Files
UHD_Media_Tool/packages/adapters/src/browser
rootandClaude Opus 5.5 1fcfcac885 feat(browser): keep sessions alive during desktop, takeover and agent use (AG-4)
camofox reaps a tab after TAB_INACTIVITY_MS (5 min) without API tool calls;
desktop-mode input goes straight to Selkies and screenshots don't count, so
tabs disappeared under the user during takeover.

- BrowserKeepalive (in-process): sessions held by an open desktop WebSocket
  (ref-counted), a human takeover lease, or a running/queued agent run that
  used browser.* get a keepalive every BROWSER_KEEPALIVE_MS (default 120s,
  0 = off). Holds are re-checked each round; release/lease expiry/agent end/
  session close stop it and camofox reaps as usual.
- Keepalive call: camofox 1.17 has no keepalive/touch endpoint and /stats does
  not bump toolCalls; use GET /tabs/{id}/downloads without includeData/consume
  (read-only metadata). Goes through BrowserService.run so restarts or reaped
  tabs self-heal via BR-3 reconnect; no lastAutomationAt, no control needed.
- Runbook browser.md §12, deploy env table, compose env; tests against the
  testkit mock camofox.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-09-30 18:05:12 +08:00
..

browser

BrowserProvider 合同(provider.ts)及其实现:业务侧访问浏览器服务只走这套接口。产品浏览器能力对接用户已有的 camofox-browser;Playwright 只用于前端截图测试,不是产品浏览器能力。设计说明见 docs/design/browser-provider.md。

  • camofox/:camofox-browser 适配器。按 jo-inc/camofox-browser v1.17.0 的 openapi.json、README 与 server.js 实现,2026-09-28 对真实容器逐项核验(docs/runbook/browser.md §7;核验表 camofox/capabilities.ts 的 CAMOFOX_VERIFIED_RELEASES),Authorization: Bearer <CAMOFOX_ACCESS_KEY>,凭据每次请求经 BrowserSecretResolver 解出,不缓存、不入日志。camofox userId = 本项目 profileKey(站点/账号隔离),sessionKey = browser_session.id。
  • none.ts:默认实现(未配置浏览器),能力全关,动作报 unavailable。
  • redact.ts:redactSnapshot / redactBrowserUrl,快照与地址进入模型前遮蔽密码框值、Cookie 值、含 passkey/token 的 URL。
  • profile-lock.ts:同一 profile 串行锁(进程内内存实现;多进程部署换持久化实现,接口不变)。
  • budget.ts:导航域名白名单(含子域),非 http(s) 拒绝。

已实现(C1-E;BR-1 对 v1.17.0 真机核验并修正实测差异)

  • 会话:openSession(POST /tabs)、closeSession(DELETE /sessions/{userId})、listSessions(内存镜像;服务层以数据库为准)、restoreSession(重启后按 DB 恢复,旧引用失效)。
  • 标签页:openTab、closeTab(DELETE /tabs/{id})、reconcileTabs(GET /tabs?userId,上游已回收的标 lost)。
  • 导航:navigate(白名单 / 超时 / 预算内重试 5xx·超时·网络)、history(/back /forward /refresh)。
  • 读取:snapshot(字节预算截断 + 脱敏)、screenshot(JSON base64 或 image/*;登录页 / 含凭据地址 modelSafe: false)、extract(POST /extract,需先快照)。
  • 交互:click(ref / selector / 视口坐标)、type(明文或凭证引用,凭证经 BrowserSecretResolver 解出后直接下发)、select、scroll、press。
  • 下载:waitForDownload 轮询 GET /tabs/{id}/downloads?includeData&consume&maxBytes,字节交 DownloadSink 暂存、算 sha256,只返回引用。
  • 控制权:TabRef.as 标明发起方(automation / human),与会话 controller 不符报 control_conflict;setController 任何变更都让旧 eN 引用失效(stale_ref);acquireControl / releaseControl 为内存版,服务层用 browser_control_lease 持久化。

未核验的上游行为(坐标点击、无目标输入、scroll / press / back 等端点、downloads 字段形状)见 docs/runbook/browser.md 第 6 节。