Jeff Kazzee

← All projects

Context Engine

prototype ·

Stack

  • TypeScript
  • Claude Code
  • Codex

Tags

agents · context · experimental

AI-generated concept illustration: an agent reads working context, edits it, and passes updated context to the next model request. Tool names are illustrative, not an implemented API. Context delivery remains experimental and under testing.
AI-generated concept illustration: an agent reads working context, edits it, and passes updated context to the next model request. Tool names are illustrative, not an implemented API. Context delivery remains experimental and under testing.

Context Engine explores file-based working context for coding agents. It is based on the Context Language Models paper, which treats context as a file the model can update.

The work

There are separate Codex and Claude Code repositories. The proposed plugins share a context core while adapting setup and hooks to each runtime. They are experimental and opt-in, starting disabled in each project.

I am developing this with AI coding agents, with attention to scoped activation, recovery, and clear setup and rollback instructions.

What is established

As of October 8, 2026, both public default branches are bootstrap pages. The implementation and validation records are in the open release PRs for Codex and Claude Code, pending merge and release approval.

Those records cover automated and offline checks. Live native-agent context delivery and long-session acceptance have not yet been established. This project does not claim the paper’s benchmark results or a production-ready integration.