mirror of
https://git.openapi.site/https://github.com/desirecore/market.git
synced 2026-09-05 20:03:43 +08:00
feat(web-access): v3.0 内置浏览器能力面同步——删除 Python Playwright 回落（S31） (#85)
## 变更内容 / What
浏览器升级计划 S31 最终验收信号：web-access 技能与内置受管浏览器新能力面对齐，**删除「用户手工启动调试 Chrome +
Python Playwright」回落路径**。
- **删除回落路径**：Prerequisites: Chrome CDP Setup、Layer 3 CDP Browser（Python
Playwright 模板）、Installation Note（pip install
playwright）整段删除；references/cdp-browser.md 文件删除；jina-reader.md 的 CDP 引用改为
page.extract-text
- **订正陈旧断言**：
- 「没有批量取文通道」→ BrowserSnapshot mode:text /
page.extract-text（maxBytes/cursor 分页，超出截断给 nextCursor）
- 「page.evaluate 基本不可用」→ 返回真实值（expression/awaitPromise，超预算截断标
truncated）；仍走人工闸门
- 「截图前必须 tab.activate / 串行截图 / BROWSER_TAB_HOST_NOT_FOUND」→ S36 订正：Agent
单标签会话免 activate；多标签后台 tab 秒级报 BROWSER_VIEWPORT_UNAVAILABLE；命令超时只 stop 不
close，标签页可重试
- 「accessibility 超限即失败」→ 尊重 depth + maxBytes 截断翻页（S8）
- 「只有整页截图」→ clip{x,y,width,height,scale≤4} + captureBeyondViewport
- **provides.tools 加 BrowserScript**（code-mode；信任级别等同 Bash）
- **新增选用规则（D4 唯一约束机制）**：反检测站点一律优先 input.*（#1808 输入拟真 +
身份一致性）；page.element 写类仅用于表单批量填充等站点不检测场景；JS 直调 el.click() 为禁止回退
- **新增 fetch.browser 配方**：page.evaluate 页面上下文跑 fetch（带 origin Cookie、同
origin、受 Grant origins 约束）——登录态取站内接口的正解
- **能力速查**：page.element 九 op / page.wait 九 until / inline wait 块 / loc=
方言 / BrowserScript / 跨源 iframe 快照（S35）
- **版本** 2.2.1 → 3.0.0（删除回落层为 breaking）；source_hash
重算；required_client_version 维持 10.0.98（新能力在正文标注 10.0.112+）
## Why
v2.x 时代回落路径存在的每一条理由（无批量取文、evaluate 不可用、截图必须串行 activate）均已被
S2–S14/S35/S36 覆盖；文档继续引导用户手工起调试 Chrome 会误导新 Agent 走已废弃路径。
双语同步修改（SKILL.md / SKILL.zh-CN.md heading 数一致，i18n-validate 通过）。
- [x] CLA
This commit is contained in:
GitHub
@@ -1,26 +1,15 @@
|
||||
# 内置受管浏览器工具速查（L3-fast）
|
||||
# 内置受管浏览器工具速查（L3）
|
||||
|
||||
> v2.1 起，本层从「cdp-proxy 驱动用户自己的 Chrome」改为「DesireCore 内置受管浏览器」。
|
||||
> v2.1 起本层从「cdp-proxy 驱动用户自己的 Chrome」改为「DesireCore 内置受管浏览器」；v3.0 补全能力面（批量取文 / 元素操作 / 等待 / 代码模式）并移除 Python Playwright 回落。
|
||||
> 每个任务跑在独立 BrowserSpace 里（Cookie / Storage / 缓存互不串扰），每个动作都经过
|
||||
> Capability → Grant → Lease → Origin → Host fencing 校验并留下可审计回执。
|
||||
>
|
||||
> 要求客户端 **v10.0.98+**。旧的 `BrowserListTabs` / `BrowserNavigate` / `BrowserEval` /
|
||||
> `BrowserClick` / `BrowserScreenshot` / `BrowserScroll` / `BrowserSetFiles` /
|
||||
> `BrowserCloseTab` 与其背后的 cdp-proxy 已停用，调用会直接返回
|
||||
> 要求客户端 **v10.0.98+**。`page.extract-text` / `page.element` / `page.wait` / 内联 wait 块 /
|
||||
> 跨源 iframe 快照需 **v10.0.112+**；`BrowserScript` 需包含 S17/S18 的版本。旧的
|
||||
> `BrowserListTabs` / `BrowserNavigate` / `BrowserEval` / `BrowserClick` / `BrowserScreenshot` /
|
||||
> `BrowserScroll` / `BrowserSetFiles` / `BrowserCloseTab` 与其背后的 cdp-proxy 已停用，调用会直接返回
|
||||
> 「该旧 BrowserXxx/cdp-proxy 入口已停用；请改用 BrowserManage、BrowserSnapshot 或 BrowserAct」。
|
||||
|
||||
## 何时用内置浏览器 vs Python Playwright
|
||||
|
||||
| 场景 | 推荐 |
|
||||
|------|------|
|
||||
| 到达并操作登录态站点（小红书 / B站 / 微博 / 飞书 / 知乎） | **内置浏览器**（登录态优先走 L3-fallback CDP；仅在已授予 `browser.import.*` 时才用 BrowserImport 导入 Cookie） |
|
||||
| 抽取登录态站点的长正文 | Python Playwright（内置浏览器没有批量取文通道，见下） |
|
||||
| 简单点击 / 滚动 / 截图 | **内置浏览器** |
|
||||
| 多个任务要互不串扰地并行 | **内置浏览器**（一任务一 Space） |
|
||||
| 需要复杂等待逻辑（wait_for_selector + race condition） | Python Playwright（cdp-browser.md） |
|
||||
| 需要在浏览器内运行长时间脚本（>30 s） | Python Playwright（内置浏览器单条命令 30 s deadline） |
|
||||
| 需要对元素做放大裁剪截图 | Python Playwright（内置浏览器只有整页截图） |
|
||||
|
||||
## 前置条件
|
||||
|
||||
1. 客户端 v10.0.98+，`~/.desirecore/config/browser.json` 里 `electron-embedded` 或
|
||||
@@ -68,27 +57,52 @@ BrowserManage:
|
||||
|
||||
### BrowserSnapshot
|
||||
|
||||
读页面的**主通道**。返回可交互元素及其 `ref` 句柄、`rect`、`disabled` 状态。
|
||||
读页面的**主通道**。四种 mode：
|
||||
|
||||
```yaml
|
||||
BrowserSnapshot:
|
||||
