<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Subomi Olagoke</title>
    <description>The latest articles on DEV Community by Subomi Olagoke (@subomi_olagoke_de43e35300).</description>
    <link>https://dev.to/subomi_olagoke_de43e35300</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4251932%2F26754eff-b30d-4164-a44f-9a6e9d1e27c0.png</url>
      <title>DEV Community: Subomi Olagoke</title>
      <link>https://dev.to/subomi_olagoke_de43e35300</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/subomi_olagoke_de43e35300"/>
    <language>en</language>
    <item>
      <title>How I turned whiteboard marks into input Claude Code can act on</title>
      <dc:creator>Subomi Olagoke</dc:creator>
      <pubDate>Wed, 18 Nov 2026 13:28:01 +0000</pubDate>
      <link>https://dev.to/subomi_olagoke_de43e35300/how-i-turned-whiteboard-marks-into-input-claude-code-can-act-on-29jm</link>
      <guid>https://dev.to/subomi_olagoke_de43e35300/how-i-turned-whiteboard-marks-into-input-claude-code-can-act-on-29jm</guid>
      <description>&lt;p&gt;Every Claude Code session I run has the same moment. Claude lays out a plan, it's 80% right, and I spend two minutes typing "skip step 4, do 6 before 5, and don't touch the auth module." I wanted to fix that plan the way I'd fix a sketch on a whiteboard, so I built one that Claude can draw on and read back. This post walks through the pattern, which works for any MCP tool that needs a human to edit something visual.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Correcting a plan in chat is fragile
&lt;/h2&gt;

&lt;p&gt;A plan in a terminal arrives as a wall of text. Changing one step means writing a paragraph back, and the more changes you make, the more likely one gets missed or misread. "The second one, not the third" is exactly the kind of sentence an agent gets wrong.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. The pattern: a board, a skill, and hooks
&lt;/h2&gt;

&lt;p&gt;Three pieces do the work:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;An MCP server that exposes a board.&lt;/strong&gt; Claude gets tools to open the board, lay out a plan as cards, set a card's status and read what the human did.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A skill that tells Claude when to use it.&lt;/strong&gt; When a task has several parts, Claude says "let's put this on the board" instead of dumping a numbered list.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hooks that hand the marks back.&lt;/strong&gt; Between turns Claude runs a small wait command in the background, so it wakes when you mark something. Panning and zooming never wake it.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  3. What a mark becomes
&lt;/h2&gt;

&lt;p&gt;The board recognises the gesture itself, so Claude gets a precise line rather than a screenshot it has to interpret:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;h1 CROSSED OUT c5 "Pinch to zoom" =&amp;gt; drop it
h2 CIRCLED c1 "Board model" =&amp;gt; focus here
h4 QUESTION "which transport?" by c3 =&amp;gt; answer on the board next to it
Plan in current board order: c1 -&amp;gt; c2 -&amp;gt; c4 -&amp;gt; c3
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A cross-out means drop it, a circle means do it first, a drag changes the order, a question mark gets an answer written beside it, and a big X over the whole board means stop and ask. For handwriting and freehand sketches, Claude looks at a snapshot instead.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Keeping it local
&lt;/h2&gt;

&lt;p&gt;Everything runs on the Mac: a small server on 127.0.0.1 and boards stored as files in your home folder. Nothing about your repo or your plans leaves the machine, and one uninstall command removes every piece it added to Claude Code.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Where it still falls short
&lt;/h2&gt;

&lt;p&gt;Marks cover the common corrections, but not everything. Some feedback really is a sentence, and the board doesn't replace typing for that. Handwriting recognition also depends on the snapshot, so messy notes are slower to read than clean gestures.&lt;/p&gt;

&lt;p&gt;I make Chalk, the app this is from. It's at getchalk.dev if you want to try the pattern yourself.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>mcp</category>
      <category>ai</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
