Skip to content

fast-dev-compaction

Codex port of fast-jev-compaction that re-injects Jev-selected verbatim history after compaction; the author advises against using it.

PluginprototypeMIT

Overview

This repository (published as fast-jev-compaction for Codex) wraps Codex session compaction with Jev decisions. Before Codex compacts, every tool call and result is scored in fast requests; after compaction, the verbatim history Jev kept is re-injected as context. User and assistant text stays verbatim and in order. It is both an npm library and a Codex plugin using lifecycle hooks, porting tamaratran's original Claude Code adapter.

The README opens with a clear warning: the author does not recommend using it and calls it an experimental proof of concept. It cites a critique by Theo (t3.gg) arguing that pruning tool calls on a probability threshold misunderstands compaction: it discards reasoning traces frontier models do not expose, ignores that models are trained on their own compaction flows, and makes history edits trigger expensive cache rewrites. It is best read as a write-up of an idea.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this tool call or result still needed?(inferred)yes · no

How it works

  1. Before compaction, score each tool call and result with Jev
  2. Let Codex compact natively
  3. Re-inject the Jev-kept verbatim history as context

Strengths

  • Illustrates a verbatim-preservation approach
  • Usable as an npm library

Caveats

  • Author explicitly does not recommend using it
  • Critiqued for breaking caches and ignoring hidden reasoning
  • Proof of concept

Getting started

Not recommended for real work; see the README for installing the Codex plugin if experimenting.

At a glance

Type
Plugin
Status
prototype
Language
TypeScript
Runtime
Codex plugin + npm library
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Context management: Codex port of the Jev-guided compaction idea, restoring context verbatim around a session compaction rather than summarizing it.

    yibie/awesome-jev
  • Codex plugin for context compaction, using Jev within session lifecycle hooks to preserve essential history while pruning noise.

    logicrw/awesome-jev-projects
  • Codex port that restores Jev-selected verbatim history around native session compaction.

    cobanov/awesome-jev
  • Codex plugin: verbatim Jev-guided context restoration around session compaction. Port of tamaratran/fast-jev-compaction to Codex lifecycle hooks.

    hellogumbo/awesome-jev

Frequently asked questions

Should I use it?

The author says no; it is an experimental proof of concept.

What is the critique?

Pruning by probability discards hidden reasoning, conflicts with trained compaction flows and breaks caching.

What is it ported from?

tamaratran/fast-jev-compaction for Claude Code.

Related projects

  • Codex plugin that builds handoff packets by asking Jev whether each old read-only tool record should stay, keeping retained text verbatim.

    ProjectbetaJavaScript
  • Local compaction proxy for Claude Code and Codex that compacts more often with local rules and keeps full history searchable.

    ProxybetaGo
  • Claude Code plugin and npm library that replaces the compaction summary with Jev keep, truncate or drop scores for every tool call.

    PluginbetaTypeScript
  • Cursor CLI plugin that asks Jev keep-or-drop nouls for each tool call and result, then re-injects kept facts after compaction.

    CLIbetaTypeScript
  • DeepSeek Harness plugin that asks Jev which tool results can be pruned early, delaying or avoiding lossy compaction.

    ProjectbetaTypeScript
  • DeepSeek Harness compaction backend that asks Jev which tool calls and outputs to keep, preserving everything else verbatim.

    ProjectbetaJavaScript