v0.23.2

改进

• 「来源」卡片顶部只放正在用的来源（已登录且打开）；没登录的、关掉的都收进折叠列表，平时看不见。
• README 和简介改为先讲三件事：📅 每日签到、🔁 一个地址中转所有模型、🧩 免费和订阅模型来源。

Changes

• The Sources card lists only sources in use (signed in and on); signed-out and turned-off ones fold away.
• README and descriptions now lead with the three things: daily check-in, one address relaying every model, and free / subscription model sources.

v0.23.0

改动：WorkBuddy 只用来签到——它会认出 dsh 发出的对话并拦下，所以新装时它的模型默认不接进 dsh（签到照常；以前装过的保持原样）。从页面打开 WorkBuddy 前会先提醒一次。「开始使用」向导只推荐两家：硅基流动（国内直连）和 Gemini（能上外网）。

Change: WorkBuddy is used for check-in only — it blocks conversations sent by dsh, so a fresh install no longer connects its models (check-in unaffected; existing installs keep what they had). Turning it on from the page first explains why. The getting-started guide suggests only SiliconFlow (reachable from China) and Gemini.

v0.22.3

改进：写明 WorkBuddy 会认出 dsh 发出的对话并拦下（国内版、国际版都是）：「最近失败」的建议不再推荐改用国内版；「测一下」WorkBuddy 的模型会注明"只是一句短测试"；来源行和 README 加了提醒。

Change: say plainly that WorkBuddy blocks conversations sent by dsh (China and international): recent-failure advice no longer suggests WorkBuddy China, a Test on a WorkBuddy model notes it is only a short test, and the source rows and README carry the warning.

v0.22.2

改进：「最近失败」能认出"被服务商拦截"：WorkBuddy 国际版的「请求被拦截」、国内版的 11-128（未经批准的渠道）。会写明原因：服务商在限制官方客户端之外的使用，请改用其他来源或建组合，继续调用有封号风险。组合切换遇到这种拦截也会换下一个模型。

Change: recent failures now recognise "blocked by the provider" (WorkBuddy AI's 请求被拦截, WorkBuddy China's 11-128 unapproved channel) and say why: the provider is restricting use outside its app, so switch source or use a combo; continuing risks the account. Combos move on past this error too.

v0.22.1

修复：重装后 dsh 里没有模型——dsh 凭据里留着上次安装的 key（dshm_…），新装的 dsh-model 认不出来，停在"已有 DSH_MODEL_API_KEY，且不是 dsh-model 写的"。现在 dsh-model 格式的旧 key、以及上次写的 dsh-model 模型分组会直接接管（并提示一句）；别人写的同名 key 照旧不覆盖。

Fix: after a reinstall dsh showed no models, because the previous install's key (dshm_…) was still in dsh's credentials and the new install refused to touch it. A key in dsh-model's own format, or a dsh-model provider left by an earlier install, is now taken over with a notice; anything else is still refused.

v0.22.0

新功能

• 组合切换：把几个模型（任意来源，订阅也行）按顺序排成一个组合，在 dsh 里就是一个模型 combo/<id>：先用第一个，限流、额度用完或出错就自动换下一个。设置页「组合切换」卡片可新建、编辑、调顺序；命令行 dsh-model combo。
• 额度什么时候恢复：套餐额度用完时，来源行标出「已用完 · X 后恢复」；WorkBuddy 显示最近一批积分的到期时间（3 天内标黄）。
• Qoder 令牌说明改为「设置 → API 令牌」。

New

• Combos: put several models (any source, subscriptions too) in order and use them as one model, combo/<id>; the next one takes over on a rate limit, used-up quota or error. Settings card to create, edit and reorder; dsh-model combo on the command line.
• When quota comes back: a used-up plan shows "Used up · back in X"; WorkBuddy shows when the next batch of credits expires (yellow within 3 days).
• Qoder token instructions now point to Settings → API Tokens.

v0.21.5

文档：README 改为先讲能得到什么（一个地址用上所有模型、免费额度、每日签到、自动切换…），补上签到功能和 checkin 命令；更新 npm 简介和关键词。

Docs: the README now leads with what you get (one address for all your models, free tiers, daily check-in, failover…), and lists check-in and the checkin command; npm description and keywords updated.

v0.21.4

修复：新增、吊销、轮换 key 后会确认引擎真的换上了。引擎偶尔会漏掉一次配置变化（新 key 用不了，或吊销的 key 还能用）；现在没生效会自动再写一次配置，还不行就重启引擎。

Fix: adding, revoking or rotating a key now confirms the engine picked up the change. The engine occasionally missed a config change (a new key rejected, or a revoked key still accepted); now the config is written again if needed, then the engine is restarted as a last resort.

v0.21.3

改进：安装（setup）时不再询问 OpenCode Zen 的 key。需要时在 dsh「设置 → dsh-model → 来源」里打开它（或执行 dsh-model opencode key）；已经填过的 key 照常刷新模型。

Change: setup no longer asks for the OpenCode Zen key. Turn it on from dsh Settings → dsh-model → Sources when you need it (or run dsh-model opencode key); an existing key is still refreshed.

v0.21.2

修复：macOS 把 dsh-model 的后台服务关掉后，「修复」和 setup 会自动重新允许并启动，并确认服务真的跑起来了；以前会把加载失败当成成功，页面一直显示「服务没在运行」。

Fix: when macOS has turned off dsh-model's background services, Repair and setup now allow them again and start them, and check that they really loaded. Before, a failed load counted as success and the page kept saying the service was not running.