mode: semantic # semantic（默认）| accessibility | visual
|
||||
mode: semantic # semantic（默认）| text | accessibility | visual
|
||||
sessionId: bss_xxx # 当前 Agent 有多个会话时用于消歧
|
||||
tabId: btab_xxx
|
||||
tabId: btab_xxx # 非活动标签页时用于消歧
|
||||
options: {...} # 各 mode 专属参数，见下
|
||||
```
|
||||
|
||||
提示：
|
||||
- `semantic` 只列 button / input / a 等可交互元素，**不含图片和正文文本节点**
|
||||
- 元素 `name` 往往是 placeholder（如「请输入」），无标签时只能靠 `rect.y` 排序定位
|
||||
- 页面重排后旧 `ref` 会失效——**每次交互前重新取快照**
|
||||
- `accessibility` 能拿到 StaticText 正文，但**只在很简单的页面上可用**：实测
|
||||
example.com 正常返回，维基百科条目一律 `BROWSER_RESULT_TOO_LARGE`（回执上限 2 MB），
|
||||
且 schema 里的 `depth` 参数目前被宿主忽略（硬编码 depth=50），调小也没用
|
||||
- 因此**取真实页面正文仍要靠 L2 Jina Reader（公开页）或 L3-fallback Playwright（登录态）**
|
||||
**semantic** —— 可交互元素清单（button / input / a 等）+ `ref` 句柄。每行形如
|
||||
`[ref=e12] [loc=...] 按钮 提交`：
|
||||
- `loc=` 是该元素的稳定选择器（S7）：仅当能产出时才给出，五级优先级 data-testid →
|
||||
唯一 CSS 安全 id → name 属性 → 链接 href（剥 query 前缀匹配）→ role+name
|
||||
- 跨源 iframe（OOPIF）的元素并入同一棵树，ref 全局连续编号（S35）
|
||||
- 页面重排后旧 `ref` 会失效——**每次交互前重新取快照**；带 ref 的命令要同时给签发该
|
||||
ref 的快照 `snapshotId`
|
||||
- `options`：`scope`（viewport=视口内元素优先截断）、`maxElements`（默认 400 上限 2000）、
|
||||
`maxBytes`（默认 256KB 上限 1MB）、`cursor` 翻页、`limit`
|
||||
|
||||
**text** —— 正文批量取文通道之一（S3/S4）：
|
||||
```yaml
|
||||
BrowserSnapshot:
|
||||
mode: text
|
||||
options:
|
||||
format: markdown # markdown（默认，保留标题/列表/表格）| text（纯文本）
|
||||
maxBytes: 65536 # 默认 65536，最大 524288；超出不报错，截断并返回 nextCursor
|
||||
cursor: t1:...:... # 上一页返回的 nextCursor，同参数续读
|
||||
scope: full_page # full_page（默认）| viewport | ref（配 ref + snapshotId 取子树）
|
||||
includeLinks: true # 链接目标写进正文（只写 http/https）
|
||||
includeTables: true # 表格保留逐行结构
|
||||
```
|
||||
|
||||
**accessibility** —— AX 树（S8 起**尊重 `depth`、预算截断不炸**）：
|
||||
`options.depth`（默认 50 上限 100）、`maxElements`、`maxBytes`、`cursor`。真实内容页不再
|
||||
`BROWSER_RESULT_TOO_LARGE` 整体失败——超预算截断给 nextCursor 续读。
|
||||
|
||||
**visual** —— 截图（像素直接进结果）：
|
||||
`options.format`（png/jpeg）、`quality`、`captureBeyondViewport`（整页）、
|
||||
`clip={x,y,width,height,scale}`（元素级裁剪放大，`scale` 最大 4）。
|
||||
|
||||
### BrowserAct
|
||||
|
||||
一次调用一个受管动作。完整 `action` 见工具 schema，常用的：
|
||||
一次调用一个受管动作。`action` 分组：
|
||||
|
||||
**tab.*** —— `navigate` / `back` / `forward` / `reload` / `activate` / `close`：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
@@ -96,10 +110,15 @@ BrowserAct:
|
||||
params: { url: https://www.xiaohongshu.com/explore/... }
|
||||
```
|
||||
|
||||
**input.*** —— `move` / `click` / `double-click` / `drag` / `wheel` / `touch` / `pinch` /
|
||||
`key` / `text`。走 #1808 输入拟真（坐标派发、拟真轨迹、身份一致性），**反检测站点的交互首选**：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: input.click
|
||||
params: { ref: bref_xxx } # 用快照 ref；坐标会因页面重排失效
|
||||
params:
|
||||
ref: e12 # 语义快照签发的短号 ref（不要裸 x/y）
|
||||
snapshotId: snap-xxx # 签发该 ref 的快照回执里的 snapshotId，必带
|
||||
```
|
||||
|
||||
```yaml
|
||||
@@ -114,37 +133,120 @@ BrowserAct:
|
||||
params: { deltaX: 0, deltaY: 720, x: 640, y: 400 }
|
||||
```
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: tab.activate # 截图前必须先做这一步
|
||||
params: { bounds: { x: 0, y: 0, width: 1280, height: 900 } }
|
||||
```
|
||||
带元素 ref 的 `input.*` 必须在 params 顶层同时带同一条快照回执的 `snapshotId`——ref 序号跨
|
||||
快照会重复，缺 snapshotId 一律拒绝执行。
|
||||
|
||||
**page.element** —— 判别式元素命令（op × selector，九 op，S10）。**表单批量填充等站点
|
||||
不检测场景用**；反检测站点交互走 input.*：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: page.screenshot
|
||||
params: { format: png } # 结果落 artifact store，回执给 artifact.id / sha256 / bytes
|
||||
action: page.element
|
||||
params:
|
||||
op: fill # 写类：fill | select-option | check | uncheck | scroll-into-view
|
||||
# 读类：get-attribute | bounding-box | count | all-inner-texts
|
||||
selector: loc=css:input#email # loc= 方言或 e12 / ref=e12（配 snapshotId）
