Claude helper resumes repeatedly return Cloudflare 403 in an existing session

未关闭
  • 哪里不对？
#36,121 3 条评论 11 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
18/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
typescript
领域
backend

调研方向

Start with ClaudeAdapterV2.ts, where the provider's result text is preserved, and trace how a persistent Claude subprocess is reused across resumed helper sessions. The issue has no deterministic reproduction, so first confirm the 403 path by reading the result handling around is_error and api_error_status. Done means a reproducible cause or a clear fix for stale process reuse, with a test covering it.

由索引模型根据 Issue 内容生成。

描述

via-triage
What happened

During normal work in the T3 desktop app, an existing Claude thread and its two helpers repeatedly failed with: Failed to authenticate. API Error: 403 Attention Required! | Cloudflare. Sending continue resumed the helpers but the error returned. Expected: the existing helpers continue working, or the UI explains how to recover.

Diagnosis

The provider event log records Claude SDK results with is_error=true, api_error_status=403, terminal_reason=api_error and the exact displayed text. T3 preserves this provider text in ClaudeAdapterV2.ts (commit 40526f96f6ae3dfe74ae60ca279e748d7838a691, lines 11075 onward); the UI labels it Provider error. The wording does not establish expired credentials; see PR #15872. Both existing helpers recovered after the persistent Claude process was restarted. This does not distinguish stale process state from a transient upstream Cloudflare block. No deterministic reproduction or confirmed root cause was established.

Steps to reproduce

Observed sequence, not a deterministic reproduction:

  1. Use an existing Claude Opus 5.5 thread with two helpers in the desktop app against its local server.
  2. Send continue; the parent resumes the existing helpers with SendMessage.
  3. Observe repeated Cloudflare 403 errors.
  4. Restart the affected persistent Claude process and resume the same helpers for diagnostic replies and harmless tool calls; both checks succeed.
Version

0.1.10-nightly.20261112.4580

Environment

Linux x64, kernel 7.2.15-1-omarchy; local desktop-managed server; Node v26.12.1; Claude Code 2.1.322; claudeAgent provider; claude-opus-5-5, high effort. Triage CLI only: 0.1.9 (npx t3 triage).

Evidence
Provider log excerpts/summary, 2026-11-13 UTC:
09:51:49.461 result: subtype=success, is_error=true, api_error_status=403, terminal_reason=api_error
result: Failed to authenticate. API Error: 403 Attention Required! | Cloudflare
09:54:28.668 and 09:54:30.513: both existing helpers replied HELPER_AUTH_OK
09:55:02.098 and 09:55:03.027: both helpers replied HELPER_TOOL_OK after successful Bash tool results
No further Cloudflare error through the inspected log at 09:56:36.219. This is a bounded recovery check, not proof against recurrence.
Related issues

#18554 is analogous process-reuse/auth behavior, but reports Not logged in after re-login rather than Cloudflare 403; not a confirmed duplicate. #15872 explains preserving Claude's 403 text.

Fix applied or workaround

A separate diagnostic session verified that the affected thread had no active run, checked the Claude subprocess owner and session identity, and sent that idle subprocess SIGTERM at 09:54:10 UTC. A new turn resumed the same conversation and helpers. Both passed reply-only and tool-call checks. Recovery after restart does not prove causation. No T3 source patch was applied.

Filed by

Codex GPT-6 Astra

主要语言
TypeScript
星标
24.8k
派生
6.4k
平均合并
59 分钟
30 天内合并 PR
418

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

pingdotgg/t3code 的其他 Issue

查看 pingdotgg/t3code 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。