3.7 KiB
Anonymous customer connectivity checks
/diagnostics is a public page independent of account/plugin initialization and browser storage. The login form has no diagnostic link; the platform/task entries and direct URL remain available.
The page now runs the same three checks whether the customer is logged in or not:
- A tiny fixed
/diagnostics-probe.jsonresource with an expected marker. /api/health, verifying the expected healthy JSON response./api/auth/bootstrap-status, verifying the public authentication-route response shape.
Every request is a fixed GET with credentials: omit, no-store and a cache-busting parameter. No /auth/me, account, project, task, timeline or customer-media request is made. An unsigned session is not a fault and does not block any check. HTTP 200 alone is not sufficient: an HTML fallback or unrelated JSON fails the content check. Each attempt is bounded to eight seconds and 64 KiB; transient errors retry once and successful recovery is displayed as normal, with both attempts retained in the report. Cancellation stops remaining checks.
Customer UI has only start/stop, a concise result and report download/copy. A successful result says 基础连接正常, not that credentials or all business functions have passed. Technical details are confined to the downloadable report: timestamps, origin, frontend/API commit, browser identification, status, latency, attempt outcome and request IDs. Raw responses, account names/phones, credentials, task identifiers and private/signed video URLs are not exported. Reports are not automatically uploaded.
No browser-version allowlist is used. HTTP, unavailable Web Locks, unavailable clipboard and blocked local storage do not mark connectivity as abnormal. Clipboard has download/manual-copy fallbacks; non-streaming fetch responses are supported. Missing essential checking APIs is reported as an incomplete check, not a proven network failure. Passing on one browser does not certify every old browser engine can render the application.
Scope limitations are recorded in the report: account/password validation, project permissions, private BOS video storage and concurrent capacity are not tested by this anonymous check. The authentication-route probe establishes reachability of the public auth route; it does not attempt login or prove that a specific password is correct. A genuine anonymous BOS video-path test would require a dedicated non-customer test object, separately approved before any storage upload. Do not expose customer videos or call a platform-hosted sample a successful BOS test.
Request tracing and the existing authorized minimal task-diagnostic API remain unchanged. Diagnostic-marked API requests and login attempts receive server-generated request IDs and sanitized log lines. No database mutation, account changes, task claiming, submission or load test is performed.
Validation uses synthetic browser responses and node tests: cookie-free requests even with an existing browser session, identical public checks without login/task gates, schema validation, bounded retries/timeouts/cancellation, secret-free export, ordinary HTTP with optional capabilities missing, clipboard fallback, compact layout and real outage reporting. Deployment and PR creation require separate user approval.
本地验证(2026-09-26)
- 免登录诊断单元测试 7 项通过;浏览器验证覆盖无登录、禁止存储、缺少可选功能、真实失败及检查能力缺失。
- 前端类型检查、生产构建通过。完整前端测试 619 项,606 项通过,13 项既有基线失败(均不属于诊断模块)。
- 本次仅本地修改,未部署公司集群、未改动客户数据。