|
||||
value: user@example.com # fill / select-option 用
|
||||
snapshotId: bsnp_xxx # selector 用 ref 时必带
|
||||
```
|
||||
|
||||
- 写类走 `browser.input.keyboard` 能力档（与 input.text 同档）；读类走 `browser.observe.snapshot` 只读档
|
||||
- `fill` 对 `input[type=password]` **一律拒绝**
|
||||
- `selector` 方言：`e<序号>` / `ref=e<序号>`、`loc=css:` / `loc=role:` / `loc=text:` /
|
||||
`loc=testid:`、裸 CSS，可叠 `internal:nth/last/scope/filter`。未知前缀（如 `loc=xpath:`）显式报错，绝不静默降级
|
||||
|
||||
**page.wait** —— 判别式等待（until，九种，S11）。等待不改页面状态，走只读档：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: page.wait
|
||||
params:
|
||||
until: networkidle # load | domcontentloaded | networkidle | selector | url | timeout
|
||||
# | request | response | download
|
||||
timeoutMs: 10000 # 默认 10000，上限 60000
|
||||
idleMs: 500 # networkidle 静默窗，默认 500
|
||||
```
|
||||
|
||||
- 轮询型（load/domcontentloaded/networkidle/selector/url/timeout）超时返回 `waited: false`，不抛错
|
||||
- 事件型（request/response/download）超时**抛错**
|
||||
- `waitForFunction` 刻意不在枚举里——任意 JS 走 page.evaluate 的能力档与审批
|
||||
|
||||
**内联 wait 块**（S14）—— `tab.navigate` / `input.click` / `input.key` / `page.element{op:"fill"}`
|
||||
支持 `params.wait`（形态与 page.wait 参数同构），一条回执完成「动作→等结果」，等待器先于动作注册，
|
||||
不跨两次 IPC 出竞态：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: tab.navigate
|
||||
params:
|
||||
url: https://example.com/login
|
||||
wait: { until: networkidle, timeoutMs: 15000 }
|
||||
```
|
||||
|
||||
**page.evaluate** —— 页面上下文求值（S2 起返回真实值）：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: page.evaluate
|
||||
params:
|
||||
expression: document.title
|
||||
awaitPromise: true # 默认 true
|
||||
```
|
||||
|
||||
- 返回值**原样过界**（字符串/对象/数组都真实返回）；超预算截断并标 `truncated`，不抛错
|
||||
- 能力档 `browser.page.evaluate` 属人工闸门：非 allow-all 模式每次调用弹审批卡
|
||||
- 登录态取站内接口走 [fetch.browser 配方]（见下节）
|
||||
|
||||
**page.extract-text** —— 与 `BrowserSnapshot mode:text` 同款取文通道，作为动作下发：
|
||||
scope（viewport/full_page/ref）+ ref + snapshotId、format（markdown/text）、maxBytes（默认
|
||||
65536 最大 524288）、cursor（`^t1:` 续读）、includeLinks、includeTables。超出预算截断 + nextCursor。
|
||||
|
||||
**page.screenshot** —— 与 `BrowserSnapshot mode:visual` 同款：`format`、`quality`、
|
||||
`captureBeyondViewport`、`clip={x,y,width,height,scale}`。结果落 artifact store，回执给
|
||||
artifact.id / sha256 / bytes；截图像素直接进结果（见「截图」节）。
|
||||
|
||||
### BrowserScript（代码模式，S17+S18）
|
||||
|
||||
一段异步 JS 在 Worker 里跑，通过注入的 `page` / `tab` / `input` / `snapshot` / `console` /
|
||||
`performance` 门面连续下发命令——**导航→快照→点击→等待→取文一气呵成**，消除逐动作往返：
|
||||
|
||||
```yaml
|
||||
BrowserScript:
|
||||
code: |
|
||||
const snap = await snapshot.semantic();
|
||||
receipt.log(snap.content);
|
||||
await input.click({ ref: 'e12', snapshotId: snap.snapshotId });
|
||||
await page.wait({ until: 'networkidle' });
|
||||
receipt.log(await page['extract-text']({ format: 'markdown' }));
|
||||
sessionId: bss_xxx # 可选，多会话消歧
|
||||
tabId: btab_xxx # 可选，缺省用会话活动 tab
|
||||
totalBudgetMs: 180000 # 默认 180000，上限 600000；审批等待不计入
|
||||
maxCommands: 500 # 每次运行命令条数上限
|
||||
```
|
||||
|
||||
- ❗**信任级别等同 Bash**：Worker 不是沙箱，脚本可访问 Node fs/net/child_process。两层审批：
|
||||
脚本源码先过一次与 Bash 同档的人类审批（allow-all / ask-external 豁免；脚本源码不可记忆），
|
||||
其内每个受管浏览器能力再按「能力 × 本次运行」各问一次
|
||||
- helper 白名单 24 个：`snapshot.*` / `page.extract-text` / `tab.*` / `input.*` /
|
||||
`page.evaluate` / `page.wait` / `page.element` / `console.read` / `performance.metrics`；
|
||||
`console.log(page)` 可输出各 helper 的 signature/params/example 文档
|
||||
- 会话被用户接管、Lease 轮换 ⇒ **硬停**：后续命令一律不再下发
|
||||
- 需要 `browser.*` 相应能力；`receipt.log(...)` 或 `console.log` 输出文本
|
||||
|
||||
### BrowserImport（**需人工审批 + 额外授权**）
|
||||
|
||||
把用户浏览器里的登录态 Cookie 导入当前 Space。
|
||||
把用户浏览器里的登录态 Cookie 导入当前 Space——**唯一的登录态复用通道**。
|
||||
|
||||
