Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

[Bug] Anthropic pool quota exhaustion is returned as retryable rate_limit_exceeded, so Codex retries and shows 'exceeded retry limit, 429' instead of a usage-limit message

Closed
  • Is something wrong?
#6,991 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
45/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
typescript
Domain
api, backend

Research direction

Start with the cooldown handling in src/oauth/anthropic-routing.ts and the response paths in src/server/responses/adapter-dispatch.ts and src/server/responses/passthrough-dispatch.ts; then inspect the Codex account path in src/codex/auth-context.ts. Check how cooldown sources and reset times reach error responses, and run the proposed tests for reset-derived and retry-after cooldowns. Done means quota exhaustion returns usage_limit_reached with resets_at while short throttles remain rate_limit_exceeded.

Written by the indexing model from the issue text.

Description

account-pool bug
Client or integration

Codex App

Area

Authentication and account pool

Summary

When every Anthropic OAuth account in the pool is out of its five-hour or weekly quota, OpenCodex answers Codex with HTTP 429 rate_limit_error / rate_limit_exceeded. Codex treats that class as transient, keeps retrying ("reconnecting 1/5 ... 5/5") and ends with exceeded retry limit, last status: 429 Too Many Requests. The user sees a long "thinking" state and never learns that the quota is spent or when it resets.

I expected the same terminal class that OpenAI returns for a spent plan: usage_limit_reached with resets_at. Codex renders that class as "Usage limit reached ... try again at " and stops at once.

This is the Anthropic-pool form of #6000, which was closed automatically for the issue template and not fixed.

Reproduction
  1. Configure an Anthropic OAuth account pool and let every account reach its five-hour (or weekly) limit.
  2. Send a Codex turn with anthropic/claude-opus-5-5.
  3. Observe: the pool logs no eligible replacement; returning 429, and every later turn is refused locally in 2-9 ms with rate_limit_exceeded. Codex retries each turn (one conversation logged 16 refused requests) and shows only the retry-limit message.
Version

2.83.0

Operating system

macOS (Bun 1.4.2), Codex 0.162.1

Provider and model

anthropic / claude-opus-5-5 (any Anthropic OAuth pool model)

Logs or error output
[anthropic-pool] 429 on <account-a>; failing over to <account-b>
[anthropic-pool] no eligible replacement; returning 429
usage.jsonl: provider=anthropic status=429 errorCode=rate_limit_exceeded failureCause=rate-limit durationMs=2..9 (78 rows)
Codex: exceeded retry limit, last status: 429 Too Many Requests
Source-level analysis (v2.83.0)
  • src/oauth/anthropic-routing.ts:75-80: AnthropicAccountCooldownError ("All Anthropic OAuth accounts are temporarily rate-limited"). The pool already knows each cooldown source (retry-after, reset-derived, default, lines 84-100), so it knows when a cooldown is a drained quota window and when it resets.
  • src/server/responses/adapter-dispatch.ts:518-520 and src/server/responses/passthrough-dispatch.ts:951-953 turn that error into formatErrorResponse(429, "rate_limit_error", ...) with Retry-After. src/lib/errors.ts:448-458 maps every 429 to rate_limit_exceeded. The quota information is lost here.
  • The comment at adapter-dispatch.ts:521-524 says "the Codex client ... does not retry a 429", but Codex 0.162.1 does retry rate_limit_exceeded and ends with the retry-limit message.
Proposed change
  • When AnthropicAccountCooldownError fires and every blocking cooldown is reset-derived (quota window), return HTTP 429 with error.type = "usage_limit_reached", resets_at = earliest reset (epoch seconds), and a message that names the provider and the reset time. Keep rate_limit_exceeded for short retry-after throttles.
  • Apply the same rule to the Codex-account cooldown path (src/codex/auth-context.ts:946).
  • Test: a reset-derived cooldown yields usage_limit_reached + resets_at; a retry-after cooldown yields rate_limit_exceeded + Retry-After.

Related: #6989 (Retry-After-sourced cooldowns not recovering).

Checks
  • I searched existing issues and documentation.
  • I removed secrets, tokens, account details, request credentials, and personal data.
Dominant language
TypeScript
Stars
16.9k
Forks
1.3k
Avg merge
7h 38m
Merged PRs (30d)
659

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from lidge-jun/opencodex

All issues in lidge-jun/opencodex

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.