More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    note-fig — article figures, checked for phone readability

    1

    Generates article figures from a JSON spec and fails the ones that would be too small to read on a phone.

    Free

    2 uses1.0 (1 review)
    note-fig — article figures, checked for phone readability

    note-fig — article figures, checked for phone readability

    Example session with this skill installed

    I just finished a draft about why our onboarding emails get ignored. It is about
    1,800 words, all prose, no images. Add figures where they would actually help,
    and make sure they are readable on a phone.

    [draft pasted below]

    • Read your context and instructions
    • Compiled the note-fig

    Summary

    Read the draft and found four places where a figure replaces a paragraph the
    reader would otherwise have to hold in their head. Below is the plan, then the
    first figure built and verified.

    Where the figures go

    #After the line…ShapeWhy this shape
    1"…so the email arrives at the worst possible moment."timelineThe argument is about ordering, not quantity
    2"…three things we changed."stepsSequence the reader is meant to follow
    3"Open rate went from 11% to 34%."barTwo numbers, one comparison
    4"The point is not the subject line."pointsTakeaways at the close

    Say the word and I will build 2–4. Here is #3, done.

    The spec

    {"type":"bar","title":"Open rate, before and after",
     "items":[
       {"label":"Before","value":11,"unit":"%"},
       {"label":"After","value":34,"unit":"%"}
     ],
     "key":1,
     "note":"Measured over 6 sends, Aug–Sep"}
    

    The check

    ✓ figs/03_open_rate.png
        2480×1432px / PC 620×358pt / phone 343×198pt
    
    1 figure / error 0 / warn 0
    

    Both labels sit above the minimum type size at phone width, the title fits on
    one line, and the accent colour lands on one bar only — the "After" one, which
    is the number the sentence is about.

    Next steps

    • Figures 1, 2 and 4 are specced but not built. Say which you want.
    • The 34% is taken from your draft. If the real figure moved since you wrote it,
      change the one number in the JSON and re-run; nothing else needs touching.
    • _preview.html shows both figures at true phone and desktop size. Open it
      before you publish — it is faster than guessing.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    The problem

    You write an article, then spend an hour in a design tool making one diagram. Or you ask an image model for a figure and get text that is subtly wrong and never reproducible.

    And then there is the part nobody checks: most readers are on a phone. A figure composed on a 1240px canvas is displayed at 343pt on a phone. Type that looked comfortable while you were making it arrives at around 8px. It is there, but nobody reads it.

    What this does

    note-fig turns a small JSON file into a finished PNG, and refuses to pass a figure whose text would be unreadable at phone size.

    No image generation. No design tool. The figure is composed in HTML with your own tokens and screenshotted by headless Chrome, so the same input always produces the same output, and the characters are never garbled.

    The 13 shapes

    Pick by what the sentence is doing, not by what looks nice.

    You are sayingType
    These options differ on these pointstable
    A versus B, before and aftercompare
    This, then this, then thisflow
    Do these in ordersteps
    This happened over timetimeline
    These sit on top of each otherlayers
    This is bigger than thatbar
    This changed over timeline
    This is the breakdownshare
    This number mattersstat
    These are the takeawayspoints
    Here is the actual screenwindow
    Here is how the conversation wentchat

    What it checks, every time

    • Minimum type size. Anything that would fall below ~12px on a phone is an error, not a warning.
    • 28 characters per line. Past that, the line is too dense at phone width.
    • Wrapped labels and orphans. A heading that breaks onto a second line, or a last line holding one character, is flagged.
    • Overflow and clipping. Text pushed outside its box fails the run.
    • Three columns maximum in a table, because a fourth cannot hold a word at 343pt.

    A run that produces an error still writes the PNG, but exits non-zero — so you cannot ship a broken figure without noticing.

    Bring your own look

    Colours, type scale and spacing live in one tokens.css. Point the renderer at your own brand folder and every shape follows it. The included set is the one this tool was built for; replacing it does not require touching the engine.

    What you need

    Python 3.9+ and Google Chrome (Chromium or Edge also work). Nothing from pip.

    Honest limits

    • The measurements behind the thresholds were taken on note.com (343pt on a phone, 620pt on desktop). Other platforms are close but not identical.
    • The documentation (PLAYBOOK.md, the catalog) is written in Japanese. The skill itself works in any language, and an agent reads the Japanese fine, but a human skimming the docs will want a translator.
    • It cannot enlarge text inside a screenshot you supply. Crop first.

    MIT licensed. Source: https://github.com/bigtree1984/note_fig

    Output is a 2480px-wide PNG (2× of a 1240px canvas), so it stays sharp on
    note.com and similar platforms that serve around 1200px.

    Height follows the content. Figures are trimmed to the drawing, with no padding
    you did not ask for.

    The engine is ~360 lines of plain JavaScript with no build step and no
    dependencies. Adding a 14th shape is one function, one CSS block and one row in
    the catalog.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    1.0
    1 review
    5
    0
    4
    0
    3
    0
    2
    0
    1
    1

    This skill has been used 2 times.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Environment Variables
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Note fig solves the common problem of unreadable graphics on mobile devices by using a deterministic engine that enforces specific font sizes and layout densities optimized for the note.com interface.

    This skill is designed for use with AI agents that have terminal or code execution capabilities, as it requires a Python environment to run the rendering script.

    You will receive the core Python rendering engine, a set of optimized JSON schema templates for different visual types, and documentation for the mobile-responsive preview system.

    Yes, you must have Python 3.9 or higher and a Chrome-based browser (like Chrome, Edge, or Chromium) installed on your local machine for the headless rendering to function.

    The skill includes a 'Mobile Validation' step that automatically checks your graphics against a 28-character limit and font rules, generating an HTML preview so you can verify the mobile look before posting.

    While optimized for the specific aspect ratios and legibility standards of note.com, the generated PNGs can be used elsewhere, though they are not intended for high-resolution print or presentation slides.

    No. The figure is composed in HTML from your JSON and screenshotted. The same input always produces the same output, and no text is ever garbled.

    Yes. One tokens.css holds the palette, type scale and spacing. Point the renderer at your own brand folder with --brand and every shape follows it.

    The thresholds were measured there, but nothing in the engine is specific to it. Any platform that serves images around 1200px wide and displays them around 343pt on a phone will behave the same.

    The run tells you, with the exact phrase and what to do — shorten the label, or move the detail into the note line. It will not silently shrink the type.

    It is the only dependency, and most machines already have it. It renders the HTML and takes the screenshot; nothing is uploaded anywhere.

    New in Workflow Automation