> **前置条件（先看这里，别直接试）**：`browser.import.discover` /
|
||||
> `browser.import.cookies.inspect` / `browser.import.cookies` 三个能力**不在 `agentDefault` 里**，
|
||||
> 而 `BrowserManage(create_space)` 建的 Agent grant 就是按 `agentDefault` 签的。也就是说
|
||||
> **仅靠 create_space / start_session 走不通 BrowserImport**，必须由 Host/用户侧另行授予 import
|
||||
> 能力（`agentElevated` 或 Workbench 路径）。没有这层授权就别在这条路上耗——**直接回落
|
||||
> L3-fallback（Python Playwright 连用户已登录的 Chrome）**，那是当前更稳的登录态复用方式。
|
||||
> 能力（`agentElevated` 或 Workbench 路径）。没有这层授权就别在这条路上耗——如实告诉用户
|
||||
> 「当前无法复用你的登录态」，按无登录态继续或放弃。
|
||||
|
||||
动作枚举只有这 6 个：`discover` | `create_plan` | `dry_run` | `apply` | `rollback` | `list_plans`
|
||||
（**没有 `plan`**）。完整流程是 `discover → create_plan → dry_run → apply`：
|
||||
|
||||
```yaml
|
||||
BrowserImport:
|
||||
action: discover # 列出可导入的来源，返回不透明 sourceProfileId
|
||||
```
|
||||
|
||||
```yaml
|
||||
BrowserImport:
|
||||
action: create_plan # 域名授权、来源、冲突策略都在这一步定死
|
||||
@@ -155,57 +257,85 @@ BrowserImport:
|
||||
conflictStrategy: newer-wins # keep-target | replace-target | newer-wins | fail-on-conflict
|
||||
```
|
||||
|
||||
```yaml
|
||||
BrowserImport:
|
||||
action: dry_run # 先看命中多少条，再决定要不要真的写入
|
||||
planId: bimp_xxx # create_plan 返回的 ID，前缀是 bimp_
|
||||
```
|
||||
|
||||
```yaml
|
||||
BrowserImport:
|
||||
action: apply # 只认 planId；域名/策略在 create_plan 时已固化
|
||||
planId: bimp_xxx
|
||||
```
|
||||
|
||||
写坏了用 `action: rollback` + 同一个 `planId` 回退；`list_plans` 查当前 Space 的历史 plan。
|
||||
|
||||
解密与过滤全在 Host 侧完成，**Cookie 值不会进入 Agent 上下文或审计日志**。
|
||||
`dry_run` 先看命中多少条再 `apply`（只认 `create_plan` 返回的 `planId`，前缀 `bimp_`）；写坏了
|
||||
用 `rollback` + 同一个 planId 回退；`list_plans` 查历史。解密与过滤全在 Host 侧完成，
|
||||
**Cookie 值不会进入 Agent 上下文或审计日志**。
|
||||
|
||||
### BrowserShare
|
||||
|
||||
把 Space / Session 委派给另一个 Agent，`shareMode` 可选 `snapshot`（只读副本）、
|
||||
`copy-on-write`（写时复制）、`live-shared`（实时共享）、`handoff`（移交控制权）。
|
||||
|
||||
## fetch.browser 配方：带登录态取接口数据
|
||||
|
||||
站内接口（列表、评论、订单等 JSON）在登录态下的正解——**页面上下文跑 `fetch`**，自动带该
|
||||
origin 的 Cookie，同 origin、受 Grant origins 约束，走 `page.evaluate` 已有闸门：
|
||||
|
||||
```yaml
|
||||
BrowserAct:
|
||||
action: page.evaluate
|
||||
params:
|
||||
expression: |
|
||||
fetch('/api/v1/comments?page=1&size=20', {
|
||||
headers: { accept: 'application/json' }
|
||||
}).then(r => r.text())
|
||||
awaitPromise: true
|
||||
```
|
||||
|
||||
- 先 `tab.navigate` 到该站任意页面建立 origin 与 Cookie，再发 fetch；路径写**相对路径**
|
||||
- 只能访问当前 tab origin；跨站接口先导航过去
|
||||
- 大 JSON 用 `.text()` 拿原文自己截取，或分页多次取
|
||||
- 非 allow-all 模式会弹审批卡（page.evaluate 人工闸门），向用户说明用途即可
|
||||
|
||||
## 交互通道选用：input.* vs page.element（反检测决策）
|
||||
|
||||
| 场景 | 用什么 | 原因 |
|
||||
|------|--------|------|
|
||||
| 反检测站点（小红书/微博/B站等）的一切点击/输入 | **`input.*`** | 坐标派发 + #1808 拟真轨迹 + 身份一致性（UA/UA-CH 无 Electron/Headless 痕迹） |
|
||||
| 表单批量填充等**站点不检测**的场景 | **`page.element` 写类** | 一条命令完成 fill/select-option/check，比逐元素 input.* 快得多 |
|
||||
| 需要元素属性/坐标/计数/批量文本 | `page.element` 读类 | 只读档（browser.observe.snapshot），无写审批 |
|
||||
| 任何「用 JS 直调 el.click()」的想法 | ❌ 禁止 | 绕开全部拟真投入（S10 红线）；指针动作只走 input.* |
|
||||
|
||||
## 截图：像素直接给你，通常不需要再 Read
|
||||
|
||||
> **需要客户端 10.0.100+。** 更老的版本上截图只回元数据；
|
||||
> 若需要画面信息请改用 `semantic` / `accessibility` 快照，或回落 Jina / Playwright。
|
||||
`BrowserSnapshot mode:visual` 与 `BrowserAct page.screenshot` 会把截图像素作为 image 块
|
||||
**直接放进工具结果**——视觉模型当场就能看，不必再调 `Read`。
|
||||
|
||||
`BrowserSnapshot` `mode: visual` 与 `BrowserAct` `page.screenshot` 会把截图像素作为
|
||||
image 块**直接放进工具结果**——视觉模型当场就能看，不必再调 `Read`。
|
||||
|
||||
只有这几种情况才需要走 `result.artifact.absolutePath` 再 `Read` 一次：
|
||||
|
||||
- 结果里明确写了「截图已保存，但…未附带像素」（超出单图预算、或体积过大）
|
||||
- 你需要的是**原始分辨率**，而附带的像素被压缩过
|
||||
|
||||
对同一张图既看 image 块又 `Read` 一遍，只会让它在上下文里占两份。
|
||||
- **元素级裁剪放大**：`options.clip={x,y,width,height,scale}`（`scale` 最大 4）。先用
