback/)
- **Feishu websocket 模式重连太慢，或连接稳定性可疑**
→ [`feishu-websocket-stability`](./skills/feishu-websocket-stability/)
- **我需要在全新或近乎全新的 macOS 环境里，从零搭建本地 LiteLLM + Vertex Gemini 网关，并完成 LaunchAgent 自启动、Claude Code 与 OpenClaw 接入**
→ [`litellm-vertex-gemini-local-gateway`](./skills/litellm-vertex-gemini-local-gateway/)
- **本地 LiteLLM 网关已经可用，我只想把 OpenClaw 稳妥接到这条 Gemini 路线上，并保留可回滚路径**
→ [`openclaw-litellm-gemini-gateway`](./skills/openclaw-litellm-gemini-gateway/)
- **本地 LiteLLM 网关已经可用，我只想让原生 OpenCode 在 `/models` 中出现 Gemini，并且不改默认模型、不使用 wrapper**
→ [`opencode-litellm-gemini-provider`](./skills/opencode-litellm-gemini-provider/)
- **OpenClaw 飞书 Bot 不回复、突然吐 pairing code，或两个飞书 bot 窗口像是对错了身份**
→ [`openclaw-feishu-bot-troubleshooting`](./skills/openclaw-feishu-bot-troubleshooting/)
---
## Topic guides / 专题索引
- [`topics/feishu-hermes-stability.md`](./topics/feishu-hermes-stability.md) — Feishu / Hermes 稳定性专题：审批按钮 200340、文本兜底、websocket 重连调优、GitHub 发布回写
---
## Local shared library sync / 本地共享技能库同步
This repository is the **publish/share/backup companion** to the local shared skill library:
- Local canonical library: `~/.agents/skills/`
- Public/shared repository: `https://github.com/davidtoby/agent-skills`
Default merge policy:
- **Remote wins on conflicts**: if the same skill path diverges both locally and on GitHub, sync the GitHub version back to local.
- **Local new active skills publish upstream**: validated local skills that are new in `~/.agents/skills/` should be copied into this repo, indexed, packaged, and pushed upstream.
- `openclaw-imports/` is an explicit imported/compatibility namespace. Those skills are allowed in this repo when they are part of the shared local library snapshot.
For local multi-agent usage, the operational entry point is `~/.agents/skills/README.md`.
Detailed sync policy: [`SYNC.md`](./SYNC.md)
---
## Why this repo exists / 为什么做这个仓库
### English
Most AI workflows die after one chat. Good skills do not.
This repo exists to turn one-off agent heroics into reusable, inspectable, shareable assets.
### 中文
大多数 AI 工作流都死在一次对话里，真正好的 Skill 不会。
这个仓库的目标，是把一次性的 Agent 灵光乍现，沉淀成**可复用、可审阅、可共享、可迭代**的资产。
---
## What lives here / 这里放什么
### English
- reusable agent skills
- source skill folders
- helper scripts
- references and troubleshooting notes
- example assets when they materially improve reuse
- optional packaged `.skill` artifacts for distribution
### 中文
- 可复用的 Agent Skill
- Skill 源码目录
- 配套脚本
- 参考文档与故障排查说明
- 只有在确实提升复用价值时才保留的示例资产
- 用于分发的可选 `.skill` 打包产物
---
## How to read this repo / 怎么理解这个仓库
| Path | Purpose |
|---|---|
| [`skills/`](./skills/) | Source-of-truth skill folders / Skill 源码主目录 |
| [`packages/`](./packages/) | Packaged `.skill` artifacts / 已打包的 `.skill` 文件 |
| [`skills/README.md`](./skills/README.md) | Directory-level skill index / `skills/` 目录索引 |
| [`CONTRIBUTING.md`](./CONTRIBUTING.md) | Contribution guide / 贡献说明 |
| [`CHANGELOG.md`](./CHANGELOG.md) | Important repository changes / 重要变更记录 |
| [`.github/workflows/`](./.github/workflows/) | Validation and release automation / 校验与发布自动化 |
---
## Skill catalog / 技能目录
| Skill | Problem it solves / 解决的问题 | Main assets / 主要内容 |
|---|---|---|
| [`academic-paper-to-chinese-insight-pdf`](./skills/academic-paper-to-chinese-insight-pdf/) | Turn an academic paper PDF into a readable Chinese insight report and polished Chinese PDF / 把学术论文 PDF 转成更易读的中文精读/洞察版 PDF | extraction script, PDF renderer, structure and quality references |
