Skip to content

Codex Cloud auto code review: silent fail + dashboard shows quota available but review says limit reached #15477

Open

Description

@AytuncYildizli

What variant of Codex are you using?

Codex Cloud (chatgpt.com/codex) — GitHub auto code review

What subscription do you have?

Pro

What platform is your computer?

macOS arm64 (Darwin 25.3.0)

What issue are you seeing?

Three bugs in one:

Bug 1: Silent failure with stale GitHub connector
Auto code review was enabled and configured correctly (auto review ON, "Review all PRs", trigger "On every push"), but PRs received zero review comments and zero error messages. No bot reaction, no 👀, nothing. Multiple PRs tested (#80, #81 on a private repo). The feature appeared to work in settings but silently did nothing.

Bug 2: Dashboard shows quota available but review says limit reached
The usage dashboard shows "Code review: 70% remaining (resets Mar 28)". But when actually triggering a review after fixing Bug 1, the response is "You have reached your Codex usage limits for code reviews." These two states directly contradict each other.

Bug 3: Disconnect/reconnect required to get ANY signal
The silent failure (Bug 1) was only resolved by disconnecting and reconnecting the GitHub connector in chatgpt.com/codex settings. After reconnect, at least an error message appeared (Bug 2). Before reconnect, there was zero indication anything was wrong.

Steps to reproduce

For Bug 1 (silent fail):

  1. Have a GitHub repo connected to Codex code review
  2. Enable auto review: "Review all PRs", trigger "On every push"
  3. Open PRs or push commits
  4. Observe: no review comments, no errors, no bot reactions
  5. Try @codex review in PR comments — still nothing

For Bug 2 (quota mismatch):

  1. Check usage dashboard — shows ~70% code review quota remaining
  2. Disconnect and reconnect GitHub connector
  3. Trigger a review (PR open or @codex review)
  4. Receive "You have reached your Codex usage limits for code reviews"
  5. Check dashboard again — still shows quota available

Expected behavior

  • If review cannot be triggered, show an error (not silent failure)
  • Dashboard quota should match actual review eligibility
  • GitHub connector state should not silently drift into a broken state

Actual behavior

  • Reviews silently fail with no indication
  • Dashboard shows 70% quota remaining while reviews are actually blocked
  • Only disconnect/reconnect surfaces the (incorrect) error

Screenshots

Settings page shows: auto review ON, repo connected, "Review all PRs", "On every push"
Usage dashboard shows: Code review 70% remaining, resets Mar 28, 2026
Review response after reconnect: "You have reached your Codex usage limits for code reviews"

Related issues

Root cause hypothesis

  1. GitHub App installation state drifts silently — events stop being ingested but UI shows everything as configured
  2. Usage quota tracking is inconsistent between the dashboard display and the actual review eligibility check
  3. No observability: no "last event received", no "last review attempted", no failure reason per repo

Suggested improvements

  1. Add per-repo diagnostics in settings: last webhook received, last review task created, last failure reason
  2. Fix quota display to match actual eligibility
  3. Fail loudly on GitHub (bot comment or check status) when a review cannot start
  4. Add a "re-sync permissions" button instead of requiring full disconnect/reconnect
  5. Detect stale connector state proactively and notify users

Activity

  1. added
    bugSomething isn't working
    codex-webIssues related to Codex Web
    code-reviewIssues relating to code reviews performed by codex
    rate-limitsIssues related to rate limits, quotas, and token usage reporting
    authIssues related to authentication and accounts
    on Mar 22, 2026
  2. AytuncYildizli commented on Mar 23, 2026

    @AytuncYildizli
    Author

    Not a duplicate. This issue covers the full failure chain that none of the suggested issues address individually:

    1. Silent connector drift — GitHub integration stops receiving events with zero indication. No error, no bot reaction, no failed check. The feature appears configured correctly in settings but silently does nothing. None of the linked issues cover this silent failure mode.

    2. Quota display mismatch — Dashboard shows "70% remaining" while the actual review endpoint returns "usage limit reached." This is a separate data inconsistency bug, not just fast depletion (Possible quota/accounting issue: Pro weekly limit depletes unusually fast #15405) or caching (Usage still being cached when switching accounts in Codex App #14730).

    3. Zero observability — No "last webhook received" timestamp, no "last review attempted" log, no failure reason per repo. The only way to discover the connector was broken was to disconnect and reconnect it manually.

    The linked issues each touch one piece, but this is the only report documenting the end-to-end experience: a user with quota, with correct settings, getting absolute silence, and only discovering the real state after a manual reconnect that surfaces a contradictory error.

    Happy to provide additional logs or screenshots if helpful.

  3. RudiPersson commented on May 26, 2026

    @RudiPersson

    I am seeing the same silent automatic-review failure.

    Repository: private GitHub repository
    Settings: Auto code review = Review my PRs; Review trigger = On PR open

    Observed:

    • Automatic review produced no comment, reaction, or review on multiple PRs.
    • Manual @codex review works.
    • Failure occurs on both codex/... and ordinary feature/... source branches.
    • I removed and reinstalled the GitHub Codex integration; the issue persisted.