|
||||
`semantic` 快照拿到元素坐标，再截那一块并放大——验证码、小按钮在整页截图里只有几十像素，
|
||||
看不清时用它。**不必走 `cdp.raw`**
|
||||
- **整页截图**：`captureBeyondViewport: true`（超出视口部分也截）
|
||||
- 只有结果里明确写了「截图已保存，但…未附带像素」（超预算/体积过大）或需要原始分辨率时，
|
||||
才走 `result.artifact.absolutePath` 再 `Read` 一次
|
||||
|
||||
当前模型不支持视觉输入时，结果会明说「你看不到它的内容」——**此时不要凭空描述画面**，
|
||||
改用 `semantic` / `accessibility` 快照拿页面信息。
|
||||
改用 `semantic` / `text` 快照拿页面信息。
|
||||
|
||||
## 视口与 tab.activate（S36 起）
|
||||
|
||||
Agent 会话（actor ≠ user）的标签页**常驻离屏原位，保住合成表面**，不再停放成 1×1：
|
||||
|
||||
| 场景 | 行为 |
|
||||
|------|------|
|
||||
| embedded，单标签会话 | 无需 `tab.activate` 即可截图/取视口（presentation 即呈现） |
|
||||
| embedded，多标签会话的后台 tab | 仍需先 `tab.activate`；未激活的视口快判**秒级**报 `BROWSER_VIEWPORT_UNAVAILABLE`，不再挂满 30s，也不会毁掉标签页 |
|
||||
| standalone | 后台 tab 可直接截图 |
|
||||
|
||||
多 Space 依然可以并发导航/快照/输入，互不串扰。
|
||||
|
||||
## 已知边界（照做，别试探）
|
||||
|
||||
| 边界 | 说明 |
|
||||
|------|------|
|
||||
| **截图前必须 `tab.activate`** | 标签页默认停在 `(-10000,-10000,1x1)`，没有合成表面。直接截图会卡满 30 s deadline，**并把标签页宿主打掉**，之后全部报 `BROWSER_TAB_HOST_NOT_FOUND` |
|
||||
| **同时只有一个可见标签页** | `tab.activate` 绑定主窗口、全局互斥。多 Space 可以并发导航/快照/输入，但截图必须逐个 activate 串行 |
|
||||
| **`page.evaluate` 基本不可用** | 每次调用需人工审批；返回的字符串/对象被替换为 `[REDACTED:browser-runtime-value]`（只有 number/boolean/null 穿透）；反调试站点会把它挂起几十秒。读页面用 `BrowserSnapshot` |
|
||||
| **`cdp.raw` 需人工审批** | `browser.raw_cdp.*` 属于永远人工闸门的能力，无人值守流程用不了。**元素级裁剪不必走它**（需 10.0.100+）——`BrowserSnapshot` 的 `options.clip={x,y,width,height,scale}` 直接支持（`scale` 最大 4）：先用 `semantic` 快照拿到元素坐标，再截那一块并放大。验证码、小按钮在整页截图里只有几十像素，看不清时用它 |
|
||||
| **没有批量取文通道** | `semantic` 不含正文，`accessibility` 在真实页面上超限，`page.evaluate` 被审批+脱敏。要正文请回落 Jina / Playwright |
|
||||
| **单条命令 30 s deadline** | 超时即判 `browser.host.gone`，会话作废 |
|
||||
| **用户真实操作会抢控制权** | 在可见标签页上**点击、按键、滚轮、触摸**会触发 `trusted-user-input` 并递增 control epoch，打断 Agent 的租约。**鼠标只是划过不会**（默认策略 `intentional-input`，需 10.0.100+；更老版本上划过也会打断），所以用户看着页面移动光标是安全的 |
|
||||
| **artifact 不要走 `/save`** | 该接口会弹系统「另存为」对话框等人点。**用回执里的 `result.artifact.absolutePath`**（需 10.0.100+；更老版本没有这个字段，只能放弃直接读，改用页面内提取）（工具会把它登记进本次会话的可读白名单），不要自己拼路径——`${DESIRECORE_ROOT}` 这类变量在路径展开里不认（只认 `~` / `$HOME` / `$USERPROFILE`），拼出来是相对路径、`Read` 会报「文件不存在」。默认保留 24 小时 |
|
||||
| **`page.evaluate` 走人工闸门** | 能力 `browser.page.evaluate` 属 always-human-gate：非 allow-all 模式每次弹审批卡。返回值已原样过界（超预算截断标 `truncated`），登录态取接口走 fetch.browser 配方 |
|
||||
| **`cdp.raw` 需人工审批** | `browser.raw_cdp.*` 属于永远人工闸门的能力，无人值守流程用不了。元素级裁剪 / 整页截图用 `clip` / `captureBeyondViewport`，别走它 |
|
||||
| **`page.element` 没有 click** | 指针动作必须走 `input.*`（拟真轨迹、可审计）；JS 直调 `el.click()` 是明确禁止的回退 |
|
||||
| **`fill` 拒绝密码框** | `input[type=password]` 一律拒绝——密码输入只走 `input.text`（拟真键入） |
|
||||
| **单条命令 30 s deadline** | 超时只掐掉**这一条命令**（stop 加载），标签页仍可用，重试即可；不再打掉宿主。等待类命令超时连 stop 都跳过 |
|
||||
| **多标签后台 tab 先 activate** | embedded 多标签会话里后台 tab 截图前先 `tab.activate`，否则秒级 `BROWSER_VIEWPORT_UNAVAILABLE` |
|
||||
| **用户真实操作会抢控制权** | 在可见标签页上**点击、按键、滚轮、触摸**会触发 `trusted-user-input` 并递增 control epoch，打断 Agent 的租约。**鼠标只是划过不会**（默认策略 `intentional-input`），所以用户看着页面移动光标是安全的 |
|
||||
| **BrowserScript 信任级别等同 Bash** | 脚本可访问 Node fs/net/child_process；源码过一次与 Bash 同档审批 + 每能力一问；接管/Lease 轮换即硬停 |
|
||||
| **artifact 不要走 `/save`** | 该接口会弹系统「另存为」对话框等人点。**用回执里的 `result.artifact.absolutePath`**（工具会把它登记进本次会话的可读白名单），不要自己拼路径——`${DESIRECORE_ROOT}` 这类变量在路径展开里不认（只认 `~` / `$HOME` / `$USERPROFILE`），拼出来是相对路径、`Read` 会报「文件不存在」。默认保留 24 小时 |
|
||||
|
||||
## SitePatternRead / SitePatternWrite
|
||||
|
||||
@@ -228,17 +358,18 @@ SitePatternWrite:
|
||||
| 错误 | 原因 | 解决 |
|
||||
|------|------|------|
|
||||
| `该旧 BrowserXxx/cdp-proxy 入口已停用` | 还在调 v2.0 的旧工具 | 改用 BrowserManage / BrowserSnapshot / BrowserAct |
|
||||
| `BROWSER_TAB_HOST_NOT_FOUND` | 标签页宿主已销毁（多因上一条命令超时） | 重建 Session；检查是否漏了 `tab.activate` |
