<?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: chipfighter</title>
    <description>The latest articles on DEV Community by chipfighter (@chipfighter_62).</description>
    <link>https://dev.to/chipfighter_62</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%2F4202191%2F18113ab8-dc3f-42b5-983c-1be1520ecca2.jpg</url>
      <title>DEV Community: chipfighter</title>
      <link>https://dev.to/chipfighter_62</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/chipfighter_62"/>
    <language>en</language>
    <item>
      <title>Nine images, a 42 MB request, and a Codex chat that wouldn't reconnect</title>
      <dc:creator>chipfighter</dc:creator>
      <pubDate>Mon, 26 Oct 2026 17:08:23 +0000</pubDate>
      <link>https://dev.to/chipfighter_62/nine-images-a-42-mb-request-and-a-codex-chat-that-wouldnt-reconnect-3232</link>
      <guid>https://dev.to/chipfighter_62/nine-images-a-42-mb-request-and-a-codex-chat-that-wouldnt-reconnect-3232</guid>
      <description>&lt;p&gt;I maintain a small image-history plugin for Codex and Claude. I wanted a test that checked more than whether the interface looked useful: could a conversation still continue after a restart or compaction, and could the model find the right earlier image?&lt;/p&gt;

&lt;p&gt;In one Codex test, nine images were enough to produce a &lt;strong&gt;42.4 MB full-history request&lt;/strong&gt;. That request stalled in our setup. With older images replaced by retrievable placeholders, the corresponding requests were &lt;strong&gt;0.12–0.13 MB&lt;/strong&gt; and completed.&lt;/p&gt;

&lt;p&gt;The useful detail is &lt;em&gt;when&lt;/em&gt; this happened.&lt;/p&gt;

&lt;h2&gt;
  
  
  The chat worked until Codex needed the whole history
&lt;/h2&gt;

&lt;p&gt;The test used nine synthetic storyboard keyframes, roughly 3.2–3.9 MB each. Across six turns, the user introduced scenes, replaced a dark frame, and said that one version had the best face.&lt;/p&gt;

&lt;p&gt;Over an existing WebSocket, Codex sent incremental updates, so those turns went through. After a restart, it needed to send the conversation again. The full JSON body was 42.4 MB, or 31.8 MB compressed.&lt;/p&gt;

&lt;p&gt;The WebSocket message was refused within seconds. HTTP fallback attempts stopped after roughly a minute. We eventually stopped the retries.&lt;/p&gt;

&lt;p&gt;A compaction did not remove the image payloads in this test. The next question still needed a full-history request and hit the same problem.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://raw.githubusercontent.com/chipfighter/context-attachment-manager/main/eval/results/2026-10-24/request-size-en.svg" rel="noopener noreferrer"&gt;Request-size chart for the tested conversation&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What the plugin changed
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/chipfighter/context-attachment-manager" rel="noopener noreferrer"&gt;Context Attachment Manager&lt;/a&gt; leaves the original chat and image files unchanged. It rewrites the outgoing view of earlier images into placeholders with IDs. The model can retrieve an original with &lt;code&gt;cam_view_image&lt;/code&gt; when it needs visual details.&lt;/p&gt;

&lt;p&gt;New images still go out on the turn that introduces them. Users can also pin a reference to keep sending it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fairn4ryndb7pbpdks1xv.gif" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fairn4ryndb7pbpdks1xv.gif" alt="Pinning and retrieving earlier images" width="760" height="428"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Same test&lt;/th&gt;
&lt;th&gt;After restart&lt;/th&gt;
&lt;th&gt;After compaction&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Without the plugin&lt;/td&gt;
&lt;td&gt;42.4 MB; did not send&lt;/td&gt;
&lt;td&gt;42.4 MB; did not send&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;With v0.9.0&lt;/td&gt;
&lt;td&gt;0.13 MB; sent&lt;/td&gt;
&lt;td&gt;0.12 MB; sent&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;For file-finding questions, v0.9.0 selected the correct image in 6/6 cases before compaction and 6/6 after it. When asked to read a small digit in a corner, it fetched the particular original it needed.&lt;/p&gt;

