Codex 0.146.0: Smarter Tool Selection and Safer Sessions
This episode breaks down Codex 0.146.0’s smarter skill selection, including Reciprocal Rank Fusion, dynamic context budgets, and graceful truncation when tool catalogs get too large. It also covers clearer warnings, session naming improvements, automatic MCP server reconnection, and new Agent Plugin manifest support across toolchains.
Chapter 1
Reciprocal Rank Fusion and Skill Selection in Codex 0.146.0
Ethan Park
So I was looking at the pull requests for Codex version 0.146.0 last night, and thanks to Jellypod for helping us bring this daily break down to life every morning. But Maya, they completely reworked how tools actually get loaded into the model context.
Maya
Wait, you mean instead of just dumping every single Model Context Protocol skill into the system prompt at the start of a session?
Ethan Park
Exactly. Pull requests 34547 and 34581 introduced Reciprocal Rank Fusion, or RRF, right into the skill selection pipeline. So now, instead of taking every tool in your workspace and shoving hundreds of tool definitions into the prompt, it runs a hybrid search over routing cards and lexical metadata to rank what you actually need for that specific turn.
Maya
Oh, that is huge. Because if you have a massive enterprise set up with, say, fifty different MCP servers attached, those skill definitions were just eating up thousands of tokens before you even typed a single word of code.
Ethan Park
Right! And pull request 34626 ties those metadata budgets directly to the active model context window. So if you are on a smaller context limit, it tightens the budget dynamically. But here is the really clever part. Pull request 34738 adds a multi stage fallback behavior.
Maya
A fallback? Like, what happens when you start pushing right up against that cap?
Ethan Park
It does not just drop the skill entirely right away. First, Codex strips out the verbose descriptions and long form documentation from the tool definitions, keeping just the raw function signatures. That way it can retain more available skills under tight context budgets before it has to resort to dropping anything.
Maya
Wow, so it degrades gracefully instead of just chopping off half your toolkit. That actually explains something I ran into last week. You know that custom database query tool I was building?
Ethan Park
The one with the three paragraph description explaining how to join four legacy tables?
Maya
Yes, that exact one! I kept wondering why the agent would randomly pretend the tool did not exist whenever my context got deep. I ended up rewriting the routing card keywords to be super tight and punchy, like three words instead of a essay, and boom, it stayed in the active set every single time.
Chapter 2
Dynamic Truncation Warnings and Session Management
Ethan Park
And that brings us right to pull request 34785, which fixes that exact silent drop issue you were dealing with. Codex now explicitly warns when skill catalogs must be truncated. You get a clear terminal warning right in the interface showing you that a tool was left out due to space constraints.
Maya
Man, where was that warning three weeks ago? That would have saved me two hours of head scratching. No more mystery drops.
Ethan Park
Seriously. And speaking of keeping your workflow clean, pull request 34605 updated session management. You can now pass string names straight into slash new or slash clear. So you can type slash new feature test or slash clear bugfix, and it labels or resets the branch right there without breaking your flow.
Maya
Oh, I love that. I am constantly spinning up quick side experiments and forgetting which temporary session was doing what. Being able to name them on creation is such a nice quality of life bump.
Ethan Park
Same here. And under the hood in pull request 34952, they added automatic MCP server reconnection. If an underlying server drops or needs to re authenticate, Codex automatically reconnects in the background without dropping your active user interface state or interrupting parallel connections that are working fine.
Maya
That is so important for reliability when you are chaining multiple remote agents together. And speaking of multi platform workflows, pull request 35105 added native support for Agent Plugin manifests. So you can publish workspace configs across Amazon Bedrock and Claude Code toolchains without rewriting your definitions.
Ethan Park
It really feels like the team is focusing on making the whole ecosystem feel like a polished product rather than a bunch of loose developer scripts held together with tape.
Maya
Absolutely. Alright, I am going to go go prune some more tool descriptions. Talk to you tomorrow, Ethan.
Ethan Park
Catch you later.