| [`agent-skills-frontmatter-compatibility`](./skills/agent-skills-frontmatter-compatibility/) | Fix brittle SKILL frontmatter issues that break packaging/validation when publishing local skills into this repo / 修复把本地 skill 发布到本仓库时，由脆弱 SKILL frontmatter 引发的打包与校验失败 | compatibility checklist, real repair examples, repo-specific rebuild guidance |
| [`chinese-pdf-report`](./skills/chinese-pdf-report/) | Generate professional Chinese PDF reports with reliable font rendering on macOS / 生成中文字体更稳、排版更专业的 PDF 报告 | renderer script, troubleshooting, font notes, example assets |
| [`consulting-pdf-from-youtube`](./skills/consulting-pdf-from-youtube/) | Download a YouTube video, extract transcript/metadata, and export premium PDF report variants / 下载 YouTube 视频、提取字幕与元数据，并导出高端报告风格 PDF | workflow guide, output package reference, style-variant reference |
| [`video-bilingual-subtitle-delivery`](./skills/video-bilingual-subtitle-delivery/) | Create, repair, audit, and deliver bilingual subtitles with shared EN/ZH timing / 创建、修复、审计并交付中英双语字幕 | subtitle scripts, workflow docs, troubleshooting, real repair lessons |
| [`humanizer-zh`](./skills/humanizer-zh/) | Rewrite Chinese text to remove templated AI-writing patterns while preserving meaning, audience, and an authentic human voice / 去除中文文本的模板化 AI 痕迹，同时保留原意、受众与真实表达 | 24-pattern checklist, rewrite workflow, quality rubric; adapted from [op7418/Humanizer-zh](https://github.com/op7418/Humanizer-zh) |
| [`github-auth-recovery`](./skills/github-auth-recovery/) | Recover from broken GitHub auth/push flows when `gh` is logged out, HTTPS credentials fail, or SSH may already work / 当 `gh` 未登录、HTTPS 凭证失效、但 SSH 可能已可用时，恢复 GitHub push/认证流程 | concise recovery workflow, SSH-vs-HTTPS reference |
| [`litellm-vertex-gemini-local-gateway`](./skills/litellm-vertex-gemini-local-gateway/) | Build a fresh local LiteLLM + Vertex Gemini gateway on macOS end to end, including LaunchAgent autostart plus Claude Code and OpenClaw integration / 在 macOS 上从零搭建本地 LiteLLM + Vertex Gemini 网关，并完成 LaunchAgent 自启动、Claude Code 与 OpenClaw 集成 | scaffold generator, config templates, verification runbook, Claude/OpenClaw integration notes |
| [`openclaw-litellm-gemini-gateway`](./skills/openclaw-litellm-gemini-gateway/) | Connect OpenClaw to an already-working local LiteLLM Gemini gateway with low-risk config edits, exact `openclaw.json` snippets, verification commands, and rollback guidance / 把 OpenClaw 接到已经可用的本地 LiteLLM Gemini 网关上，包含低风险配置修改、精确 `openclaw.json` 片段、验证命令与回滚指南 | provider snippet, model alias workflow, verification order, rollback notes |
| [`opencode-litellm-gemini-provider`](./skills/opencode-litellm-gemini-provider/) | Add a local LiteLLM-routed Gemini model to native OpenCode as a selectable `/models` provider without wrappers or changing the default model / 把本地 LiteLLM 路由的 Gemini 模型加入原生 OpenCode 的 `/models`，不使用 wrapper，也不修改默认模型 | setup script, verification script, config-file secret reference workflow |
| [`litellm-vertex-proxy-repair`](./skills/litellm-vertex-proxy-repair/) | Diagnose and repair a local LiteLLM + Vertex AI proxy on macOS when `/v1` is down, startup hangs, or Prisma/Admin UI failures need to be isolated from API recovery / 诊断并修复 macOS 本地 LiteLLM + Vertex AI 代理：当 `/v1` 不通、启动卡住、或 Prisma/Admin UI 故障需要和 API 恢复拆开处理时使用 | startup triage, loopback-proxy diagnosis, lite/full split, verification checklist |