|
||||
| `BROWSER_COMMAND_DEADLINE_EXCEEDED` | 单条命令超 30 s | 截图先 activate；避免 `page.evaluate` |
|
||||
| `BROWSER_TAB_HOST_NOT_FOUND` | 标签页宿主真的没了（渲染进程崩溃/销毁） | 重建 Session。命令超时**不再**引发此错误——超时只 stop 这一条命令，标签页保留 |
|
||||
| `BROWSER_COMMAND_DEADLINE_EXCEEDED` | 单条命令超 30 s | 标签页仍停在上一页/空白页，直接重试；慢加载常态不算故障 |
|
||||
| `BROWSER_VIEWPORT_UNAVAILABLE` | embedded 多标签会话的后台 tab 没有可用视口 | 先 `tab.activate` 再截图（秒级快判，不是挂死） |
|
||||
| `BROWSER_TOOL_SESSION_FORBIDDEN` | 会话已关闭/崩溃，或不属于当前 Agent | 重新 `list_sessions`，必要时重建 |
|
||||
| `BROWSER_TOOL_SESSION_AMBIGUOUS` | 当前 Agent 有多个会话且未传 sessionId | 显式传 `sessionId` |
|
||||
| `BROWSER_HUMAN_APPROVAL_REQUIRED` | 触到人工闸门能力（evaluate / raw_cdp / 上传下载 / Cookie 导入等） | 向用户说明用途并等待审批，或换用无需审批的路径 |
|
||||
| `BROWSER_RESULT_TOO_LARGE` | 回执超过 2 MB（`accessibility` 快照最常见） | 改用 `semantic` 快照 + 截图；取正文回落 Jina / Playwright |
|
||||
| `BROWSER_HUMAN_APPROVAL_REQUIRED` | 触到人工闸门能力（evaluate / raw_cdp / 上传下载 / Cookie 导入等） | 向用户说明用途并等待审批，或换用无需审批的路径（取文用 extract-text，裁剪用 clip） |
|
||||
| `BROWSER_RESULT_TOO_LARGE` | 回执超过结果上限 | text/semantic/accessibility 都有 maxBytes + cursor 分页，调小 maxBytes 续读即可，不再整体失败 |
|
||||
|
||||
## 调用链路
|
||||
|
||||
```
|
||||
Agent → BrowserManage/Snapshot/Act → browser-use service（Capability/Grant/Lease/Policy 校验）
|
||||
Agent → BrowserManage/Snapshot/Act/Script → browser-use service（Capability/Grant/Lease/Policy 校验）
|
||||
→ BrowserHost（electron-embedded 或 standalone-managed）→ Chromium
|
||||
↑ 每步产出带 digest 的回执，写入审计事件流
|
||||
```
|
||||
|
||||
@@ -1,330 +0,0 @@
|
||||
# CDP Browser Access — Login-Gated Sites Manual
|
||||
|
||||
Detailed recipes for accessing sites that require the user's login session, via Chrome DevTools Protocol (CDP) + Python Playwright.
|
||||
|
||||
**Precondition**: Chrome is already running with `--remote-debugging-port=9222` and the user has manually logged in to the target sites. See the main SKILL.md `Prerequisites` section for the launch command.
|
||||
|
||||
---
|
||||
|
||||
## Why CDP attach, not headless
|
||||
|
||||
| Approach | Login state | Anti-bot | Speed | Cost |
|
||||
|----------|-------------|----------|-------|------|
|
||||
| Headless Playwright (new context) | ❌ Empty cookies | ❌ Flagged as bot | Slow cold start | Re-login pain |
|
||||
| `playwright.chromium.launch(headless=False)` | ❌ Fresh profile | ⚠ Sometimes flagged | Slow | Same |
|
||||
| **CDP attach (`connect_over_cdp`)** | ✅ User's real cookies | ✅ Looks human | Instant | Zero friction |
|
||||
|
||||
**Rule**: For any login-gated site, always attach to the user's running Chrome.
|
||||
|
||||
---
|
||||
|
||||
## Core Template
|
||||
|
||||
Every CDP script follows this shape:
|
||||
|
||||
```python
|
||||
from playwright.sync_api import sync_playwright
|
||||
|
||||
def fetch_with_cdp(url: str, wait_selector: str | None = None) -> str:
|
||||
"""Attach to user's Chrome via CDP, fetch URL, return HTML."""
|
||||
with sync_playwright() as p:
|
||||
browser = p.chromium.connect_over_cdp("http://localhost:9222")
|
||||
# browser.contexts[0] is the user's default context (with cookies)
|
||||
context = browser.contexts[0]
|
||||
page = context.new_page()
|
||||
try:
|
||||
page.goto(url, wait_until="domcontentloaded", timeout=30000)
|
||||
if wait_selector:
|
||||
page.wait_for_selector(wait_selector, timeout=10000)
|
||||
else:
|
||||
page.wait_for_timeout(2000) # generic settle
|
||||
return page.content()
|
||||
finally:
|
||||
page.close()
|
||||
# DO NOT call browser.close() — that would close the user's Chrome!
|
||||
|
||||
if __name__ == "__main__":
|
||||
html = fetch_with_cdp("https://example.com")
|
||||
print(html[:1000])
|
||||
```
|
||||
|
||||
**Critical**: Never call `browser.close()` when using CDP attach — you'd kill the user's Chrome. Only close the page you opened.
