Recently, I’ve been using the Codex app for longer coding tasks, and the context compaction has honestly been driving me crazy.
After the recent updates, it feels like this issue happens more often: once the context gets full, Codex starts compacting it, and sometimes the whole session just fails.
The most painful part is starting a new conversation and asking it to “extract the previous context and continue working.”
On the surface, it seems like it is working hard.
In reality, it feels more like:
“Who am I? Where am I? What was I supposed to do again?”
It understands the words, but the actual result is often not good enough.
So I built a small tool called codexgo.
Instead of asking Codex to reconstruct the task from natural language, codexgo reads the local Codex state database directly, finds the previous session thread, and identifies the real user request more accurately.
This makes it much easier to continue a task without having to explain everything again.
If you find it useful, I’d really appreciate a star ⭐
If you run into any issues, feel free to DM me or open an issue on GitHub.
Also, feel free to check out my other projects too :)
Top comments (0)