Skip to content
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -452,7 +452,7 @@ The proxy checks npm on startup and every 3 minutes; a newer version is installe

While the running process is behind the on-disk install ("stale"), the web UI shows a banner (running vs installed version, auto-restart state) and `GET /__bili/status` returns `{version, diskVersion, stale, autoRestartOnUpdate, advisory, updatePolicy, inFlight}` for scripting (`advisory` is the active critical-defect entry or `null`, see below; `updatePolicy` reports the #3619 rollout state — policy source, `hold`/`pinned`/`blocked`, and the pending staggered upgrade with its due time). Disable permanently via `"autoUpdate": false` in config or `ACP_AUTO_UPDATE=0`.

In **plugin mode** (`bili opencode` / `bili pi` / `bili dsh` — the proxy runs embedded in a long-lived host), "restart bili" means restarting the **host**: the embedded process cannot exit on its own, so a host left open keeps serving whatever code it started with, silently missing every fix auto-update reports as installed (#1603). Enable `--auto-restart-on-update` to have it self-restart at a safe point, or run `/acp` (OpenCode hosts today — pi/dsh coverage tracked in #2084) — the status panel appends a one-line staleness warning (running vs installed version) whenever the on-disk install is ahead of the live process. Auto-update only ever replaces disk files and (optionally) the proxy process; **injected code inside the host process only takes effect after the host itself restarts** — so if the injection breaks at runtime mid-session (a disposed agent scope, a disabled plugin), model traffic goes direct and uncompressed until then. Since #3581 this state is visible instead of silent: the web UI shows a "takeover broken" banner (agent + refusal count) whenever a plugin reports its attribution gate degraded while no session is live, and `GET /__bili/status` / `/__bili/overview` expose the per-agent `attribution` state (`ok`/`none`/`threw`, refusal count, first-refusal timestamp).
In **plugin mode** (`bili opencode` / `bili pi` / `bili dsh` — the proxy runs embedded in a long-lived host), "restart bili" means restarting the **host**: the embedded process cannot exit on its own, so a host left open keeps serving whatever code it started with, silently missing every fix auto-update reports as installed (#1603). Enable `--auto-restart-on-update` to have it self-restart at a safe point, or run `/acp` (OpenCode hosts today — pi/dsh coverage tracked in #2084) — the status panel appends a one-line staleness warning (running vs installed version) whenever the on-disk install is ahead of the live process. Auto-update only ever replaces disk files and (optionally) the proxy process; **injected code inside the host process only takes effect after the host itself restarts** — so if the injection breaks at runtime mid-session (a disposed agent scope, a disabled plugin), model traffic goes direct and uncompressed until then. Since #3581 this state is visible instead of silent: the web UI shows a "takeover broken" banner (agent + refusal count) whenever a plugin reports its attribution gate degraded while no session is live, and `GET /__bili/status` / `/__bili/overview` expose the per-agent `attribution` state (`ok`/`none`/`threw`, refusal count, first-refusal timestamp). Background lanes that are agentless **by design** (e.g. dsh's title-generation sidecar) never carry an initiator, so their small one-shot calls go direct by definition; under the #3651 contract these normal interceptions are NEVER reported and therefore can never raise the banner. The banner fires only when the attribution lookup itself throws (an internal host error, e.g. a disposed agent scope), and then names what failed: the thrown error text, the masked endpoint, when the episode started, how many requests went uncompressed, the remedy (restart the host process), plus copy/submit-diagnostics buttons that prefill a GitHub issue.

**Update visibility is silent by default since v0.1.183 (#1977):** routine/recommended releases no longer render an update line on the `/acp` panel footer or in `acp_status` — only a `critical`-tier release surfaces there (`CRITICAL UPDATE READY/AVAILABLE`). The standing ways to check an installed-but-not-restarted release are the stale cues above: web UI banner, the `/acp` one-line warning, and `GET /__bili/status`. Deliberately no always-on nag; the advisory channel (#1481) remains the only auto-visible critical path.