|
||||
|
||||
---
|
||||
|
||||
## Site Recipes
|
||||
|
||||
### 小红书 (xiaohongshu.com)
|
||||
|
||||
```python
|
||||
from playwright.sync_api import sync_playwright
|
||||
from bs4 import BeautifulSoup
|
||||
|
||||
NOTE_URL = "https://www.xiaohongshu.com/explore/XXXXXXXX"
|
||||
|
||||
with sync_playwright() as p:
|
||||
browser = p.chromium.connect_over_cdp("http://localhost:9222")
|
||||
page = browser.contexts[0].new_page()
|
||||
page.goto(NOTE_URL, wait_until="domcontentloaded")
|
||||
page.wait_for_selector("#detail-title", timeout=10000)
|
||||
page.wait_for_timeout(1500) # let images/comments load
|
||||
html = page.content()
|
||||
page.close()
|
||||
|
||||
soup = BeautifulSoup(html, "html.parser")
|
||||
title = (soup.select_one("#detail-title") or {}).get_text(strip=True) if soup.select_one("#detail-title") else None
|
||||
desc = (soup.select_one("#detail-desc") or {}).get_text(" ", strip=True) if soup.select_one("#detail-desc") else None
|
||||
author = soup.select_one(".author-wrapper .username")
|
||||
print("Title:", title)
|
||||
print("Author:", author.get_text(strip=True) if author else None)
|
||||
print("Desc:", desc)
|
||||
```
|
||||
|
||||
**Selectors** (may drift over time — update if they fail):
|
||||
- Title: `#detail-title`
|
||||
- Description: `#detail-desc`
|
||||
- Author: `.author-wrapper .username`
|
||||
- Images: `.swiper-slide img`
|
||||
- Comments: `.parent-comment .content`
|
||||
|
||||
### B站 (bilibili.com)
|
||||
|
||||
```python
|
||||
from playwright.sync_api import sync_playwright
|
||||
from bs4 import BeautifulSoup
|
||||
|
||||
VIDEO_URL = "https://www.bilibili.com/video/BVxxxxxxxxx"
|
||||
|
||||
with sync_playwright() as p:
|
||||
browser = p.chromium.connect_over_cdp("http://localhost:9222")
|
||||
page = browser.contexts[0].new_page()
|
||||
page.goto(VIDEO_URL, wait_until="networkidle")
|
||||
page.wait_for_timeout(2000)
|
||||
html = page.content()
|
||||
page.close()
|
||||
|
||||
soup = BeautifulSoup(html, "html.parser")
|
||||
print("Title:", soup.select_one("h1.video-title").get_text(strip=True) if soup.select_one("h1.video-title") else None)
|
||||
print("UP:", soup.select_one(".up-name").get_text(strip=True) if soup.select_one(".up-name") else None)
|
||||
print("Desc:", soup.select_one(".desc-info-text").get_text(" ", strip=True) if soup.select_one(".desc-info-text") else None)
|
||||
```
|
||||
|
||||
**Tip**: For B站 evaluations, the [公开 API](https://api.bilibili.com/x/web-interface/view?bvid=XXXX) often returns JSON without needing CDP. Try it first:
|
||||
|
||||
```bash
|
||||
curl -s "https://api.bilibili.com/x/web-interface/view?bvid=BVxxxxxxxxx" | python3 -m json.tool
|
||||
```
|
||||
|
||||
### 微博 (weibo.com)
|
||||
|
||||
```python
|
||||
WEIBO_URL = "https://weibo.com/u/1234567890" # or /detail/xxx
|
||||
|
||||
# Same CDP template
|
||||
# Selectors:
|
||||
# .Feed_body_3R0rO .detail_wbtext_4CRf9 — post text
|
||||
# .ALink_default_2ibt1 — user link
|
||||
# article[aria-label="微博"] — each feed item
|
||||
```
|
||||
|
||||
**Note**: Weibo uses React + heavy obfuscation. Selectors change frequently. If selectors fail, pipe the HTML through Jina for clean Markdown:
|
||||
|
||||
```python
|
||||
html = fetch_with_cdp(WEIBO_URL)
|
||||
# Save to temp file, then:
|
||||
import subprocess
|
||||
result = subprocess.run(
|
||||
["curl", "-sL", f"https://r.jina.ai/{WEIBO_URL}"],
|
||||
capture_output=True, text=True
|
||||
)
|
||||
print(result.stdout)
|
||||
```
|
||||
|
||||
### 知乎 (zhihu.com)
|
||||
|
||||
```python
|
||||
ANSWER_URL = "https://www.zhihu.com/question/123/answer/456"
|
||||
|
||||
# Selectors:
|
||||
# h1.QuestionHeader-title — question title
|
||||
# .RichContent-inner — answer body
|
||||
# .AuthorInfo-name — author
|
||||
```
|
||||
|
||||
Zhihu works with CDP but often also renders enough metadata server-side for Jina to work:
|
||||
|
||||
```bash
|
||||
curl -sL "https://r.jina.ai/https://www.zhihu.com/question/123/answer/456"
|
||||
```
|
||||
|
||||
Try Jina first, fall back to CDP if content is truncated.
|
||||
|
||||
### 飞书文档 (feishu.cn / larksuite.com)
|
||||
|
||||
```python
|
||||
DOC_URL = "https://xxx.feishu.cn/docs/xxx"
|
||||
|
||||
# Feishu uses heavy virtualization — must scroll to load all content.
|
||||
# Recipe:
|
||||
|
||||
from playwright.sync_api import sync_playwright
|
||||
|
||||
with sync_playwright() as p:
|
||||
browser = p.chromium.connect_over_cdp("http://localhost:9222")
|
||||
page = browser.contexts[0].new_page()
|
||||
page.goto(DOC_URL, wait_until="domcontentloaded")
|
||||
page.wait_for_selector(".docs-render-unit", timeout=15000)
|
||||
|
||||
# Scroll to bottom repeatedly to load lazy content
|
||||
last_height = 0
|
||||
for _ in range(20):
|
||||
page.evaluate("window.scrollTo(0, document.body.scrollHeight)")
|
||||
page.wait_for_timeout(800)
|
||||
h = page.evaluate("document.body.scrollHeight")
|
||||
if h == last_height:
|
||||
break
|
||||
last_height = h
|
||||
|
||||
# Extract text
|
||||
text = page.evaluate("() => document.body.innerText")
|
||||
page.close()
|
||||
|
||||
print(text)
|
||||
```
|
||||
|
||||
### Twitter / X
|
||||
|
||||
```python
|
||||
TWEET_URL = "https://x.com/username/status/1234567890"
|
||||
|
||||
# Selectors:
|
||||
# article[data-testid="tweet"] — tweet container
|
||||
# div[data-testid="tweetText"] — tweet text
|
||||
# div[data-testid="User-Name"] — author
|
||||
# a[href$="/analytics"] — view count anchor (next sibling has stats)
|
||||
```
|
||||
|
||||
Twitter is aggressive with anti-bot. CDP attach usually works, but set a generous wait:
|
||||
|
||||
```python
|
||||
page.goto(url, wait_until="networkidle", timeout=45000)
|
||||
page.wait_for_selector('article[data-testid="tweet"]', timeout=15000)
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Common Patterns
|
||||
|
||||
### Pattern 1: Scroll to load lazy content
|
