Devotd Github
NewMultimodal

chipfighter/context-attachment-manager — Renaming and README Rewrite

The latest commit in chipfighter/context-attachment-manager is a documentation-only change that renames FrameKeeper to Context Attachment Manager and rewrites the README in plainer language.

Repositorychipfighter/context-attachment-manager ↗

The latest commit in chipfighter/context-attachment-manager is a documentation-only change that renames FrameKeeper to Context Attachment Manager and rewrites the README in plainer language.

GitHub description: Codex and Claude Code plugin (works in Claude Desktop): choose which past images go to the model with your next message. Unchecked images become placeholders, the conversation goes on, and requests stay small.

stars
★ 235
Forks
23
Language
TypeScript
License
MIT
Latest release
v0.7.1
Read original source ↗View on GitHub ↗

chipfighter/context-attachment-manager is a local plugin that stops images from previous conversations in Codex and Claude Code from being attached over and over in subsequent messages. The November 6, 2026 commit 2fd3b5d is a merge of the 'docs/plain-tool-readme' branch; it touches no feature code at all, only five documentation files (ADAPTERS.md, README, README_CN, README_JA, docs/spec.md). Among the 90 added lines and 93 deleted lines, the project name changes from FrameKeeper to Context Attachment Manager, and the README description is rewritten in plainer language.

The key point is not to remove images but to set them aside. Each image is given a number like IMG-001, only a short placeholder is sent for earlier images, and the model can fetch the original again by number when it needs it. The README states that in Codex a conversation with nine keyframes grew to as much as 42.4 MB and could no longer be sent after a restart or a conversation summary, and that with the plugin the same request was 0.12 MB. The panel introduction GIF explains that on a task with fifteen keyframes the next request shrank from 7.5 MB to 1.7 MB.

The structure has three layers. A platform-agnostic core decides what to strip out, assigns numbers to images and keeps them, and a request format converter injects that decision into four formats: OpenAI Responses and Chat Completions, Anthropic Messages, and Gemini generateContent. The panel for selecting, pinning and searching images exists only in Codex and Claude Desktop. The license is MIT and the target platforms are Windows, macOS and Linux, as the README badges indicate.

That said, this commit touches documentation only. ADAPTERS.md states that the adapters for WorkBuddy, OpenCode, Gemini CLI, pi, DeepSeek Harness, Qwen Code, Cline, Kilo Code, goose and Cherry Studio were written by reading each agent's code and documentation but have not yet been live-tested. In other words, only Codex and Claude Code are verified paths, so people using other agents will find it hard to trust the behavior right away even though the installation instructions have been cleaned up. For those using Codex or Claude Code, this change is worth following as it makes clear what has been verified.