Guide · Agent skill
How to use the Paper Illustration via Codex Image2 agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The paper-illustration-image2 skill is an experimental agent skill that helps Claude Code draft scientific paper figures, such as method or pipeline diagrams. Claude plans the layout and wording, reviews the draft against academic style rules, and asks a local Codex app-server bridge to render the image. This guide explains what the skill does, how to install it, how to trigger it and what it needs before you start.
Paper Illustration via Codex Image2wanshuiyin/Auto-claude-code-research-in-sleep · MIT Download the skill Quick start
- 1Download the skill. Download the ZIP for skills/paper-illustration-image2 from the directory page and unzip it.
- 2Place the folder. Move the paper-illustration-image2 folder into ~/.claude/skills/ or into your project's .claude/skills/ folder.
- 3Set up the bridge. Make sure a local Codex app-server and the codex-image2 MCP bridge are running, and that Python 3 is available.
- 4Write a clear request. Describe the components, labels and data flow of the figure, or point to a method file.
- 5Let the loop run. Claude plans, reviews and renders the figure, scoring each draft and iterating for up to 5 rounds.
- 6Check the output. Open the accepted image in figures/ai_generated/ and verify every label and arrow manually.
What is an agent skill?
An agent skill is a folder containing a SKILL.md file and optional scripts that Claude Code can load when a request matches the skill's description. Instead of you writing a long prompt each time, the skill gives the agent a fixed workflow to follow.
This skill is distributed as a ZIP from the directory page and is identified by the name paper-illustration-image2.
What the skill makes the agent do
The skill runs a multi-stage loop. Claude first parses your request and writes a paper-ready figure brief. Claude or a Codex text critic then refines the layout, and Claude checks the prompt against palette, arrow and label standards. The prompt is sent to a local Codex app-server MCP bridge (codex-image2) that uses Codex native image generation.
Claude then reviews the rendered image strictly and gives it a score from 1 to 10. A score of 9 or higher is accepted. Otherwise the agent writes specific feedback and loops back, for up to 5 rounds. Accepted images are saved under figures/ai_generated/ in your project.
- Planner: Claude extracts components, labels and data flow.
- Layout and style review: refines positioning, spacing and grouping.
- Rendering: native image generation through the codex-image2 bridge only; shell or Python fallbacks are rejected.
- Review: strict visual scoring with a target of 9 out of 10.
How to install the skill in Claude Code
Download the ZIP for skills/paper-illustration-image2 from this page and unzip it. For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.
If you use the Claude apps instead of Claude Code, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/paper-illustration-image2/
- Project install: .claude/skills/paper-illustration-image2/ inside your project
What you need before running it
The skill cannot render images without a working bridge. You need a local Codex app-server and the codex-image2 MCP bridge. Python 3 is needed to run the helper script scripts/paper_illustration_image2.py.
Codex MCP (mcp__codex) is used for an optional text-only critique. Reviewer MCP backends such as claude-review, grok or antigravity are optional and depend on your configuration.
- Claude Code with the skill installed
- Local Codex app-server and the codex-image2 MCP bridge
- Python 3
- Optional: Codex MCP for text critique
- Optional: reviewer MCP backends such as claude-review, grok or antigravity
How to trigger the skill with natural requests
Describe the figure you want in plain language and name the kind of diagram. The skill accepts a description or a path to a method file as its argument, so you can point it at a text file that describes your pipeline.
Be specific about components, labels and data flow. Text in the figure defaults to English unless you ask for another language.
- Draw a method overview figure for my pipeline described in docs/method.md.
- Make a pipeline diagram with an encoder, a fusion module and a classifier, labels in English.
- Create a draft architecture figure for the method section of this paper and save it to figures/ai_generated/.
Safety notes: scripts and network use
This skill runs scripts and uses the network. It is allowed to run shell commands and calls the helper script, and it uses MCP tools that send prompts and drafts to the Codex and reviewer services.
Do not use it with sensitive or unpublished methods unless you are comfortable with those prompts and drafts passing through these tools. Review the SKILL.md and scripts before installing.
- The skill runs scripts (scripts/paper_illustration_image2.py) and shell commands.
- Prompts and draft images pass through the Codex and any configured reviewer tools.
- Generated images can contain wrong labels or arrows; check every figure manually.
Tips for better figures
Give a precise component list and say which arrows show data flow. Clear inputs produce clearer first drafts and fewer review rounds.
Treat the output as a draft. Redraw or polish it by hand for a final submission, especially where exact labels or data matter.
- Name every block and arrow in your brief.
- Specify the palette and label style you want.
- Use the output as a starting point for hand redrawing.
What people use it for
- Drafting a method or pipeline diagram for a paper before polishing it by hand.
- Comparing a GPT-image-style renderer against the original paper-illustration skill.
- Producing a rough architecture figure for a student report or thesis draft.
Frequently asked questions
Is the paper-illustration-image2 skill free to use?
The skill is released under the MIT licence, as stated on this page. Running it requires a local Codex app-server and codex-image2 bridge, so check the costs and terms of those tools separately.
What do I need to install paper-illustration-image2?
You need Claude Code with the skill installed, a local Codex app-server with the codex-image2 MCP bridge, and Python 3 to run the helper script. Optional reviewer backends such as claude-review, grok or antigravity can be added depending on configuration.
Where do I put the skill folder in Claude Code?
Place the paper-illustration-image2 folder in ~/.claude/skills/ for a personal install, or in a project's .claude/skills/ folder for one project. In the Claude apps, upload it under Settings → Capabilities → Skills.
How is this different from the paper-illustration skill?
This is a separate experimental alternative to paper-illustration. It is intended for Claude Code users who want a GPT-image-style renderer, and it does not modify the original skill.
Can I use this skill without a Codex setup?
No. The skill cannot render images without the local Codex app-server bridge, so you need that setup before the image generation step will work.
Why does my generated figure have wrong labels or arrows?
Generated images can contain errors in labels and arrows. Check every figure manually and redraw or polish it before using it in a camera-ready paper.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you want a quick draft of a method or pipeline diagram and already run a local Codex app-server bridge. Pick something else if you need an exact camera-ready figure, if you cannot run the bridge, or if your method is unpublished and must not pass through external services.
Download the skill