&lt;h2&gt;
  
  
  The results also limit the pitch
&lt;/h2&gt;

&lt;p&gt;This was &lt;strong&gt;one scenario on Windows 11&lt;/strong&gt;, with Codex app-server 0.163.0-alpha.11 and &lt;code&gt;gpt-6-sol&lt;/code&gt; at low reasoning effort. There was one final run per setup, plus pilot runs. Claude was not evaluated in this experiment.&lt;/p&gt;

&lt;p&gt;A faster connection might handle the large HTTP request differently. We did not test that.&lt;/p&gt;

&lt;p&gt;Token costs were less straightforward than request size. Total input tokens fell 16%, but uncached input tokens rose 36%. Whether the bill falls depends on the prompt-cache discount.&lt;/p&gt;

&lt;p&gt;The older plugin version, without image feature notes, also found 6/6 files after compaction in this short task. This result does not establish that feature notes improve every long conversation.&lt;/p&gt;

&lt;p&gt;One pilot was especially instructive: the model misread the simple synthetic scenes, wrote inaccurate descriptions, then sometimes said the requested image did not exist instead of retrieving it. Those failures are included in the report.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reproduce it or try another setup
&lt;/h2&gt;

&lt;p&gt;The &lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/eval/README.md" rel="noopener noreferrer"&gt;evaluation write-up&lt;/a&gt; includes the task, raw answers, request measurements, pilot failures, and scripts to reproduce the test.&lt;/p&gt;

&lt;p&gt;The plugin is free and MIT-licensed. It supports Codex desktop and &lt;strong&gt;local Code sessions in Claude Desktop&lt;/strong&gt;. Installation instructions are &lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/INSTALL.md" rel="noopener noreferrer"&gt;here&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If you reproduce the reconnect failure, or find that it works on your setup, &lt;a href="https://github.com/chipfighter/context-attachment-manager/issues" rel="noopener noreferrer"&gt;open an issue&lt;/a&gt; with the client version and result. An independent reproduction would be useful.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>opensource</category>
      <category>testing</category>
      <category>showdev</category>
    </item>
    <item>
      <title>First PR welcome: help improve an image-history plugin for Codex and Claude</title>
      <dc:creator>chipfighter</dc:creator>
      <pubDate>Mon, 26 Oct 2026 09:13:53 +0000</pubDate>
      <link>https://dev.to/chipfighter_62/first-pr-welcome-help-improve-an-image-history-plugin-for-codex-and-claude-3n57</link>
      <guid>https://dev.to/chipfighter_62/first-pr-welcome-help-improve-an-image-history-plugin-for-codex-and-claude-3n57</guid>
      <description>&lt;p&gt;I maintain &lt;a href="https://github.com/chipfighter/context-attachment-manager" rel="noopener noreferrer"&gt;Context Attachment Manager&lt;/a&gt;, a free MIT-licensed image-history plugin for Codex desktop and local Code sessions in Claude Desktop. I'd like more people to try it and help improve it.&lt;/p&gt;

&lt;p&gt;It keeps older images out of later requests until they're needed. You can pin references, search earlier images and let the model fetch an original back by ID.&lt;/p&gt;

&lt;h2&gt;
  
  
  Want to make a small first contribution?
&lt;/h2&gt;

&lt;p&gt;There are three starter tasks with starting points and acceptance notes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;a href="https://github.com/chipfighter/context-attachment-manager/issues/11" rel="noopener noreferrer"&gt;Keep keyboard focus inside the image preview&lt;/a&gt;: a small HTML/JavaScript accessibility improvement.&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/chipfighter/context-attachment-manager/issues/12" rel="noopener noreferrer"&gt;Add an endpoint troubleshooting checklist&lt;/a&gt;: documentation work.&lt;/li&gt;
&lt;li&gt;
&lt;a href="https://github.com/chipfighter/context-attachment-manager/issues/13" rel="noopener noreferrer"&gt;Write a synthetic demo-panel walkthrough&lt;/a&gt;: try the panel and document the checks.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The panel has a synthetic demo you can run without a Codex or Claude account. Setup instructions are in &lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/CONTRIBUTING.md" rel="noopener noreferrer"&gt;CONTRIBUTING.md&lt;/a&gt;. The project uses TypeScript and Node.js, with a plain HTML/CSS/JavaScript panel.&lt;/p&gt;