| [`claude-code-litellm-gateway`](./skills/claude-code-litellm-gateway/) | Configure Claude Code to use an existing LiteLLM gateway safely, especially for Gemini/Vertex-backed routes, without breaking other Claude or agent setups / 安全地把 Claude Code 接到现有 LiteLLM 网关上，尤其适用于 Gemini/Vertex 路由，同时不破坏已有 Claude 或其他 Agent 配置 | Anthropic-style gateway config, wrapper pattern, settings conflict isolation, real working example |
| [`feishu-approval-fallback`](./skills/feishu-approval-fallback/) | Recover and harden Hermes Feishu command approvals when card buttons fail, 200340 appears, or users need text fallback commands / 修复并加固 Hermes 的 Feishu 命令审批流程：按钮失效、出现 200340、或需要文本兜底时使用 | fallback workflow, callback-sync guidance, real fix reference |
| [`feishu-websocket-stability`](./skills/feishu-websocket-stability/) | Stabilize Hermes Feishu websocket mode when reconnects are too slow, ping timing needs tuning, or duplicate local clients compete for one app_id / 稳定 Hermes 的 Feishu websocket 模式：重连过慢、ping 需调优、或本地重复客户端冲突时使用 | reconnect tuning, app-lock pattern, runtime override guidance |
More detail: see [`skills/README.md`](./skills/README.md).
---
## Featured skills / 精选技能
Each featured skill follows the same reading order:
1. what it does / 它做什么
2. why it is useful / 为什么有用
3. quick entry / 快速入口
### `academic-paper-to-chinese-insight-pdf`
**What it does / 它做什么**
- Reads an academic paper PDF and turns it into a readable Chinese insight report
- Exports the result as a polished Chinese PDF
- Better suited for “understand this paper” tasks than raw page-by-page translation
- 读取学术论文 PDF，并整理成可读性更高的中文洞察版报告
- 输出成品级中文 PDF
- 更适合“帮我读懂这篇论文”类任务，而不是机械逐页直译
**Why it is useful / 为什么有用**
- prefers an insight edition when that is more useful than literal translation
- keeps paper facts separate from your own commentary
- uses a clear output structure so different reports stay consistent
- 在“洞察版比直译版更有用”的情况下，默认优先洞察版
- 明确区分论文事实与个人分析
- 输出结构清晰，方便形成稳定交付风格
**Quick entry / 快速入口**
- Source: [`skills/academic-paper-to-chinese-insight-pdf/`](./skills/academic-paper-to-chinese-insight-pdf/)
- Package: [`packages/academic-paper-to-chinese-insight-pdf.skill`](./packages/academic-paper-to-chinese-insight-pdf.skill)
### `chinese-pdf-report`
**What it does / 它做什么**
- Generates professional Chinese-first PDF reports
- Focuses on reliable Chinese font rendering and stronger typography on macOS
- Useful when HTML-to-PDF output looks technically OK but visually weak
- 生成中文优先的专业 PDF 报告
- 重点解决 macOS 上中文字体渲染和排版质量问题
- 适合“HTML 转 PDF 勉强能出，但成品不够专业”的场景
**Why it is useful / 为什么有用**
- uses explicit font control instead of fragile fallback chains
- includes troubleshooting notes for garbling and missing glyphs
- comes with example assets, not just abstract advice
- 用显式字体控制替代脆弱的自动回退链路
- 带有乱码、缺字、字体注册失败等问题的排查说明
- 不只是抽象原则，还附带真实示例资产
**Quick entry / 快速入口**
- Source: [`skills/chinese-pdf-report/`](./skills/chinese-pdf-report/)
- Package: [`packages/chinese-pdf-report.skill`](./packages/chinese-pdf-report.skill)
### `consulting-pdf-from-youtube`
**What it does / 它做什么**
- Downloads a YouTube video
- extracts transcript/subtitle + metadata
- synthesizes viewpoints, takeaways, and insights
- exports premium PDF report variants such as consulting, McKinsey-style, BCG-style, and Apple-inspired branding editions
- 下载 YouTube 视频
- 提取字幕/转录与元数据
