403 Forbidden mid-session with no cause given; "continue" retry prompt (CLI 1.86.3, Windows)

未关闭
  • 哪里不对？
#1,054 3 条评论 0 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
1-2 天
新手友好度
48/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
领域
cli

调研方向

Start by tracing where the CLI handles HTTP 403 responses and displays the retry prompt; the issue does not name a file or test. Compare this generic response with the model-specific errors mentioned in #930 and #736, then check whether the trace ID or session sequence narrows the cause. Done means the denial reason is surfaced clearly when available and the behavior is covered by a test.

由索引模型根据 Issue 内容生成。

描述

Summary

Command Code stopped mid-session with a bare 403 Forbidden and the prompt Type "continue" to try again. The error gives no cause (auth, plan, quota, permissions), so there is no way to tell what to fix.

⚠ Error: 403 Forbidden

Type "continue" to try again. If the issue persists, contact support: https://commandcode.ai/discord
Trace ID: ccd031590f1d2d9d981332b79202560a

This is different from #930 and #736. Those return 403 "Authentication failed" for specific models. Here the message is a plain 403 Forbidden with no model-specific text.

Expected Behavior

The request succeeds, or the error states why access was denied.

Actual Behavior

A bare 403 Forbidden with only a trace ID.

Steps to reproduce the issue

Updated from the session transcript (see the comment below for the raw evidence). This is not a one-off: 18 occurrences in one session.

  1. Start a session with a long tool-heavy conversation.
  2. Either send a prompt, or complete a turn that runs tool calls and returns tool results.
  3. The request immediately following that point 403s.

The transcript shows the error arriving at two specific points in a turn's lifecycle and not others:

  • at the same timestamp as an outgoing user prompt, before any assistant turn is produced, or
  • within ~1s of an assistant turn finishing its tool calls and returning tool results.

Turns in between succeed, so it is intermittent rather than a dead credential or a plan limit. A full turn can complete normally and the very next request 403s.

Confirmed: continue clears it and preserves the full conversation context. That held on all 18.

Not confirmed: what the two landing points have in common at the request level. My earlier note here suggested a 60s shell timeout was the trigger; the transcript does not support that, and I have removed it as a lead.

Command Code Version

1.86.3

Operating System

Windows 10.0.26200 (win32)

Terminal/IDE

WezTerm

Shell

MSYS2 UCRT64 bash

Model

xiaomi/mimo-v2.6-pro

Additional context

First seen 2026-10-21T03:01:18Z (11:01 MYT), last seen 2026-10-21T06:18:40Z. 18 injected events across 3h17m, each with its own trace ID, so the trace ID appears to be per-request. Failures cluster: two arrived 17 seconds apart, and six arrived inside the final three minutes.

All 18 trace IDs are in the comment below.

Also seen once in the same session: Error: 400 Request failed at 2026-10-21T23:20:29Z, trace 58ad3b29bc7a, in exactly the same shape (injected into the transcript as a role: user message rather than raised as an assistant or tool error). Possibly the same upstream path returning a different status.

主要语言
没有语言数据
星标
4.2k
派生
377
平均合并
1 天 17 分钟
30 天内合并 PR
7

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

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。

CommandCodeAI/command-code 的其他 Issue

查看 CommandCodeAI/command-code 的全部 Issue

相似的 Issue

更多 CLI Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。