&lt;p&gt;First-time contributors are welcome. Comment on a task if you want to work on it or need help finding the code. Small, focused documentation fixes and PRs are welcome too.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prefer trying it to writing code?
&lt;/h2&gt;

&lt;p&gt;Use the &lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/INSTALL.md" rel="noopener noreferrer"&gt;installation guide&lt;/a&gt;, then &lt;a href="https://github.com/chipfighter/context-attachment-manager/issues/new" rel="noopener noreferrer"&gt;open an issue&lt;/a&gt; with what you tried and what happened. An installation question, confusing wording or a feature suggestion is useful feedback; you don't need to diagnose the cause first.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/chipfighter/context-attachment-manager/contribute" rel="noopener noreferrer"&gt;Browse the starter tasks&lt;/a&gt;&lt;/p&gt;

</description>
      <category>typescript</category>
      <category>opensource</category>
      <category>beginners</category>
      <category>github</category>
    </item>
    <item>
      <title>I built an image-history manager for long Codex and Claude conversations</title>
      <dc:creator>chipfighter</dc:creator>
      <pubDate>Fri, 23 Oct 2026 09:38:10 +0000</pubDate>
      <link>https://dev.to/chipfighter_62/i-built-an-image-history-manager-for-long-codex-and-claude-conversations-4dph</link>
      <guid>https://dev.to/chipfighter_62/i-built-an-image-history-manager-for-long-codex-and-claude-conversations-4dph</guid>
      <description>&lt;p&gt;I was iterating on images with GPT in one long Codex task when the conversation started repeatedly reconnecting. In that task, requests carrying the accumulated image history had grown to about &lt;strong&gt;44 MB&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Starting a new conversation reduced the baggage, but separated me from the references, previous edits, and feedback I wanted to keep using.&lt;/p&gt;

&lt;p&gt;That led to &lt;strong&gt;Context Attachment Manager&lt;/strong&gt;, an MIT-licensed local plugin for image-heavy conversations in &lt;strong&gt;Codex desktop&lt;/strong&gt; and &lt;strong&gt;Claude Desktop's Code tab&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/chipfighter/context-attachment-manager" rel="noopener noreferrer"&gt;GitHub repository&lt;/a&gt; · &lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/INSTALL.md" rel="noopener noreferrer"&gt;Install instructions&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzfg1h2928y0riohjfdcg.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzfg1h2928y0riohjfdcg.png" alt="Context Attachment Manager: automatic image selection, searching past images, and a full-size preview" width="800" height="440"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep the conversation; retrieve images when needed
&lt;/h2&gt;

&lt;p&gt;With Auto-select enabled, images from earlier turns can be left out of later requests. When the model needs to inspect an earlier image, it can fetch that image back by its ID.&lt;/p&gt;

&lt;p&gt;For a character reference or another image that should always remain visible, pin it to send every turn. If you prefer to decide yourself, turn off Auto-select and use the panel's checkboxes.&lt;/p&gt;

&lt;p&gt;The aim is to keep working in the same conversation while reducing repeated image transfers. It does not require moving the conversation to another service.&lt;/p&gt;

&lt;h2&gt;
  
  
  Find an image by what you remember
&lt;/h2&gt;

&lt;p&gt;The panel supports searching visual descriptions and your own feedback. Instead of remembering a filename, search for something like &lt;strong&gt;"rainy night"&lt;/strong&gt; or &lt;strong&gt;"best face"&lt;/strong&gt;, then open the original.&lt;/p&gt;