- 整理核心观点、Key Takeaways 和个人洞察
- 导出咨询风、麦肯锡风、BCG 风、Apple 风品牌版等高端 PDF 报告
**Why it is useful / 为什么有用**
- starts from transcript reliability, not from vague summarization
- keeps Markdown, HTML, and PDF as layered deliverables
- includes explicit QA expectations such as page count and text extraction checks
- 先保证可分析的字幕/转录，再做报告
- Markdown、HTML、PDF 分层交付，便于修改和复用
- 明确要求做页数、文本提取、视觉层级等 QA 验证
**Quick entry / 快速入口**
- Source: [`skills/consulting-pdf-from-youtube/`](./skills/consulting-pdf-from-youtube/)
- Package: [`packages/consulting-pdf-from-youtube.skill`](./packages/consulting-pdf-from-youtube.skill)
### `video-bilingual-subtitle-delivery`
**What it does / 它做什么**
- Creates, repairs, audits, and delivers bilingual subtitles
- keeps English timing as the source of truth
- aligns Chinese to the same subtitle event
- supports both softsub and hardcode delivery
- 创建、修复、审计并交付中英双语字幕
- 以英文时间轴为基准
- 把中文挂在同一字幕事件下
- 同时支持 softsub 与 hardcode 交付
**Why it is useful / 为什么有用**
- prevents “wrong source cut” and “missing Chinese lines” from slipping into final delivery
- includes scripts for audit and hardcode fallback
- captures real lessons from subtitle rebuild work
- 能防止“用错视频 cut”与“中文漏挂”进入最终交付
- 自带审计脚本和硬字幕 fallback 脚本
- 总结了真实字幕修复任务中的经验
**Quick entry / 快速入口**
- Source: [`skills/video-bilingual-subtitle-delivery/`](./skills/video-bilingual-subtitle-delivery/)
- Package: [`packages/video-bilingual-subtitle-delivery.skill`](./packages/video-bilingual-subtitle-delivery.skill)
---
## Packaged `.skill` artifacts / 已打包的 `.skill` 文件
Use a packaged `.skill` artifact when your runtime supports direct import. Otherwise, the source skill folder is the safest and most portable format.
如果你的运行环境支持直接导入 `.skill`，可以直接使用打包文件；如果不支持，最稳妥的方式仍然是直接使用源码目录。
Current packaged artifacts:
<!-- package-list:start -->
- [`packages/academic-paper-to-chinese-insight-pdf.skill`](./packages/academic-paper-to-chinese-insight-pdf.skill)
- [`packages/agent-skills-frontmatter-compatibility.skill`](./packages/agent-skills-frontmatter-compatibility.skill)
- [`packages/agent-skills-repo-publishing.skill`](./packages/agent-skills-repo-publishing.skill)
- [`packages/andrej-karpathy-perspective.skill`](./packages/andrej-karpathy-perspective.skill)
- [`packages/beautiful-feishu-card.skill`](./packages/beautiful-feishu-card.skill)
- [`packages/beautiful-feishu-whiteboard.skill`](./packages/beautiful-feishu-whiteboard.skill)
- [`packages/chinese-pdf-report.skill`](./packages/chinese-pdf-report.skill)
- [`packages/chinese-video-transcribe-pdf.skill`](./packages/chinese-video-transcribe-pdf.skill)
- [`packages/claude-code-litellm-gateway.skill`](./packages/claude-code-litellm-gateway.skill)
- [`packages/consulting-pdf-from-youtube.skill`](./packages/consulting-pdf-from-youtube.skill)
- [`packages/dogfood.skill`](./packages/dogfood.skill)
- [`packages/elon-musk-perspective.skill`](./packages/elon-musk-perspective.skill)
- [`packages/feishu-approval-fallback.skill`](./packages/feishu-approval-fallback.skill)
- [`packages/feishu-websocket-stability.skill`](./packages/feishu-websocket-stability.skill)
- [`packages/feynman-perspective.skill`](./packages/feynman-perspective.skill)
- [`packages/find-skills.skill`](./packages/find-skills.skill)
- [`packages/github-auth-recovery.skill`](./packages/github-auth-recovery.skill)
- [`packages/hermes-agent-litellm-vertex-gateway.skill`](./packages/hermes-agent-litellm-vertex-gateway.skill)
