Repository navigation
Mingbird v2.0.3
鸣鸟 Mingbird v2.0.3
🇨🇳 中文在上 / English below.
来自第一份外部深度测试报告的修复(issue #3,Fedora/Podman,qwen3.5 2B/4B/9B)。测试 508 项全绿。报告人 @kinosho-ichigo —— 感谢这份极其扎实的现场取证,本版大部分修复由它驱动。
修复
- finish 门拒绝词现在直接告诉 agent 工作区在哪:模型把"工作区"猜成主目录或幻觉的
/workspace,产物写错位置后对着核对门无限打转。三道 finish 门的拒绝词全部附上绝对工作目录路径与相对路径模板。 .bak安全网补上 shell 就地改写的大洞:sed -i、>/>>重定向、tee、cp/mv等此前完全绕开备份。现在改写型命令执行前,其点名的既有工作区文件逐个快照为<file>.bak,输出注明备份;README 措辞同步对齐。- 只读任务可以用正文收尾:finish 附带实质回答不再触发"无产物"拦截——分析/讲解类任务(读文件→讲解)以文字交付即完成。裸 finish 仍被拒,拒绝词新增文字交付出路。
AGENT_LANG=en强制英文应答(英文界面自动设置):中文优先的系统提示会让 Qwen 系在英文会话里跟风输出中文。默认行为不变。
新增
AGENT_APITIMEOUT超时旋钮(秒,默认 240),作用于流式打开超时——高负载+大上下文的机器可能超掉;非流式本来就是 900 秒。- todo 工具标注
index is 1-based:三个尺寸的模型都先猜 0-based 白烧轮次,schema 提前一句省掉。
English
Fixes from the first external deep-dive bug report (issue #3 — Fedora/Podman, qwen3.5 2B/4B/9B). Tests: 508 green. Reported by @kinosho-ichigo — thank you for exceptionally solid field forensics; most of this release is driven by it.
Fixes
- Finish-gate rejections now tell the agent where the workspace is: models guessed the home directory or a hallucinated
/workspace, wrote artifacts there, and looped against the artifact gate forever. Every finish-gate rejection now includes the absolute workspace path plus a relative-path template. - The
.baksafety net now covers in-place shell edits:sed -i,>/>>redirection,tee,cp/mvetc. previously bypassed backups entirely. Mutating commands now snapshot every existing workspace file they name to<file>.bakbefore running; README wording aligned. - Read-only tasks can finish on their text: a finish carrying a substantive answer no longer trips the no-artifact guard — analysis/explanation tasks (read → explain) complete on their content. Bare no-work finishes are still rejected.
AGENT_LANG=enforces English responses (set automatically by the English UI): the Chinese-first system prompt made Qwen models mirror Chinese in English sessions. Default behavior unchanged.
New
AGENT_APITIMEOUTenv knob (seconds, default 240) for the streaming-open timeout; non-stream calls already had 900 s.- Todo schema declares
index is 1-based— all three model sizes guessed 0-based first.
Install / 安装
Windows:下载 Mingbird-v2.0.3-CN-Setup.exe(中文)或 Mingbird-v2.0.3-EN-Setup.exe(English),双击安装。需要 Ollama;2~9B 小模型核显即可跑,全程数据不出本机。
Windows (EN): download Mingbird-v2.0.3-CN-Setup.exe (Chinese UI) or Mingbird-v2.0.3-EN-Setup.exe (English UI) and run the installer. Requires Ollama; 2–9B models run on integrated graphics, and your data never leaves the machine.
macOS / Linux(实验性):下载 mingbird-v2.0.3-macos-arm64.tar.gz / mingbird-v2.0.3-macos-x64.tar.gz / mingbird-v2.0.3-linux-x64.tar.gz,解压后运行 LocalAgent。系统语言自动识别(中文系统直接显示中文;英文系统输出自动跟随英文。也可在可执行文件旁建 app_lang.txt 写 zh/en 强制)。
macOS / Linux (experimental, EN): download the tar.gz for your arch, extract, run LocalAgent. System language is auto-detected (English systems get English output); you can also force it by placing an app_lang.txt (zh/en) next to the executable.
From source:pip install -r requirements.txt → python agent_gui.py(Python 3.12 推荐)。 / From source (EN): pip install -r requirements.txt → python agent_gui.py (Python 3.12 recommended).