&lt;p&gt;The model records short image features and user feedback as retrieval cues. Those cues can survive conversation compaction, helping it locate the corresponding image later. They are not a substitute for looking at the original when a question depends on fine visual details.&lt;/p&gt;

&lt;h2&gt;
  
  
  A measured example, with a narrow claim
&lt;/h2&gt;

&lt;p&gt;In one 15-image test in the project, the next request body dropped from &lt;strong&gt;24.9 MB to 212 KB&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;That is a result from that particular scenario. It is not a promise for every conversation, a token-cost measurement, or proof that every reconnect has the same cause. The effect depends on the images retained and the images fetched back during the turn.&lt;/p&gt;

&lt;p&gt;The panel estimates request size so the change is visible before sending.&lt;/p&gt;

&lt;h2&gt;
  
  
  How the local plugin works
&lt;/h2&gt;

&lt;p&gt;A loopback-only engine sits between the client and its configured model endpoint. It replaces unneeded historical image payloads with numbered placeholders and forwards the request. An MCP tool lets the model retrieve an original image when necessary.&lt;/p&gt;

&lt;p&gt;The panel handles search, previews, pinning, and manual selection. Original conversation records and image files remain read-only. The plugin does not send an extra copy of the conversation to a separate cloud service; normal model requests still go to the configured provider.&lt;/p&gt;

&lt;p&gt;Version &lt;strong&gt;0.9.0&lt;/strong&gt; binds each route to both the client platform and its original upstream. Compatible custom endpoints can use image management too. They must support the client's API format, image inputs, and tool calls: the plugin does not translate between Responses and Anthropic Messages.&lt;/p&gt;

&lt;h2&gt;
  
  
  Current scope
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Codex desktop and local sessions in Claude Desktop's &lt;strong&gt;Code&lt;/strong&gt; tab. Ordinary Claude Chat is outside the integration.&lt;/li&gt;
&lt;li&gt;Windows, macOS, and Linux. Windows is the primary manual-testing platform; three-platform CI covers unit tests and installation, startup, upgrade, and uninstall checks.&lt;/li&gt;
&lt;li&gt;English and Simplified Chinese interfaces.&lt;/li&gt;
&lt;li&gt;MIT licensed, written in TypeScript/Node.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Live third-party gateway model calls have not been tested by the maintainer. The plugin also does not shrink the original history files or repair unrelated networking, local rendering, or model-capacity problems.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it, report a problem, or contribute
&lt;/h2&gt;

&lt;p&gt;You can ask Codex, or Claude in Claude Desktop's Code tab, to install the plugin by following &lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/INSTALL.md" rel="noopener noreferrer"&gt;INSTALL.md&lt;/a&gt;. After installation, restart Codex or start a new Claude Code session.&lt;/p&gt;

&lt;p&gt;In Codex, open &lt;strong&gt;Context Assets&lt;/strong&gt; in the task's side panel. In Claude Desktop Code, the panel opens once the conversation has images, and &lt;code&gt;/cam&lt;/code&gt; opens it again.&lt;/p&gt;

&lt;p&gt;I'm looking for feedback from people who repeatedly edit images, compare visual references, or accumulate screenshots during a long task. Reports from real desktop workflows on macOS/Linux and compatible custom endpoints would be particularly useful.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;a href="https://github.com/chipfighter/context-attachment-manager" rel="noopener noreferrer"&gt;Source and screenshots&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/chipfighter/context-attachment-manager/issues" rel="noopener noreferrer"&gt;Report a bug or suggest an improvement&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a href="https://github.com/chipfighter/context-attachment-manager/blob/main/CONTRIBUTING.md" rel="noopener noreferrer"&gt;Contribution guide&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Please include the client version, platform, and reproducible steps when reporting a problem. PRs are welcome.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Disclosure: this article was drafted and published with an AI agent, using the project's existing documentation, implementation, and recorded test results.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>opensource</category>
      <category>ai</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
