Claude account suspended after roughly half a day with CodexBar running (v0.45.2, Auto source)

[CCLCK]

Summary

My Claude account was suspended after CodexBar had been running for roughly half a day. The account had been stable before this, and identity verification / KYC was already complete.

This is a temporal correlation report, not a claim of proven causation. I do not yet have evidence tying the suspension to a particular CodexBar request. I am reporting it so the Claude polling paths can be audited and other users can be warned if necessary.

Timeline and observed behavior

• Before CodexBar: Claude and Claude Code had been working normally without an account suspension.
• Installed and ran CodexBar 0.45.2 (build 109), then enabled the Claude provider.
• Claude quota was initially visible in CodexBar.
• Claude source was Auto; refresh frequency was Every 5 minutes.
• After roughly half a day, the Claude account became suspended / disabled.
• Afterwards, CodexBar reported: No available fetch strategy for claude.
• claude auth status --json now reports loggedIn: false and authMethod: none.
• Local CodexBar state indicates that both the delegated OAuth refresh path and the Claude CLI probe path were exercised while Auto was selected.
• KYC / identity verification had already been completed before this happened.

No account email, OAuth token, cookie, or other credential is included in this report.

Environment

• CodexBar: 0.45.2 (109)
• Claude Code CLI: 2.1.216
• macOS: 26.5.2 (25F84), Apple Silicon
• Claude provider source: Auto
• Refresh frequency: Every 5 minutes

Related issue

#2251 reported that the Claude CLI usage probe could create a new empty account-side session for each cold probe. #2263 fixed that behavior by reusing a persisted session ID, and that fix is present in v0.45.2. Therefore this suspension report may be distinct, but the earlier issue shows that passive quota probing can have provider-side effects and seems relevant to the risk review.

Policy context

Anthropic's current Claude Code legal/compliance documentation says that subscription OAuth authentication is intended for Claude Code and Anthropic's native applications, and directs third-party products to API-key authentication:

https://code.claude.com/docs/en/legal-and-compliance

CodexBar only reads usage data rather than running model inference, so it is unclear whether the quota-only OAuth path is considered supported. A clear statement or warning in CodexBar would help users make an informed choice.

Expected behavior

A passive quota monitor should not materially increase account-enforcement risk. If a provider path is not explicitly supported, CodexBar should avoid selecting it automatically or should show a prominent warning before enabling it.

Requested investigation / safeguards

1. Confirm exactly which OAuth, Web, and CLI requests may run under Claude Auto, including fallback behavior and request cadence.
2. Check whether five-minute polling or delegated OAuth refresh can resemble unsupported third-party token use or automated abuse to Anthropic.
3. Expose the last resolved Claude source and recent request count in a redacted diagnostic export.
4. Consider making Claude polling manual or CLI-only by default until the supported authentication boundary is clear.
5. Warn users before CodexBar reads or reuses Claude subscription OAuth credentials.

I can provide a redacted screenshot of the suspension notice or additional diagnostics if maintainers specify what is safe and useful. Please do not request OAuth tokens, cookies, or full Keychain exports.

[clawsweeper]

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed July 20, 2026, 11:15 PM ET / July 21, 2026, 03:15 UTC.

Summary
Keep this issue open for a maintainer-led Claude provider-boundary audit. The report does not prove causation, but it raises a concrete account-safety concern beyond the earlier session-pollution fix, and the repository’s conservative onboarding policy does not support auto-closing it.

Reproducibility: no. The reported suspension is a provider-side outcome with only temporal correlation, and no safe current-main reproduction path has been established.

Ways to help us reproduce this

• Add a screenshot or short recording showing the behavior.
• Add expected vs actual behavior.
• Include redacted logs or terminal output.

Next step
A maintainer must decide the acceptable Claude OAuth boundary before any implementation work; do not queue an automated fix from a temporal-correlation report.

Maintainer decision needed

• Question: Should CodexBar continue to auto-select delegated Claude OAuth polling until a maintainer confirms that the active request pattern is within Anthropic’s supported third-party boundary?
• Rationale: The account-suspension report is safety-relevant but causal attribution is unresolved, so the tradeoff between monitoring convenience and provider-policy caution needs explicit maintainer judgment.
• Likely owner: steipete — The recent merged work on the adjacent Claude usage-probe behavior provides the clearest available ownership signal.
• Options:
  • Pause delegated Auto polling (recommended): Temporarily make the delegated OAuth path opt-in or unavailable in Auto while the request audit and provider-boundary review are completed.
  • Keep current behavior with disclosure: Retain the current Auto selection only after documenting exact request paths and showing an explicit warning before credential reuse.
  • Collect more diagnostics first: Request redacted source-resolution and request-cadence diagnostics before changing behavior, accepting continued uncertainty in the interim.

Review details

Best possible solution:

Audit and document every Claude Auto request path and cadence, then add a narrowly scoped warning, safer default, or redacted diagnostics only where that audit demonstrates avoidable account-risk exposure.

Do we have a high-confidence way to reproduce the issue?

No. The reported suspension is a provider-side outcome with only temporal correlation, and no safe current-main reproduction path has been established.

Is this the best way to solve the issue?

Unclear. An audit of the active Auto source request paths is required before selecting a warning, default change, or opt-in policy.

AGENTS.md: found and applied where relevant.

Remaining risk / open question:

• The report does not establish which Auto request path, request cadence, or provider policy caused the suspension.
• Changing the default or adding a warning before the current fallback behavior is audited could remove useful monitoring or make unsupported claims about Anthropic policy.

Codex review notes: model internal, reasoning high; reviewed against cc8da27cec92.

Label changes

Label changes:

• add P1: A potentially unsupported background credential-use path may expose active users to account suspension, although causation and blast radius are not yet confirmed.
• add impact:security: The report concerns third-party reuse of subscription OAuth credentials and a possible account-enforcement boundary.
• add impact:auth-provider: The affected surface is Claude Auto source selection, OAuth refresh, CLI probing, and resulting authentication state.
• add issue-rating: 🦪 silver shellfish: Current issue advisory state selects this label.
• add clawsweeper:no-new-fix-pr: Current issue advisory state selects this label.
• add clawsweeper:needs-maintainer-review: Current issue advisory state selects this label.
• add clawsweeper:needs-product-decision: Current issue advisory state selects this label.
• add clawsweeper:needs-security-review: Current issue advisory state selects this label.

Label justifications:

• P1: A potentially unsupported background credential-use path may expose active users to account suspension, although causation and blast radius are not yet confirmed.
• impact:security: The report concerns third-party reuse of subscription OAuth credentials and a possible account-enforcement boundary.
• impact:auth-provider: The affected surface is Claude Auto source selection, OAuth refresh, CLI probing, and resulting authentication state.

Evidence reviewed

What I checked:

• Reporter evidence: The report documents a Claude suspension after using Auto source with a five-minute refresh interval, while explicitly acknowledging that it establishes temporal correlation rather than proven causation.
• Adjacent merged fix: Fix Claude probe session pollution #2263 merged on 2026-07-17 to reuse a Claude CLI probe session identity; its supplied description distinguishes delegated-refresh policy gating from that session-pollution fix. (cfb7b8036087)
• Inspection limitation: The required read-only local inspection command could not start because the sandbox failed to create its loopback interface, so current-main code, VISION.md, maturity data, and git provenance could not be independently verified.

Likely related people:

• steipete: The supplied related-item context identifies this person as author of the recent merged Claude CLI probe-session fix, which is adjacent to the requested polling audit. (role: recent area contributor; confidence: medium; commits: cfb7b8036087)

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.