Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

[bug] vision describe: default base URL is the Coding Plan endpoint → HTTP 404 for all VL/OCR models

未关闭 适合新手
#241 0 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
2/5
预计耗时
1-3 小时
新手友好度
78/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
typescript

调研方向

找到 vision describe 命令,并比较它与其他 API 密钥命令解析基础 URL 的方式。跟踪 --base-url、DASHSCOPE_BASE_URL 和所选配置文件的 base_url 对应的端点选择;运行 vision 命令测试。完成条件是使用已配置的端点,或者默认端点指向提供 VL/OCR 模型的公共端点。

由索引模型根据 Issue 内容生成。

描述

Environment

  • CLI: bl 2.1.0
  • Skill: 2.1.0
  • Node: v22.23.3
  • OS: Linux 5.15.0-198-generic x86_64 (Ubuntu)
  • Region: cn (domestic)

Reproduce

bl vision describe without an explicit --base-url posts to
https://coding.dashscope.aliyuncs.com/compatible-mode/v1/chat/completions,
which does not serve VL/OCR models, so the request always 404s.

# fails
bl vision describe --image ./test.png --model qwen3.5-ocr --prompt "Output the text" --verbose

# identical command + explicit public endpoint succeeds
bl vision describe --image ./test.png --model qwen3.5-ocr --prompt "Output the text" \
  --base-url https://dashscope.aliyuncs.com

Expected

bl vision describe should resolve its endpoint the same way other API-key
commands do (--base-url → DASHSCOPE_BASE_URL → selected profile base_url),
or fall back to a public DashScope endpoint that actually serves VL models.

Actual

Every call without --base-url fails with HTTP 404, for every model tried
(qwen3.5-ocr, qwen-vl-ocr, qwen3-vl-plus). The same command with
--base-url https://dashscope.aliyuncs.com exits 0 and returns correct OCR text.

The configured profile's base_url is a private workspace endpoint
([REDACTED_URL]), and vision does not use that either — so neither the
profile setting nor a working default is applied.

Full output

> POST https://coding.dashscope.aliyuncs.com/compatible-mode/v1/chat/completions
> x-dashscope-source-config: {"channel":"bailian-cli","tags":{"t1":"public","t2":"bl","t3":"2.1.0"}}
< 404 Not Found

Error:       HTTP 404
Status:      HTTP 404
Exit code:   1

JSON error

{
  "error": {
    "code": 1,
    "message": "HTTP 404",
    "http_status": 404
  }
}

Already tried

  • bl update to 2.1.0; skill version aligned with the CLI
  • bl auth status healthy (API key + console gateway configured)
  • Reproduces on every call (100%), on several models, and also on a plain
    320×90 test image with no document content
  • --base-url https://dashscope.aliyuncs.com works every time

Notes

  • Frequency: always
  • Invoked via: terminal and agent (DeepSeek Harness)
  • Impact: the error carries no hint about the endpoint, so it reads like a model
    or auth problem. Users cannot self-diagnose without --verbose, and the
    documented bl model search catalog lists these models as available. This
    breaks the entire VL/OCR surface of the CLI for anyone who does not pass
    --base-url explicitly.
主要语言
TypeScript
星标
527
派生
33
平均合并
6 小时 43 分钟
30 天内合并 PR
23

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

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。

modelstudioai/cli 的其他 Issue

查看 modelstudioai/cli 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。