- [`packages/huashu-design.skill`](./packages/huashu-design.skill)
- [`packages/huashu-nuwa.skill`](./packages/huashu-nuwa.skill)
- [`packages/humanizer-zh.skill`](./packages/humanizer-zh.skill)
- [`packages/ilya-sutskever-perspective.skill`](./packages/ilya-sutskever-perspective.skill)
- [`packages/lark-approval.skill`](./packages/lark-approval.skill)
- [`packages/lark-apps.skill`](./packages/lark-apps.skill)
- [`packages/lark-attendance.skill`](./packages/lark-attendance.skill)
- [`packages/lark-base.skill`](./packages/lark-base.skill)
- [`packages/lark-calendar.skill`](./packages/lark-calendar.skill)
- [`packages/lark-contact.skill`](./packages/lark-contact.skill)
- [`packages/lark-doc.skill`](./packages/lark-doc.skill)
- [`packages/lark-drive.skill`](./packages/lark-drive.skill)
- [`packages/lark-event.skill`](./packages/lark-event.skill)
- [`packages/lark-im.skill`](./packages/lark-im.skill)
- [`packages/lark-mail.skill`](./packages/lark-mail.skill)
- [`packages/lark-markdown.skill`](./packages/lark-markdown.skill)
- [`packages/lark-minutes.skill`](./packages/lark-minutes.skill)
- [`packages/lark-note.skill`](./packages/lark-note.skill)
- [`packages/lark-okr.skill`](./packages/lark-okr.skill)
- [`packages/lark-openapi-explorer.skill`](./packages/lark-openapi-explorer.skill)
- [`packages/lark-shared.skill`](./packages/lark-shared.skill)
- [`packages/lark-sheets.skill`](./packages/lark-sheets.skill)
- [`packages/lark-skill-maker.skill`](./packages/lark-skill-maker.skill)
- [`packages/lark-slides.skill`](./packages/lark-slides.skill)
- [`packages/lark-task.skill`](./packages/lark-task.skill)
- [`packages/lark-vc.skill`](./packages/lark-vc.skill)
- [`packages/lark-vc-agent.skill`](./packages/lark-vc-agent.skill)
- [`packages/lark-whiteboard.skill`](./packages/lark-whiteboard.skill)
- [`packages/lark-wiki.skill`](./packages/lark-wiki.skill)
- [`packages/lark-workflow-meeting-summary.skill`](./packages/lark-workflow-meeting-summary.skill)
- [`packages/lark-workflow-standup-report.skill`](./packages/lark-workflow-standup-report.skill)
- [`packages/litellm-vertex-gemini-local-gateway.skill`](./packages/litellm-vertex-gemini-local-gateway.skill)
- [`packages/litellm-vertex-proxy-repair.skill`](./packages/litellm-vertex-proxy-repair.skill)
- [`packages/mrbeast-perspective.skill`](./packages/mrbeast-perspective.skill)
- [`packages/munger-perspective.skill`](./packages/munger-perspective.skill)
- [`packages/naval-perspective.skill`](./packages/naval-perspective.skill)
- [`packages/openclaw-feishu-bot-troubleshooting.skill`](./packages/openclaw-feishu-bot-troubleshooting.skill)
- [`packages/openclaw-litellm-gemini-gateway.skill`](./packages/openclaw-litellm-gemini-gateway.skill)
- [`packages/opencode-litellm-gemini-provider.skill`](./packages/opencode-litellm-gemini-provider.skill)
- [`packages/output-workspace-cleanup.skill`](./packages/output-workspace-cleanup.skill)
- [`packages/paul-graham-perspective.skill`](./packages/paul-graham-perspective.skill)
- [`packages/precise-bilingual-subtitle.skill`](./packages/precise-bilingual-subtitle.skill)
- [`packages/steve-jobs-perspective.skill`](./packages/steve-jobs-perspective.skill)
- [`packages/steve-jobs-skill.skill`](./packages/steve-jobs-skill.skill)
- [`packages/sun-yuchen-perspective.skill`](./packages/sun-yuchen-perspective.skill)
