SkillFoxx

Codex Overload Fix skill

A local Python relay that retries Codex requests rejected for capacity reasons, before any output has started streaming.

  • 265 ★ on the repository
  • Last commit
  • MIT
Skill

Broad access

Why this level

  • The relay carries all Codex traffic, including the OAuth token, and installs autostart.
  • The installer edits the Codex `config.toml` and the user autostart of the system.
All reasons and checks

newbeem4n/codex-overload-fixAuthor:NewBeeM4n

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/codex-overload-fix

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The relay carries all Codex traffic, including the OAuth token, and installs autostart.
  • The installer edits the Codex `config.toml` and the user autostart of the system.

This is third-party code. Review the repository files before installing.

What it does

Codex Overload Fix is a local Python relay with no external dependencies. It sits between Codex and the official OpenAI endpoint, recognises overload errors (`server_is_overloaded`, `queue_full`, HTTP 429 and 5xx) and retries a request while the client has received nothing yet. Installers for Windows, macOS and Linux change two Codex `config.toml` entries, `model_provider` and `openai_base_url` (address `http://127.0.0.1:8787`), back the file up and register autostart. Headers, including the OAuth token, and request bodies pass through untouched. An optional dashboard shows retry counters at `127.0.0.1:8899`, and a skill bundle packages the same runtime.

Who it is for. Codex users interrupted every few turns by the Selected model is at capacity error.

Good fit when

  • A long Codex task breaks off on the second or third turn because of overload.
  • You need to see how many requests the server declines: the dashboard shows retries and exhausted attempts.
  • An agent is asked to install it: the skill explains the changes first, then sets up the relay.
  • Codex settings have to return to how they were, and the uninstaller restores `config.toml` byte for byte.
  • Network access goes through a local proxy, and its address is passed with `--proxy`.

Not a fit when

  • The model started answering worse: the relay only fixes interruptions.
  • A quota is exhausted or a login expired: those errors pass through unchanged.
  • A background process with autostart is not allowed on the work machine.

Example request

Diagnose my recurring Codex capacity errors, explain the changes and install the local relay only after I agree.

Limitations

The relay reduces interruptions but adds no capacity and does not improve model quality. The author does not promise zero account risk, since any client that changes request timing departs from default behaviour. Retries happen only before the answer stream starts, afterwards errors go to the client. Long retries hold a turn open, so waiting stretches on a slow link. A current macOS and Ubuntu are verified, while Intel Macs, older macOS releases and Linux without systemd are untested.

How to disable. Run the uninstaller for your system: `bash scripts/uninstall-mac.sh`, `bash scripts/uninstall-linux.sh` or `scripts\uninstall.ps1`. It restores `config.toml` from the backup and only then stops the relay. Restart Codex afterwards.

FAQ

Does the relay bypass limits?

No. It re-sends a request the server already declined, like pressing retry. An exhausted quota and an expired login pass through unchanged.

What happens to the token?

Headers, including the OAuth token, are forwarded as they are. The author says the relay neither stores nor logs them, and its log holds event types and request sizes.

How can I check it is in the path?

During a turn look at connections on `127.0.0.1:8787`. `python dashboard/server.py` shows retry counters at `http://127.0.0.1:8899/`.

How do I restore the old settings?

Run the uninstaller for your system. It restores `config.toml` from the backup and stops the relay.

Which systems are supported?

Windows, macOS and Linux, each with an installer and an uninstaller. Python 3.9 or newer is needed and no third-party packages.

Discussion

No questions or reviews yet.

Skills
Editors’ pick

OpenAI's coding agent that runs locally in your terminal, with IDE and desktop versions

CLIBroad access135.1K

Codex Auto Resume on Windows continues the same Codex conversation after a usage limit resets, finding it by UUID

PluginExtended access92

oh-my-codex

oh-my-codex (OMX)

Needs a VPN

A layer on top of Codex CLI: one flow from clarifying a task to verification, agent teams, hooks, a HUD and state in .omx/

CLIBroad access33.8K

CodexPotter loops Codex with a clean context each round until the code matches the goal, as a replacement for the built-in /goal.

SkillBroad access636
Needs a VPN

An MCP server that hands tasks to the Codex CLI from inside Claude Code: code questions, change review, web search and resumable sessions

MCP serverBroad access649

A Codex skill for an autonomous loop: Claude Code computes, Codex reviews hourly, critiques and sends the next prompt.

SkillBroad access80

README in short

Retries run only before the answer stream starts. Original headers and request bodies are forwarded unchanged, and the log keeps event types and sizes.

README badge

Report a problem