    Latest reproduction:

    • Non-draft PR, authored by me.
    • No Codex activity more than 9 minutes after opening.
  4. bassings commented on Jun 5, 2026

    @bassings

    Seeing the same failure pattern on a ChatGPT Pro account with GitHub code review.

    Context:

    Timeline observed today:

    1. Initial @codex review request returned:

      You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

    2. Codex dashboard for the same ChatGPT account showed available general/shared usage, not an exhausted state.
    3. Disconnecting and reconnecting the GitHub connector changed behaviour: a fresh @codex review did run once and produced a real review finding on commit 3f011a98.
    4. After fixing that finding and pushing commit 68fe80b7, another @codex review immediately returned the same usage-limit message again.

    This does not look like a normal real quota exhaustion case from the user side. The connector can run a review after reconnect, then the quota error returns on the next request despite the dashboard not presenting an exhausted code-review state clearly enough to explain it.

    The main product issue is the same as this report describes: GitHub review eligibility and the Codex dashboard/connector state disagree, and the user has no reliable way to tell whether the failure is quota, stale connector state, workspace/account mapping, or a backend accounting bug.

    Useful diagnostics that would help:

    • Per-repo last webhook received / last review attempted / last failure reason
    • The exact account/workspace/quota bucket used by the GitHub app when posting a quota error
    • A dashboard row specifically matching the GitHub bot’s “code reviews” quota decision
    • A connector “resync” or health check instead of requiring disconnect/reconnect
  5. hagelien commented on Jun 14, 2026

    @hagelien

    Same issue here.

    Codex GitHub code review appears to be blocked by a usage-limit check even though my visible quota is not exhausted.

    What I see:

    • GitHub bot chatgpt-codex-connector comments on PRs:
      “You have reached your Codex usage limits for code reviews.”

    • Codex Analytics → Code review shows:

      • 0 PRs reviewed
      • 0 issues found
      • no data for the period
    • Codex Usage & billing shows:

      • Pro plan
      • General 5-hour usage limit: 93% left
      • General weekly usage limit: 84% left
      • GPT-5.3-Codex-Spark: 100% left
    • This has persisted for several days, despite no code reviews being completed.

    Another interesting detail: Code review analytics shows zero activity going back months, even though I know Codex was reviewing PRs heavily before the past 1–2 weeks. So the analytics page itself also appears not to reflect actual historical code-review activity on my account/repo.

    Things I have already tried, with no effect:

    • Uninstalled the Codex GitHub app.
    • Revoked GitHub access.
    • Reinstalled/re-authorized the Codex GitHub app.
    • Removed and added back the relevant environments in Codex.

    The issue still persists after all of this, so it does not seem to be fixed by reconnecting GitHub or recreating the Codex environment configuration.

  6. rudrakshbhandari commented on Jun 21, 2026

    @rudrakshbhandari

    Same issue here. ChatGPT Pro (20X) account.

  7. hagelien commented on Jun 25, 2026

    @hagelien

    After more research, I think my problem is I used to have a business account which is now deactivated. Apparently Codex might try to use quota from that when reviewing code. More here: #8503

  8. Tabberwocky commented on Jul 1, 2026

    @Tabberwocky

    Exact same issue.

    Image Image Image
  9. james-lancelot commented on Jul 9, 2026

    @james-lancelot

    Adding another current reproduction from a ChatGPT Pro account with a deactivated legacy workspace in its account history.

    Live state on 2026-07-09

    • Active personal ChatGPT account: Pro
    • Private organization repository is present in Codex Code Review settings
    • Personal Auto review: on
    • Repository setting: Review all PRs
    • Trigger: Smart Trigger (Experimental)
    • Enable credits use: on
    • General usage dashboard: 96% of the 5-hour limit remaining and 99% of the weekly limit remaining
    • Code review analytics: 0 PRs reviewed, 0 issues found, and no data for the period
    • Credits remaining: 0
    • A fresh manual @codex review on a PR authored by the connected GitHub user received the usage-limit response after 9 seconds
    • The GitHub App installation is healthy: the connector receives the event and posts the failure response, with the expected repository and pull-request permissions

    There is also a legacy/deactivated ChatGPT workspace associated with the account history. Earlier requests returned:

    This workspace is deactivated. Select an active workspace and try again.

    Later requests changed to:

    You have reached your Codex usage limits for code reviews.

    OpenAI Support previously summarized the problem as reviews routing to a deactivated workspace. The active Pro account's UI now shows no deactivated-workspace warning, but the GitHub connector does not expose which account, workspace, or quota bucket it is using.

    The smallest one-time credits option currently shown in checkout is 1,000 credits for $40, so the only available diagnostic workaround requires nontrivial spend and may still route to the wrong workspace.