||||
|
||||
```python
|
||||
def scroll_to_bottom(page, max_steps=30, pause_ms=800):
|
||||
last = 0
|
||||
for _ in range(max_steps):
|
||||
page.evaluate("window.scrollTo(0, document.body.scrollHeight)")
|
||||
page.wait_for_timeout(pause_ms)
|
||||
h = page.evaluate("document.body.scrollHeight")
|
||||
if h == last:
|
||||
return
|
||||
last = h
|
||||
```
|
||||
|
||||
### Pattern 2: Screenshot a specific element
|
||||
|
||||
```python
|
||||
element = page.locator("article").first
|
||||
element.screenshot(path="/tmp/article.png")
|
||||
```
|
||||
|
||||
### Pattern 3: Extract structured data via JavaScript
|
||||
|
||||
```python
|
||||
data = page.evaluate("""() => {
|
||||
const items = document.querySelectorAll('.list-item');
|
||||
return Array.from(items).map(el => ({
|
||||
title: el.querySelector('.title')?.innerText,
|
||||
url: el.querySelector('a')?.href,
|
||||
}));
|
||||
}""")
|
||||
print(data)
|
||||
```
|
||||
|
||||
### Pattern 4: Fill a form and click
|
||||
|
||||
```python
|
||||
page.fill("input[name=q]", "search query")
|
||||
page.click("button[type=submit]")
|
||||
page.wait_for_load_state("networkidle")
|
||||
```
|
||||
|
||||
### Pattern 5: Clean HTML via Jina after extraction
|
||||
|
||||
When selectors are unreliable, dump the full page HTML and let Jina do the cleaning:
|
||||
|
||||
```python
|
||||
html = page.content()
|
||||
# Save to file, serve via local HTTP, or just pipe the original URL:
|
||||
import subprocess
|
||||
clean_md = subprocess.run(
|
||||
["curl", "-sL", f"https://r.jina.ai/{url}"],
|
||||
capture_output=True, text=True
|
||||
).stdout
|
||||
print(clean_md)
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Troubleshooting
|
||||
|
||||
### `connect_over_cdp` fails with `ECONNREFUSED`
|
||||
|
||||
Chrome is not running with remote debugging. Tell the user:
|
||||
> "请先用下面的命令启动 Chrome：
|
||||
> `/Applications/Google\\ Chrome.app/Contents/MacOS/Google\\ Chrome --remote-debugging-port=9222 --user-data-dir=\"${DESIRECORE_ROOT}/chrome-profile\"`
|
||||
> 然后手动登录需要抓取的网站，再让我继续。"
|
||||
|
||||
### `browser.contexts[0]` is empty
|
||||
|
||||
Chrome was launched but no windows are open. Ask the user to open at least one tab and navigate anywhere.
|
||||
|
||||
### Playwright not installed
|
||||
|
||||
```bash
|
||||
pip3 install playwright beautifulsoup4
|
||||
# No need for `playwright install` — we're attaching to existing Chrome, not downloading a new browser
|
||||
```
|
||||
|
||||
### Site detects automation
|
||||
|
||||
Despite CDP attach, some sites (Cloudflare-protected, Instagram) may still detect automation. Options:
|
||||
1. Use Jina Reader instead (`curl -sL https://r.jina.ai/<url>`) — often succeeds where Playwright fails
|
||||
2. Ask the user to manually copy the visible content
|
||||
3. Use the site's public API if available
|
||||
|
||||
### Content is truncated
|
||||
|
||||
The page uses virtualization or lazy loading. Apply Pattern 1 (scroll to bottom) before calling `page.content()`.
|
||||
|
||||
### `page.wait_for_selector` times out
|
||||
|
||||
The selector is stale — the site updated its DOM. Dump `page.content()[:5000]` and inspect manually, or fall back to Jina Reader.
|
||||
|
||||
---
|
||||
|
||||
## Security Notes
|
||||
|
||||
- **Never log or print cookies** from `context.cookies()` even during debugging
|
||||
- **Never extract and store** the user's session tokens to files
|
||||
- **Never use the CDP session** to perform writes (post, comment, like) unless the user explicitly requested it
|
||||
- The `${DESIRECORE_ROOT}/chrome-profile` directory contains the user's credentials — treat it as sensitive
|
||||
- If the user asks to "log in automatically", refuse and explain they must log in manually in the Chrome window; the skill only reads already-authenticated sessions
|
||||
|
||||
---
|
||||
|
||||
## When NOT to use CDP
|
||||
|
||||
- **Public static sites** → use L1 `WebFetch`, it's faster
|
||||
- **Heavy SPAs without login walls** → use L2 Jina Reader, it's cheaper on tokens
|
||||
- **You need thousands of pages** → CDP is not built for scale; look into proper scrapers
|
||||
|
||||
CDP is specifically the "right tool" for: **small number of pages + login required + human-like behavior needed**.
|
||||
@@ -95,19 +95,8 @@ curl -sL "https://r.jina.ai/$URL" > /tmp/jina-cache.md
|
||||
### Handle very long articles
|
||||
Jina returns the full article in one response. For articles > 50K chars, pipe through `head` or extract specific sections with Python/awk before feeding back to the model context.
|
||||
|
||||
### Combine with CDP
|
||||
When you use L3 CDP to fetch a login-gated page, you can pipe the resulting HTML through Jina for clean Markdown instead of parsing with BeautifulSoup:
|
||||
|
||||
```python
|
||||
html = fetch_with_cdp(url) # from references/cdp-browser.md
|
||||
# Now convert via Jina (note: Jina fetches the URL itself, not your HTML)
|
||||
# So this only works if the content is already visible without login:
|
||||
import subprocess
|
||||
md = subprocess.run(["curl", "-sL", f"https://r.jina.ai/{url}"],
|
||||
capture_output=True, text=True).stdout
|
||||
```
|
||||
|
||||
For truly login-gated content, you must parse the HTML directly (BeautifulSoup) since Jina can't log in on your behalf.
|
||||
### Combine with the built-in browser
|
||||
For login-gated pages, the built-in browser already has its own extraction channel with paged budgets — `BrowserAct({ action: 'page.extract-text', params: { format: 'markdown', maxBytes: 65536 } })` returns clean Markdown directly (beyond maxBytes it truncates and hands back a nextCursor). Jina can't log in on your behalf, so use Jina for public pages and `page.extract-text` for logged-in ones; there is no need to round-trip HTML through Jina anymore.
|
||||
|
||||
---
|
||||
|
||||
|
||||
Reference in New Issue
Block a user