Expand Down
2 changes: 1 addition & 1 deletion README.zh-CN.md
Original file line number Diff line number Diff line change
Expand Up @@ -337,7 +337,7 @@ bili --no-auto-update # 本次启动禁用自动更新

**升级撒点(#3619):** 发现新版本后,自动更新不再立即应用。每台实例在延迟窗口内随机抽取**一个**时刻(内置默认 **15 分钟 – 48 小时**),到点再升;抽中的时刻落盘持久化,重启不重摇。全量升级因此摊开在整个窗口内,而不是发版后十几分钟内全体升完,给坏发版留出中途拦截的响应窗(#3401/#3600/#3394)。窗口与紧急控制放在一个公开的策略文件里(本仓库的 `auto-update-policy.json`,匿名 GET + ETag 缓存):`hold` 全局冻结自动升级、`blocked` 跳过指定版本、`pinned` 设置升级天花板。按设计 fail-open——文件不可达或格式非法时直接回退本地默认;而且它只能调节奏/跳过/钉住,**永远控制不了装什么**。本机退出方式:手动 `bili update` 绕过 `hold`/撒点(但仍遵守 `blocked`/`pinned`)、`"staggerMinDelayMs": 0` + `"staggerMaxDelayMs": 0`(或 `BILI_UPDATE_STAGGER_MIN_MS`/`BILI_UPDATE_STAGGER_MAX_MS`)恢复旧版立即升级、`ACP_AUTO_UPDATE=0` 完全停用自动更新。完整格式与语义见 [CONFIGURATION.zh-CN.md → `update`](CONFIGURATION.zh-CN.md#update)。

**插件模式**(`bili opencode` / `bili pi` / `bili dsh` —— 代理内嵌在长驻宿主里运行时),"重启 bili"指的是重启**宿主**:内嵌进程无法自行退出,所以一直开着的宿主会持续运行它启动时的那份代码,默默地错过自动更新报告为"已安装"的每一处修复(#1603)。启用 `--auto-restart-on-update` 可让它在安全点自重启;或运行 `/acp`(当前仅 OpenCode 宿主,pi/dsh 覆盖见 #2084)—— 当磁盘安装领先于运行进程时,状态面板会追加一行过期警告(运行版本 vs 已安装版本)。
**插件模式**(`bili opencode` / `bili pi` / `bili dsh` —— 代理内嵌在长驻宿主里运行时),"重启 bili"指的是重启**宿主**:内嵌进程无法自行退出,所以一直开着的宿主会持续运行它启动时的那份代码,默默地错过自动更新报告为"已安装"的每一处修复(#1603)。启用 `--auto-restart-on-update` 可让它在安全点自重启;或运行 `/acp`(当前仅 OpenCode 宿主,pi/dsh 覆盖见 #2084)—— 当磁盘安装领先于运行进程时,状态面板会追加一行过期警告(运行版本 vs 已安装版本)。若注入代码在运行期损坏(如 agent scope 被销毁),模型流量会静默直连(不压缩),直到宿主重启;自 #3581 起该状态可见:无会话运行时插件上报归属闸门降级,Web UI 显示「接管已失效」横幅,`GET /__bili/status` / `/__bili/overview` 暴露每代理的 `attribution` 状态。**按设计无归属**的后台车道(如 dsh 标题生成 sidecar)本身不携带 initiator,其小体量一次性请求按定义直连——按 #3651 契约这类正常拦截**永不报告**,因此永远不会触发横幅。横幅只在归属查询本身抛错(宿主内部错误,如 agent scope 被销毁)时出现,并给出可诊断信息:抛出的错误文本、脱敏后的端点、起始时间、已直连的请求数、解决方法(重启宿主进程),以及复制/提交诊断按钮(预填 GitHub issue)。

**自 v0.1.183 起,更新提示默认静默(#1977):**routine/recommended 版本不再在 `/acp` 面板脚注或 `acp_status` 中渲染更新行 —— 只有 `critical` 级别的版本会在那里显示(`CRITICAL UPDATE READY/AVAILABLE`)。检查"已安装未重启"版本的常设途径即上文所列的过期线索:Web UI 横幅、`/acp` 单行警告、以及 `GET /__bili/status`。刻意不设常驻提醒;严重缺陷公告通道(#1481)仍是唯一自动可见的严重路径。

Expand Down
Loading
Loading