grok token导入时无法有效检测出失效账号

[ZORfree]

如题，导入了几千个号，检测全部正常，然后 测试grok模型，提示400上游错误

导入了一个真正有效token，响应比较慢，但可以正常出

由此还有一个问题，批量删除很麻烦，因为最多一页只有100个

[github-actions]

AI issue labeler summary (classification based on trusted dev branch evidence): Labels: bug, feature, needs information, priority: medium Evidence-sensitive labels considered: bug Confidence: 0.63 Reason: Issue reports two actionable problems: (1) Grok invalid accounts pass import check but fail later with upstream 400, which is consistent with a likely account-validation defect; (2) batch deletion is cumbersome due to paging limits, which is a product enhancement request. Dev evidence shows import triggers refresh, but no direct dev-branch tests/evidence were found proving invalid Grok token detection is already fixed in the import/refresh path, so this is not labeled fixed-in-dev/awaiting-release. Code evidence summary: Dev branch evidence supports applying bug (not fixed-in-dev): account import calls refresh immediately, and Grok provider has upstream-status mapping helpers (including abnormal/limited), so there is intended detection logic; however, provided dev tests focus on payload/header/response parsing and do not verify invalid-token detection during bulk import/refresh, and no matching fix + tests were shown for this reported behavior. Therefore fixed-in-dev/awaiting-release is not supported by current dev evidence. The second complaint about deletion ergonomics is supported as a feature request: API supports token-based/bounded deletion modes, but provided evidence does not show a dedicated bulk-delete-all-across-pages UX flow. Bounded dev branch evidence used: - api/accounts.py: create_accounts() invokes account_service.refresh_accounts(tokens) after import - api/accounts.py: delete_accounts() supports mode 'tokens' and 'limited' only - services/providers/grok.py: _feedback_status maps 401/403->异常 and 402/429->限流 - services/providers/gro... [truncated]

[zqbxdev]

已在 cdc7121 修复该问题，并在 7e7e9ea 补充了回归测试。

根本原因：Grok 账号导入后会调用账号刷新检测，但原来的 refresh_accounts() 只保留 GPT provider，Grok token 被直接跳过，因此导入时会保持默认“正常”状态，实际调用模型时才暴露上游 400/认证失败。

修复内容：

• Grok 账号刷新现在会通过 grok.com app-chat 的 rate-limits 接口做低成本校验，不再跳过。
• 认证/session/token 失效会按现有账号策略标记异常或自动移除。
• 402/429 会标记为限流。
• Grok 校验错误返回已做脱敏，避免把底层 Cookie/token 信息透出到 API 响应。
• 已补充单元测试覆盖有效、失效、错误脱敏和 malformed response 场景。

验证：python3 -m unittest test.test_account_provider test.test_grok_provider 通过，100 个测试 OK。

感谢 @ZORfree 反馈这个问题。

[zqbxdev]

问题已修复并已合入 dev，先关闭该 issue。如后续仍遇到 Grok token 检测异常，可以带上新的错误上下文继续反馈。