    Could OpenAI clarify:

    1. Whether individual Pro currently includes any GitHub code-review allowance.
    2. Which account/workspace/quota bucket the GitHub connector is evaluating.
    3. Whether Support can clear a stale GitHub-to-workspace binding server-side.
    4. Whether credits purchased on the active personal Pro account would definitely be used by this installation.

    We are disabling automatic reviews for now to avoid posting a quota-error comment on every PR, and will retest after this issue has a confirmed fix or maintainer guidance.

  10. assafkip commented on Aug 3, 2026

    @assafkip

    Same failure, reproduced on 2026-08-03 with Codex CLI/Desktop 0.146.0 and a ChatGPT Plus account.

    Backend state from account/rateLimits/read at the same moment:

    • weekly usage: 83% used, so 17% remains
    • rateLimitReachedType: null
    • hasCredits: true
    • credit balance: 2500
    • spendControlReached: false

    Observed behavior:

    • codex exec works
    • a minimal codex cloud exec task is accepted, stays pending, then becomes ERROR with no diff
    • GitHub @codex review immediately says the code-review usage limit was reached
    • purchased credits record zero usage events
    • disconnecting GitHub from a second ChatGPT account, then disconnecting and reconnecting it to the Plus account, did not fix the routing

    I uploaded a diagnostic bundle from the affected installation through feedback/upload.

    Receipt: no-active-thread-019fc964-c85b-7a63-8871-d11096ebac99

    Could OpenAI trace the receipt and confirm which ChatGPT account/workspace the GitHub connector resolves to? There is still no user-facing UI that exposes that binding.

  11. assafkip commented on Aug 3, 2026

    @assafkip

    Confirmed account-routing mismatch from Codex desktop logs.

    Correct ChatGPT Plus account:
    82e4557b-6b10-4cd3-b32e-8f58fc91e56d

    Workspace still being resolved by the cloud surface:
    108e5ab5-d7fc-43d3-a0f8-dd0551833be1

    The second workspace is explicitly logged as:
    bannerType=workspace_owner_credits_depleted
    limitReason=credits

    This explains the exact behavior:

    • CLI works under the correct Plus identity.
    • GitHub and Linear cloud tasks immediately report exhausted usage.
    • The 2,500 credits on the Plus account record zero usage because the cloud tasks are routed to the other workspace.
    • Disconnecting and reconnecting GitHub did not clear the stale server-side binding.

    Diagnostic upload receipt:
    no-active-thread-019fc964-c85b-7a63-8871-d11096ebac99

    Requested fix: remove the stale connector binding to 108e5ab5-d7fc-43d3-a0f8-dd0551833be1 and bind cloud tasks to 82e4557b-6b10-4cd3-b32e-8f58fc91e56d.

  12. tom-takeru commented on Sep 15, 2026

    @tom-takeru

    Summary

    Since approximately September 14, 2026 UTC, Codex Cloud has stopped automatically starting code reviews for newly opened pull requests in a private GitHub repository.

    Automatic reviews had worked reliably for approximately six months up to this incident, and no settings were changed.

    Expected behavior

    With personal Auto review enabled and the review trigger set to On PR open, Codex should automatically react to and review a newly opened, non-draft pull request.

    Actual behavior

    • No Codex reaction (👀 or 👍), review task, review comment, or GitHub timeline event appears.
    • Manually commenting @codex review still starts and completes a review successfully.
    • The Codex Code review settings currently show:
      • Personal Auto review: ON
      • Review trigger: On PR open
      • Repository Auto review: Follow personal
      • Repository trigger: Follow personal
    • The settings page also shows:
      Action required — Update your GitHub permissions to enable code reviews
    • Clicking the Update button leads to a 404 page.

    Steps to reproduce

    1. Open a non-draft pull request in the connected private repository.
    2. Wait for the automatic review; no Codex activity appears.
    3. Add @codex review as a comment; the manual review works.
    4. Open Codex Code review settings and click Update; it leads to a 404 page.

    Expected result

    Automatic review should start when the pull request is opened, and the GitHub permission update flow should resolve successfully.

    Could this be related to a regression in pull_request.opened event ingestion, a stale GitHub connector state, or a broken permission-update link?

    Update (September 16, 2026 UTC)

    I performed an additional test on the affected private repository.

    • With repository-level Auto review set to Review my PRs and the trigger set to On PR open, a newly opened, non-draft pull request authored by me did not receive an automatic review.
    • A manual @codex review request on that pull request completed successfully.
    • I then changed only the repository-level Auto review setting to Review all PRs. The trigger remained On PR open.
    • After that change, automatic review started working.

    I also completed the GitHub organization Codex Connector authorization flow. The installation has access to all repositories and read/write access to pull requests. The Codex Code review settings page still displays the same permission warning.

    No other repository-level code review settings were changed. I am leaving the repository set to Review all PRs as a workaround.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    authIssues related to authentication and accountsbugSomething isn't workingcode-reviewIssues relating to code reviews performed by codexcodex-webIssues related to Codex Webrate-limitsIssues related to rate limits, quotas, and token usage reporting

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions