REA
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
TikZ for paper diagrams, without hand-writing TikZ — copyable icons, editable templates, and a Claude Code skill
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/skills/opentikz-opentikz)在同一工作中把 AI 连到服务的 MCP
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
Codebase intelligence for TypeScript and JavaScript, exposed as structured MCP analysis tools.
MCP server for terminal commands, file operations, and process management
同类工作,排序与 Skill 列表相同
| 本项 opentikz | rea | andrej-karpathy-skills | graphify | MoneyPrinterTurbo | |
|---|---|---|---|---|---|
| 作者 | opentikz | morluto | multica-ai | Graphify-Labs | harry0703 |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 245 | 468,733 | 234,332 | 150,287 | 144,566 |
| 最近代码更新 | 2026年7月12日 | 2026年11月19日 | 2026年4月20日 | 2026年11月18日 | 2026年11月18日 |
| 许可证 | – | MIT | – | Apache-2.0 | MIT |
opentikz 提供的 1 个 Skill(指令集),AI 读取后照做 · 不是 MCP server
MCP = 手
把 AI 连到外部服务或数据,例如读取 Notion 或开发票。需要有程序或服务器在运行。
Skill = 手册
教 AI 怎么做某项工作的文本文件,不连接任何服务。
有些会重叠:少数 Skill 带有调用工具的脚本,少数 MCP 只提供知识。请从你想完成的工作出发。
Claude CodeCodexCursorGemini CLI
依据开发者 README 和 repo 中的文件夹
GitHub account opentikz (organization) is only 5 months old: take extra care
★ 245 stars, few users so far · code last changed 2026年7月12日
No, plain text files the AI reads and follows
No scripts, so it does not go online by itself
No hooks or writes to machine settings found
Checked automatically from the repo files at commit 359befb (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
第一步:你用哪个 AI?
选择你用的应用,本页只显示该应用的安装方法
This skill has no license, so we do not hand out a ZIP from our site. Download it from GitHub and arrange the folder yourself, or use "Let the AI install it" above, which is easier
Download from GitHub (whole repo) ↗
Unzip it, then zip only the skills/using-opentikz folder again. The ZIP must have a using-opentikz folder with SKILL.md at the top level, no loose files
+Create skillUpload a skillpick using-opentikz.zip
SettingsCapabilitiesCode execution and file creation
Turn on Code execution and file creation first (newer menus: Create and edit files) · personal plans: Settings → Capabilities · Team/Enterprise: an admin turns it on under Organization settings → Plugins & skills · every plan, including Free
Upload once; it works in claude.ai and Claude Desktop, and in recent Claude Code signed in with a Claude account (not with an API key)
Cowork and cloud sessions use the skills turned on in your claude.ai account, not ~/.claude/skills on your computer (Claude Desktop scheduled tasks run locally and still load it)
Checked against official docs 2026年11月12日 · Source: support.claude.com, support.claude.com, code.claude.com
复制这句话,粘贴到 Claude Code 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
重新打开 Claude Code,输入 /using-opentikz,或直接提出相关任务,Claude 会自动选用
Checked against official docs 2026年11月12日 · Source: code.claude.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
已隐藏作者的命令 · 它不是从 GitHub 上的这个仓库安装 · 命令未通过检查,已隐藏 · 请改从作者的 GitHub 页面安装
摘自开发者 README,会下载并运行开发者的代码
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
ChatGPT 桌面应用内置 Codex,Codex 从 ~/.agents/skills 文件夹读取 Skill
在 ChatGPT 桌面应用中打开 Codex,把这句话粘贴到对话中
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
把 using-opentikz 文件夹放到 ~/.agents/skills/using-opentikz
~ 是你的主文件夹($HOME)
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
不能使用你自己安装的 Skill,只能使用目录中插件自带的 Skill
想在 ChatGPT 中使用这个 Skill,请按上面的方法使用 ChatGPT 桌面应用
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
复制这句话,粘贴到 Codex 的对话中 · 适用于 ChatGPT 桌面应用中的 Codex、Codex CLI 和 IDE 扩展
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex · 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
复制这句话,粘贴到 Cursor 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
在 Cursor 的 Agent 聊天中输入 / 并选择 using-opentikz
Checked against official docs 2026年11月12日 · Source: cursor.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
适用于其他能读取 Skill 的应用。这句话列出了 Claude Code、Codex 和 Cursor 的路径,AI 会按你的应用选择 · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs 2026年11月12日 · Source: code.claude.com, learn.chatgpt.com, cursor.com
其他支持 Agent Skills 的应用(如 Gemini CLI、GitHub Copilot)使用不同文件夹,请查看该应用的指南,再把含 SKILL.md 的文件夹复制过去。
已于 2026年11月11日 对照 Claude Code、Codex 和 Cursor 官方指南核实 skill 文件夹
以下文字来自开发者的 repo,不是 PopNx MCP 的指令
显示前 8,000 个字符 · 在 GitHub 打开完整文件
---
name: using-opentikz
description: Use when a user wants a TikZ figure for a paper — an icon, an editable architecture/pipeline/flow template, or an example — and wants you to find, edit, and verify it. The single repo-wide skill for working with the OpenTikZ library.
---
# Using OpenTikZ
OpenTikZ is a library of copyable TikZ **icons**, editable **templates** (neural
nets, encoder-decoder, training pipelines, system block diagrams, flowcharts),
and full **examples**. This skill is how you go from a user's request to a
finished, compiling `.tex` figure — reliably, and without guessing.
There is **one** skill (this file). Per-template knowledge lives in each
template's `edit_contract` (inside its `meta.json`), which you read at edit time.
Do not look for per-template `skill.md` files; they no longer exist.
## 0. First: which mode are you in, and where is the library?
**Pick the mode:**
- **Mode A — produce a figure for the user's own project** (the default). The
OpenTikZ library is read-only and lives elsewhere; the user is working in their
own paper/LaTeX project. You will *copy* a figure out of the library and edit the
copy in their project. Never modify the library.
- **Mode B — contribute back to the OpenTikZ repo itself.** The user is editing the
library: their working directory *is* the repo. Edit in place and run the repo
tooling (see §6).
**Locate the library root (`OTROOT`) for Mode A:**
1. If the environment variable `${CLAUDE_PLUGIN_ROOT}` is set, you were installed as
a Claude Code plugin: `OTROOT = ${CLAUDE_PLUGIN_ROOT}`.
2. Otherwise (cloned repo / another agent), `OTROOT` is the OpenTikZ repo root — the
directory two levels above this `SKILL.md` (`skills/using-opentikz/SKILL.md` →
`../../`).
3. If you can only read the repo remotely (e.g. over GitHub, no local clone), treat
the GitHub repo as `OTROOT` and fetch raw file contents on demand.
Confirm `OTROOT` once (e.g. list `OTROOT`/catalog.json) before relying on it. The
**output target is always the user's current working directory**, never `OTROOT`.
## 1. Communicate precisely (do this first, throughout)
Your job is to produce the figure the user actually wants, not a plausible guess.
Classify every ambiguity before acting:
**Material ambiguity → ask, or offer two concrete alternatives.**
Trigger when the answer changes the figure's *structure or identity*, is *hard to
reverse*, or you would otherwise be guessing between genuinely different intents.
Ask one crisp question with concrete options; never a vague "what do you want?".
- "draw a transformer" → encoder-only, decoder-only, or encoder-decoder? (ask)
- "add attention" → self-attention or cross-attention, and where? (ask)
- "highlight the new part" → which part is new? (ask)
- Use **offer-two-alternatives** when seeing beats describing — an aesthetic or
layout fork (two color schemes, two layouts). Produce both and let the user
pick. Don't use it for continuous quantities (widths, counts).
**Safe-default ambiguity → proceed, state the assumption in one line, keep it
cheap to reverse.** When there is a sensible default and changing it is a
one-line tweak, don't interrupt — just decide and tell the user what you assumed.
- palette: default light Okabe-Ito unless "dark" is mentioned.
- width: default standalone (no `\resizebox`) unless a venue/column width is named.
- spacing / size / an unspecified-but-conventional count: use the template default
(e.g. a 4-layer net `4,6,6,2`).