- [`packages/sync-shared-skills-repo.skill`](./packages/sync-shared-skills-repo.skill)
- [`packages/taleb-perspective.skill`](./packages/taleb-perspective.skill)
- [`packages/trump-perspective.skill`](./packages/trump-perspective.skill)
- [`packages/video-bilingual-subtitle-delivery.skill`](./packages/video-bilingual-subtitle-delivery.skill)
- [`packages/x-mastery-mentor.skill`](./packages/x-mastery-mentor.skill)
- [`packages/youtube-bilingual-subtitle-delivery.skill`](./packages/youtube-bilingual-subtitle-delivery.skill)
- [`packages/yuanbao.skill`](./packages/yuanbao.skill)
- [`packages/zhang-yiming-perspective.skill`](./packages/zhang-yiming-perspective.skill)
- [`packages/zhangxuefeng-perspective.skill`](./packages/zhangxuefeng-perspective.skill)
<!-- package-list:end -->
### How packaging works / 打包方式
```bash
python3 /opt/homebrew/lib/node_modules/openclaw/skills/skill-creator/scripts/package_skill.py \
/path/to/skill-folder \
/path/to/output-dir
```
This flow validates the skill structure first, then emits the `.skill` bundle.
这个流程会先校验 Skill 结构，再生成 `.skill` 打包文件。
---
## Validation and release automation / 校验与发布自动化
### Validation CI
- Workflow: [`.github/workflows/validate-skills-repo.yml`](./.github/workflows/validate-skills-repo.yml)
- Script: [`scripts/validate_skills_repo.py`](./scripts/validate_skills_repo.py)
- Sync helper: [`scripts/sync_package_lists.py`](./scripts/sync_package_lists.py)
Checks include:
- every skill folder has `SKILL.md`
- required markers exist in `SKILL.md`
- markdown fences are balanced
- every skill has a matching `.skill` package
- package contents are fresh and in sync with source files
- package lists in `README.md` and `skills/README.md` exactly match the real `packages/` directory
- repo-level indexes mention every skill
校验内容包括：
- 每个 skill 目录都有 `SKILL.md`
- `SKILL.md` 含有必要字段标记
- Markdown 代码块成对闭合
- 每个 skill 都有对应 `.skill` 包
- `.skill` 包内容与源码目录保持同步，不允许陈旧包漂移
- `README.md` 与 `skills/README.md` 里的 package 列表必须和真实 `packages/` 目录完全一致
- README 与 skills/README 不漏掉任何 skill
### Release workflow
- Workflow: [`.github/workflows/release-skill-packages.yml`](./.github/workflows/release-skill-packages.yml)
- Trigger: tag push like `v0.2.0` or manual dispatch
- Result: create a GitHub Release and upload all `packages/*.skill` as release assets
### Rebuild helper
- Script: [`scripts/rebuild_all_packages.py`](./scripts/rebuild_all_packages.py)
- Purpose: rebuild every `.skill` package from the current source folders, or rebuild only selected skills, sync package lists in README files, then run repository validation
Common usage:
```bash
# rebuild all packages, then validate
python scripts/rebuild_all_packages.py
# rebuild one skill only, then validate
python scripts/rebuild_all_packages.py --skill consulting-pdf-from-youtube
# just run validation without rebuilding
python scripts/rebuild_all_packages.py --check-only
# sync README package lists with packages/
python scripts/sync_package_lists.py
# only check whether README package lists drifted from packages/
python scripts/sync_package_lists.py --check-only
```
- 触发方式：推送 `v*` tag，或手动运行 workflow
- 结果：创建 GitHub Release，并把所有 `packages/*.skill` 作为 release assets 上传
- 日常维护时，如果改了 skill 源码又想批量重打包，可以直接运行上面的脚本
- 如果只改了某一个 skill，也可以用 `--skill <name>` 精确重打
---
## Star History / Star 趋势
If you find this repository useful, please consider starring it and contributing improvements.
A visible star-history chart helps maintainers and users see whether the project is gaining traction over time.
如果这个仓库对你有帮助，欢迎顺手点个 Star，也欢迎提 PR 一起维护。
把 Star History 放到仓库首页，可以让维护者和使用者都更直观看到这个项目是否在持续获得关注。
[](https://www.star-history.com/#davidtoby/agent-skills&Date)
> Like the repo? Star it, use it, and help improve it.
> 觉得有用的话，欢迎点 Star、用起来、顺手提改进。
---
## Repository structure / 仓库结构
```text
agent-skills/
├── README.md
├── CHANGELOG.md
├── CONTRIBUTING.md
├── scripts/
│ └── validate_skills_repo.py
├── skills/
│ ├── README.md
│ ├── academic-paper-to-chinese-insight-pdf/
│ ├── chinese-pdf-report/
│ ├── consulting-pdf-from-youtube/
│ └── video-bilingual-subtitle-delivery/
├── packages/
│ ├── academic-paper-to-chinese-insight-pdf.skill
│ ├── chinese-pdf-report.skill
│ ├── consulting-pdf-from-youtube.skill
│ └── video-bilingual-subtitle-delivery.skill
└── .github/workflows/
├── validate-skills-repo.yml
└── release-skill-packages.yml
```
---
## Complete active skill names / 完整技能名快照
This generated block helps humans and validators confirm repository coverage.
<details>
<summary>Active skill names</summary>
- `1password`
- `academic-paper-to-chinese-insight-pdf`
- `agent-browser`
- `agent-platform-operations`
- `agent-skill-lifecycle`
- `agent-skills-repo-publishing`
- `airtable`
- `andrej-karpathy-perspective`
- `api-learner`
- `apple-notes`
- `apple-reminders`
- `architecture-diagram`
- `arxiv`
- `ascii-art`
- `ascii-video`
- `audiocraft`
- `auto-updater`
- `automation-browser-api-workflows`
- `axolotl`
- `baoyu-comic`
- `baoyu-infographic`
- `bear-notes`
- `blogwatcher`
- `blucli`
- `bluebubbles`
- `brave-search`
- `caldav-calendar`
- `camsnap`
- `canvas`
- `chinese-asr-transcript-polishing`
- `chinese-pdf-document-workflows`
- `chinese-pdf-report`
- `chinese-video-transcribe-pdf`
- `claude-code`
- `claude-design`
- `clawddocs`
- `clawhub`
- `clawk`
- `codebase-inspection`
- `codex`
- `coding-agent`
- `comfyui`
- `communications-productivity-integrations`
- `consulting-pdf-from-youtube`
- `creative-ideation`
- `debugging-hermes-tui-commands`
- `design-md`
- `discord`
- `dogfood`
- `dspy`
- `eightctl`
- `elon-musk-perspective`
- `excalidraw`
- `feishu-approval-fallback`
- `feishu-websocket-stability`
- `feynman-perspective`
- `find-skills`
- `findmy`
- `gemini`
- `gh-issues`
- `gif-search`
- `gifgrep`
- `github`
- `github-auth`
- `github-auth-recovery`
- `github-code-review`
- `github-issues`
- `github-operations`
- `github-pr-workflow`
- `github-repo-management`
- `gmail`
- `godmode`
- `gog`
- `google-workspace`
- `goplaces`
- `healthcheck`
- `heartmula`
- `hermes-agent`
- `hermes-agent-skill-authoring`
- `himalaya`
- `huashu-design`
- `huashu-nuwa`
- `huggingface-hub`
- `humanizer`
- `ilya-sutskever-perspective`
- `imessage`
- `imsg`
- `jupyter-live-kernel`
- `kanban-orchestrator`
- `kanban-worker`
- `linear`
- `llama-cpp`
- `llm-wiki`
- `lm-evaluation-harness`
- `local-media-smart-device-workflows`
- `local-places`
- `manim-video`
- `maps`
- `markdown-converter`
- `mcporter`
- `minecraft-modpack-server`
- `model-cli-and-review-workflows`
- `model-usage`
- `mrbeast-perspective`
- `multi-search-engine`
- `munger-perspective`
- `nano-banana-pro`
