fix(responses): surface provider 413 as terminal context overflow by Ingwannu · Pull Request #3177 · lidge-jun/opencodex

Ingwannu · 2026-09-01T16:58:29Z

Summary

convert a confirmed streaming Responses input-size 413 into one terminal response.failed with context_length_exceeded, so Codex stops its 5/5 reconnect loop and can apply next-turn compaction
cover local input admission, direct passthrough, translated adapters, exhausted Anthropic image tightening, and the outer combo boundary without silently deleting prompts or images
preserve non-streaming 413 status/body fidelity and unrelated 4xx/5xx behavior; use a bounded proxy-owned failure message so an upstream 413 body cannot echo request content
document the transport decision and user-visible continuation behavior

Verification

Exact head 6d5eea6a163106232242af14335f69f34ca6cad5 on base 22a643a00b5974fa53b084a04491f60d56ec9ee2:

exact-head cross-platform CI — passed, including all four Linux shards, macOS, keyring checks, npm-global checks, storage policy, typecheck/gates, and React Doctor
bun run typecheck — passed
bun test ./tests/responses-context-overflow.test.ts — 6 passed, 0 failed
translator-budget production contract — 1 passed, 0 failed
bun run test:changed — 2,887 passed, 4 skipped, 0 failed across 147 import-connected files
cd docs-site && bun run build — passed, 409 pages built
local full repository suite — 16,952 passed, 16 skipped, with only the three shutdown-launcher.test.ts 20-second startup timeouts failing under the constrained runner; the same three failures reproduced unchanged on clean origin/dev at 22a643a00b5974fa53b084a04491f60d56ec9ee2
git diff --check origin/dev...HEAD — passed
all local validation used isolated temporary HOME, OPENCODEX_HOME, and CODEX_HOME, nice -n 10, and no more than two CPUs; protected runtime configuration hashes remained unchanged

Checklist

Scope stays focused and avoids unrelated cleanup.
Docs or release notes were updated when needed.
Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults. Streaming 413 bodies are not relayed; non-streaming compatibility stays unchanged.

Closes #3170.

github-actions · 2026-09-01T16:58:40Z

✅ Deterministic PR hygiene checks passed.

chatgpt-codex-connector · 2026-09-01T17:08:17Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-01T17:12:19.533891Z	`6d5eea6`	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6d5eea6a16

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-09-01T17:12:17Z

      // The bounded reader owns the original body, deadline, abort settlement, and lock.
      // Unsafe partial data falls back to #452's non-empty status-only JSON.
      const errorText = await readDisplaySafeErrorText(upstreamResponse, upstream.signal, "");
+      if (upstreamResponse.status === 413 && clientRequestedStream) {


Preserve 413 for translated non-streaming callers

When a request enters through /v1/chat/completions or /v1/messages with stream:false, both wrappers force the internal Responses body to stream:true, so this condition—and the equivalent translated-adapter condition below—converts the provider's 413 into an HTTP-200 Responses SSE failure. The wrappers then fold that failure into HTTP 502, changing a terminal client-input error into a retryable server error instead of retaining the upstream 413 contract documented by this change. Restrict the conversion to genuine streaming Responses ingress (for example via options.inboundWire) or carry the original caller's stream mode through the replay.

AGENTS.md reference: src/AGENTS.md:L19-L19

Useful? React with 👍 / 👎.

@olddonkey

…dies Reimplements #3142 (thanks @olddonkey) with the guard off by default, and adds the rebuild site and refusal shape that version was missing. The measurement, local refusal, image diagnostics, body-observation release and probe-lease handling are that PR's work and are kept. Why default-off rather than the 15 MiB default: the only measured ceiling in this codebase belongs to the WebSocket transport, and the comment recording it says the same body still succeeds over HTTP SSE. #2473 acts on that by falling back to HTTP rather than refusing, and #2426 records an 18.2 MB HTTP 200. A 15 MiB default would therefore refuse turns that work today - on canonical ChatGPT as well as on Azure and custom Responses gateways whose limits were never measured at all. An unset proxy now measures nothing and sends exactly what it sends today. Two fixes beyond the original: - The stored/main pool 401 replay rebuilds its body and sent it unchecked. It is now guarded like every other build site; a replay is precisely when a grown payload reappears. - A streaming refusal returns terminal response.failed / context_length_exceeded instead of a JSON 413. Codex treats HTTP 413 as a retryable transport error and resends the same oversized body, which is the loop this feature exists to stop. That is the contract the upstream-413 path already uses (#3177). RequestLogContext gains a proxy-owned errorCode so a locally refused request is named as such instead of being classified from a status with no upstream message behind it. Refs #3142. Related to #2511, which asks for per-provider downscaling and pruning and is deliberately not implemented here.

@olddonkey

…dies (#3196) * feat(responses): opt-in ceiling for oversized outbound passthrough bodies Reimplements #3142 (thanks @olddonkey) with the guard off by default, and adds the rebuild site and refusal shape that version was missing. The measurement, local refusal, image diagnostics, body-observation release and probe-lease handling are that PR's work and are kept. Why default-off rather than the 15 MiB default: the only measured ceiling in this codebase belongs to the WebSocket transport, and the comment recording it says the same body still succeeds over HTTP SSE. #2473 acts on that by falling back to HTTP rather than refusing, and #2426 records an 18.2 MB HTTP 200. A 15 MiB default would therefore refuse turns that work today - on canonical ChatGPT as well as on Azure and custom Responses gateways whose limits were never measured at all. An unset proxy now measures nothing and sends exactly what it sends today. Two fixes beyond the original: - The stored/main pool 401 replay rebuilds its body and sent it unchecked. It is now guarded like every other build site; a replay is precisely when a grown payload reappears. - A streaming refusal returns terminal response.failed / context_length_exceeded instead of a JSON 413. Codex treats HTTP 413 as a retryable transport error and resends the same oversized body, which is the loop this feature exists to stop. That is the contract the upstream-413 path already uses (#3177). RequestLogContext gains a proxy-owned errorCode so a locally refused request is named as such instead of being classified from a status with no upstream message behind it. Refs #3142. Related to #2511, which asks for per-provider downscaling and pruning and is deliberately not implemented here. * docs(devlog): record live inventory after 3190 --------- Co-authored-by: jun <jun@lidge.dev>

…dge-jun#3177) Co-authored-by: Ingwannu <ingwannu@users.noreply.github.com>

@olddonkey

…dies (lidge-jun#3196) * feat(responses): opt-in ceiling for oversized outbound passthrough bodies Reimplements lidge-jun#3142 (thanks @olddonkey) with the guard off by default, and adds the rebuild site and refusal shape that version was missing. The measurement, local refusal, image diagnostics, body-observation release and probe-lease handling are that PR's work and are kept. Why default-off rather than the 15 MiB default: the only measured ceiling in this codebase belongs to the WebSocket transport, and the comment recording it says the same body still succeeds over HTTP SSE. lidge-jun#2473 acts on that by falling back to HTTP rather than refusing, and lidge-jun#2426 records an 18.2 MB HTTP 200. A 15 MiB default would therefore refuse turns that work today - on canonical ChatGPT as well as on Azure and custom Responses gateways whose limits were never measured at all. An unset proxy now measures nothing and sends exactly what it sends today. Two fixes beyond the original: - The stored/main pool 401 replay rebuilds its body and sent it unchecked. It is now guarded like every other build site; a replay is precisely when a grown payload reappears. - A streaming refusal returns terminal response.failed / context_length_exceeded instead of a JSON 413. Codex treats HTTP 413 as a retryable transport error and resends the same oversized body, which is the loop this feature exists to stop. That is the contract the upstream-413 path already uses (lidge-jun#3177). RequestLogContext gains a proxy-owned errorCode so a locally refused request is named as such instead of being classified from a status with no upstream message behind it. Refs lidge-jun#3142. Related to lidge-jun#2511, which asks for per-provider downscaling and pruning and is deliberately not implemented here. * docs(devlog): record live inventory after 3190 --------- Co-authored-by: jun <jun@lidge.dev>

…dge-jun#3177) Co-authored-by: Ingwannu <ingwannu@users.noreply.github.com>

@olddonkey

…dies (lidge-jun#3196) * feat(responses): opt-in ceiling for oversized outbound passthrough bodies Reimplements lidge-jun#3142 (thanks @olddonkey) with the guard off by default, and adds the rebuild site and refusal shape that version was missing. The measurement, local refusal, image diagnostics, body-observation release and probe-lease handling are that PR's work and are kept. Why default-off rather than the 15 MiB default: the only measured ceiling in this codebase belongs to the WebSocket transport, and the comment recording it says the same body still succeeds over HTTP SSE. lidge-jun#2473 acts on that by falling back to HTTP rather than refusing, and lidge-jun#2426 records an 18.2 MB HTTP 200. A 15 MiB default would therefore refuse turns that work today - on canonical ChatGPT as well as on Azure and custom Responses gateways whose limits were never measured at all. An unset proxy now measures nothing and sends exactly what it sends today. Two fixes beyond the original: - The stored/main pool 401 replay rebuilds its body and sent it unchecked. It is now guarded like every other build site; a replay is precisely when a grown payload reappears. - A streaming refusal returns terminal response.failed / context_length_exceeded instead of a JSON 413. Codex treats HTTP 413 as a retryable transport error and resends the same oversized body, which is the loop this feature exists to stop. That is the contract the upstream-413 path already uses (lidge-jun#3177). RequestLogContext gains a proxy-owned errorCode so a locally refused request is named as such instead of being classified from a status with no upstream message behind it. Refs lidge-jun#3142. Related to lidge-jun#2511, which asks for per-provider downscaling and pruning and is deliberately not implemented here. * docs(devlog): record live inventory after 3190 --------- Co-authored-by: jun <jun@lidge.dev>

fix(responses): surface provider 413 as terminal context overflow

6d5eea6

github-actions Bot added the bug Something isn't working label Sep 1, 2026

Ingwannu mentioned this pull request Sep 1, 2026

Handle provider input size limit reached gracefully in codex #3170

Closed

3 tasks done

lidge-jun marked this pull request as ready for review September 1, 2026 17:08

lidge-jun self-requested a review as a code owner September 1, 2026 17:08

lidge-jun merged commit 0d6424f into dev Sep 1, 2026
22 checks passed

lidge-jun deleted the ingw/fix-3170-responses-413-context-overflow branch September 1, 2026 17:08

chatgpt-codex-connector Bot reviewed Sep 1, 2026

View reviewed changes

lidge-jun mentioned this pull request Sep 1, 2026

feat(responses): opt-in ceiling for oversized outbound passthrough bodies #3196

Merged

5 tasks done

lidge-jun mentioned this pull request Sep 9, 2026

accept 413 gracefully #3675

Closed

2 tasks done

DamnUi mentioned this pull request Sep 9, 2026

Non-streaming 413 never reaches the terminal context-overflow mapping, so compaction fails with a raw 413 #4112

Closed

2 tasks done

tarunravi pushed a commit to tarunravi/opencodex that referenced this pull request Sep 14, 2026

fix(responses): surface provider 413 as terminal context overflow (li…

0ce269f

…dge-jun#3177) Co-authored-by: Ingwannu <ingwannu@users.noreply.github.com>

agentHits pushed a commit to agentHits/opencodex that referenced this pull request Sep 17, 2026

fix(responses): surface provider 413 as terminal context overflow (li…

d8893df

…dge-jun#3177) Co-authored-by: Ingwannu <ingwannu@users.noreply.github.com>

Navigation Menu

Repository navigation

fix(responses): surface provider 413 as terminal context overflow - #3177

fix(responses): surface provider 413 as terminal context overflow#3177
lidge-jun merged 1 commit into
devfrom
ingw/fix-3170-responses-413-context-overflow

Ingwannu commented Sep 1, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Sep 1, 2026

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Sep 1, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Sep 1, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

Ingwannu commented Sep 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Verification

Checklist

Uh oh!

github-actions Bot commented Sep 1, 2026

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Sep 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Sep 1, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Ingwannu commented Sep 1, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Sep 1, 2026 •

edited

Loading