**Anti-interrogation guardrail.** Ask at most ~2–3 questions up front, batched;
resolve everything else with assume-and-state. Every delivery ends with a
one-line summary: *"Assumptions: light palette, single-column width, 4 layers —
say the word to change any."*
## 2. Workflow
1. **Understand & classify.** Restate the goal in one sentence; classify
ambiguity per §1; ask the (few) material questions, batched.
2. **Discover.** Read `catalog.json` at `${OTROOT}` (see §0). Match the request to
icons / templates / examples by `name`, `tags`, `domain`. If several fit,
present the candidates (with their `id`s) and let the user choose; if one
clearly fits, name it and proceed.
3. **Select & confirm.** Confirm the chosen base plus the material parameters
(e.g. layer counts, labels, which part is highlighted).
4. **Copy, then edit (Mode A) / edit in place (Mode B).**
- **Mode A:** copy the chosen item's `.tex` from `OTROOT` into the user's project
(a sensible path like `figures/<name>.tex`); tell the user where you put it.
Edit *that copy*, never the file under `OTROOT`. If the figure is already in
the user's project, edit it there. For a **template**, read its `edit_contract`
from `OTROOT`'s `catalog.json` (or the item's `meta.json`) and: edit only the
contract's `parameters`; follow its `operations` as the recipe; keep every
`invariant`; preserve the `node_naming` scheme; colours via palette names only.
For an **icon/example** (no contract), edit under the same hard rules.
- **Mode B:** edit the item in place inside the repo, under the same rules.
5. **Verify by compiling — in the user's project (Mode A).** Run the user's LaTeX on
the copied/edited file (`latexmk -pdf <file>.tex`, or `pdflatex`). Fix failures
before returning — never hand back a figure you didn't compile. (Regenerating
`.svg` previews and `catalog.json` is a *Mode B / contributor* task — skip it
when producing a figure for a user.)
6. **Deliver.** Return the edited `.tex` and the one-line assumptions summary.
## 3. Hard rules (never violate)
- The `.tex` must stay **standalone-compilable** (`\documentclass{standalone}`).
- **Colors only via the five palette names** (`otblue otorange otteal otpurple
otgray`); tints/shades like `otblue!15` are fine. Never inline a hex value or a
stock xcolor name (`blue`, `red`). See `reference/color-palettes/`.
- **Preserve node names / the `node_naming` scheme** — they are the contract that
lets edits target the right parts. Give any new node a clear semantic name.
- **Keep figures parametric** — drive counts/spacing/labels through the `\def`
block at the top; don't hard-code what a parameter already controls.
- **Separate icon from label — never overlay a `\pic` on a node's centred text.**
For any box that carries *both* a glyph and a label, make the box node empty
(`{}`), then place the icon with a standalone `\pic` in the box's **upper half**
and the label as a *separate* text node in the **lower half** (or icon-left /
text-right for a wide chip). Size the box tall enough for both bands. Dropping a
`\pic` on top of a node's own `{text}`, or faking the offset with `\hspace`,
collides every time — this is the single most common rework.
- **Budget padding — size a box to its content *plus* margin, never flush to it.**
Set `minimum width`/`minimum height` to the widest/tallest content **plus a
per-side padding budget of ≥0.25 cm**. For a multi-column in-box layout (e.g.
two icon+label stacks), also give an explicit inter-column gap (≥0.3 cm) and keep
each column's widest line clear of the frame. If any text touches the border,
widen the box or push the columns inward — do **not** shrink the text.
## 4. Common cross-template operations
These work the same across templates; the per-template `edit_contract` lists the
specific parameter/style names to touch.
**Recolor.** Change the palette color *name* in the relevant style or node, never
a hex. "Make it teal" = `otblue` → `otteal` in that style's `fill`/`draw`.
**Switch to the dark palette.** Replace the light `\definecolor` block with the
dark block from `reference/color-palettes/color-palettes.md` — the names are
identical, so the body is unchanged. The dark palette is for **dark
backgrounds**: also set `\pagecolor{otpaper}`
(`\definecolor{otpaper}{HTML}{1E1E1E}`), or the tints render washed-out grey on a
white page.
**Adapt to a venue / column width.** Wrap the whol提问、分享技巧或报告无法使用。按“有帮助”数量排序。