v0.12.1 — Stripe referrals and capture safety

Changes

• Add incoming/outgoing Similarweb referral report capture with the sw_referrals extractor, source periods and cross-page site deduplication.
• Include capture completion time for daily history.
• Fail captures that exceed the parsed-data size limit instead of uploading partial reports. Normal captures ending at the configured page limit remain supported.
• Keep extension permissions, pairing and the existing queue/pager framework. The source response body limit increases from 3,000,000 to 6,000,000 characters; the parsed-result limit remains 3,800,000 characters.

Installation / 安装

Download gefei-seo-agent-0.12.1.zip, rather than the automatically generated source archives. SHA256SUMS contains its checksum.

1. 在插件侧边栏取消勾选“接远程任务”，等当前任务执行完毕。
2. 解压ZIP，将其中gefei-seo-agent文件夹内的内容覆盖到原插件加载目录；manifest.json仍应直接位于原目录内。
3. 在chrome://extensions找到原插件，点击“重新加载”。保留同一个扩展条目。
4. 确认版本为0.12.1，NEW连接、Similarweb地址和来源权限正确，再勾选“接远程任务”。先确认原有任务正常。
5. 如升级异常，暂停接任务，恢复已本地保存的旧版文件并重载，核对连接后再恢复。

新版可先安装验证原有兼容任务。新收入及每日历史任务需要配套NEW后端部署；安装此包本身不会开启这些任务。NEW数据库和后端、SEO API属于独立发布步骤。旧插件只能继续它支持的任务，不能执行新版收入/每日任务。

Validation

npm test, JavaScript syntax checks and the full PR base/head whitespace check pass. A separate read-only reviewer checked the complete PR and ran local tests. The repository has no configured CI check runs.

The capture tests include normal 10/20-page results and size-limit rejection. Earlier Haze Chrome testing used local synthetic pages and an isolated queue/database, including twenty pages and reload/pairing preservation. Real-source twenty-page capture and acceptance on the always-on Mac remain rollout checks.