JevMade field notes / Technical guide

Keep selected evidence when Codex shortens a conversation

fast-jev-compaction is an experimental Codex add-on. It selects old tool evidence to retain, lets Codex make its normal conversation summary, and adds the saved text beside it.

Original by leonaaardobAgent workflows

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits and license

“fast-jev-compaction” by leonaaardob. Read the original source.

This expanded guide is an AI-narrated adaptation of the source’s essential explanation, examples and caveats, not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

License: MIT

MIT License

Copyright (c) 2025

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Our summary

A coding assistant may shorten a long conversation into a summary so it can continue working. This experimental Codex add-on first asks which old tool calls and results should be kept exactly. It adds that saved evidence beside Codex's normal summary rather than replacing the summary.

The add-on pairs tool calls with results and protects the first and newest messages. Separate Jev questions assess older pairs, while decisions and retained text are saved locally. Missing keys, failed requests, oversized information, or too little expected reduction leave Codex's normal process unchanged.

The maintainer explicitly advises against relying on this for real work. A high keep-or-drop score does not prove evidence is safe to lose. Retained text has a size limit, history formats may change, and rewriting conversation history can interfere with reuse of earlier processing.

Key takeaways

Keep recent messages and retained evidence exact.
Leave normal summarization unchanged when the extra step cannot run safely.
Treat this as an experiment, not a proven improvement.

More guides →See the experiments →