- `nano-pdf`
- `native-mcp`
- `naval-perspective`
- `node-connect`
- `node-inspect-debugger`
- `notion`
- `obliteratus`
- `obsidian`
- `ocr-and-documents`
- `ontology`
- `openai-image-gen`
- `openai-whisper`
- `openai-whisper-api`
- `openclaw-backup`
- `openclaw-browser-auto`
- `openclaw-cheatsheet`
- `openclaw-cost-auditor`
- `openclaw-feishu-bot-troubleshooting`
- `openclaw-memory-docs`
- `openclaw-shield`
- `opencode`
- `openhue`
- `oracle`
- `ordercli`
- `outlines`
- `output-workspace-cleanup`
- `p5js`
- `paul-graham-perspective`
- `peekaboo`
- `pixel-art`
- `plan`
- `pokemon-player`
- `polymarket`
- `popular-web-designs`
- `powerpoint`
- `precise-bilingual-subtitle`
- `pretext`
- `proactive-agent`
- `python-debugpy`
- `qmd`
- `requesting-code-review`
- `research-paper-writing`
- `sag`
- `segment-anything`
- `self-improving-agent`
- `session-logs`
- `sherpa-onnx-tts`
- `sketch`
- `skill-creator`
- `skill-pdf-content-extractor`
- `slack`
- `songsee`
- `songwriting-and-ai-music`
- `sonoscli`
- `spike`
- `spotify`
- `spotify-player`
- `steve-jobs-perspective`
- `steve-jobs-skill`
- `subagent-driven-development`
- `summarize`
- `sun-yuchen-perspective`
- `sync-shared-skills-repo`
- `systematic-debugging`
- `taleb-perspective`
- `taskflow`
- `taskflow-inbox-triage`
- `tavily-search`
- `test-driven-development`
- `things-mac`
- `tmux`
- `touchdesigner-mcp`
- `trello`
- `trl-fine-tuning`
- `trump-perspective`
- `unsloth`
- `video-bilingual-subtitle-delivery`
- `video-frames`
- `video-transcription-subtitle-workflows`
- `vllm`
- `voice-call`
- `wacli`
- `weather`
- `web-research-monitoring-workflows`
- `webhook-subscriptions`
- `weights-and-biases`
- `writing-plans`
- `x-mastery-mentor`
- `xurl`
- `youtube-bilingual-subtitle-delivery`
- `youtube-content`
- `youtube-watcher`
- `yuanbao`
- `zhang-yiming-perspective`
- `zhangxuefeng-perspective`
</details>
---
## Design principles / 设计原则
1. **Real tasks first / 先有真实任务，再有 Skill**
Skills should come from real delivery work, not hypothetical demos.
2. **Source over magic / 用脚本、文档、流程替代玄学提示词**
Prefer auditable scripts, references, and workflows over opaque prompt tricks.
3. **Progressive disclosure / `SKILL.md` 要轻，细节放 `references/`，稳定逻辑放 `scripts/`**
Keep the entry doc readable, and move heavier detail where it belongs.
4. **Failure notes matter / 失败经验同样重要**
A strong skill records what failed, not only what worked.
5. **Deliverables matter / 交付物和验收标准要写清楚**
A reusable skill should make outputs and verification explicit.
---
## Contributing / 如何贡献
### English
A good contribution usually has:
- a real task behind it
- a clear trigger condition
- runnable scripts when scripts are included
- references that help another agent succeed faster
- explicit output naming and quality checks when output quality matters
### 中文
一个好的贡献通常具备：
- 来自真实任务，而不是想象中的 demo
- 触发条件清楚
- 如果带脚本，脚本是真的能跑
- `references/` 里的内容确实能帮另一个 Agent 少踩坑
- 如果交付质量重要，要写清楚输出命名与验收方式
Read the full guide: [`CONTRIBUTING.md`](./CONTRIBUTING.md)
---
## License / 许可证
MIT
- `agent-skills-repo-publishing` — Publish or update skills in Toby's `davidtoby/agent-skills` repository, rebuild packages, validate, sync local Hermes copies, and push via SSH.