Add page --all full-page reading mode with scroll settling

README.en.md

@@ -149,7 +149,7 @@ browser-cli close <session-id> [--json]
 browser-cli close --all [--json]
 browser-cli --version
 
-browser-cli page <session-id> [-p <page>] [--next] [--prev] [--fresh] [--json] [--verbose]
+browser-cli page <session-id> [-p <page>] [--next] [--prev] [--all] [--settle <ms>] [--fresh] [--json] [--verbose]
 browser-cli click <session-id> <element-id|number|query> [-p <page>] [--new-session] [--fresh] [--quiet] [--json]
 browser-cli type <session-id> <element-id|number|query> <text> [-p <page>] [--fresh] [--quiet] [--json]
 browser-cli search <session-id> <query> [--fresh] [--json] [--verbose]
@@ -188,11 +188,14 @@ browser-cli teardown [--browser chrome|firefox]
 - `t1`, `t2`, ... are IDs for truncated text blocks, readable with `text`; both `t1` and `1` are accepted
 - `b1`, `b2`, ... are IDs for paginated list/table blocks, readable with `block`; both `b1` and `1` are accepted
 - `--next` / `--prev` paginate relative to the current scroll position
+- `--all` auto-scrolls through every logical page, waits for lazy content to render, and aggregates the result into one reading view; the output is always `current=1 total=1`
+- `--settle <ms>` only applies to `page --all` and controls the fixed delay after each scroll; defaults to `500`
 - `--fresh` bypasses the Relay cache and fetches a fresh browser snapshot
 - `--version` prints the version injected at build time, or `unknown` if not set
 - `open` returns the page structure by default; use `--quiet` for session info only, `--wait 0` to skip the post-open stability wait
 - `open` / `close` / `list` / `search` / `wait` / `plugin` / `view` all support `--json`
 - `page` / `search` / `block` / `view` support `--verbose`; this mainly matters for JSON mode, where the default `--json` output is compact and `--verbose` returns full detail
+- `page --all` is optimized for full-page reading; if you need to click or type afterwards, run a normal `page` / `page -p N` again to get current actionable IDs
 - The `<target>` for `click` / `type` accepts a prefixed ID (`e1`), a bare number (`1` maps to `e1`), or a text query matching button text, link text, or input placeholder/value
 - `click` / `type` output the updated full page XML by default; use `--quiet` for a success summary, `--json` for a structured response
 - `wait` returns the latest page on success; use `--quiet` in automation pipelines when you only need the success/timeout result

README.md

@@ -149,7 +149,7 @@ browser-cli close <session-id> [--json]
 browser-cli close --all [--json]
 browser-cli --version
 
-browser-cli page <session-id> [-p <页码>] [--next] [--prev] [--fresh] [--json] [--verbose]
+browser-cli page <session-id> [-p <页码>] [--next] [--prev] [--all] [--settle <毫秒>] [--fresh] [--json] [--verbose]
 browser-cli click <session-id> <目标> [-p <页码>] [--new-session] [--fresh] [--quiet] [--json]
 browser-cli type <session-id> <目标> <文本> [-p <页码>] [--fresh] [--quiet] [--json]
 browser-cli search <session-id> <关键词> [--fresh] [--json] [--verbose]
@@ -188,11 +188,14 @@ browser-cli teardown [--browser chrome|firefox]
 - `t1`, `t2`, ... — 被截断的长文本 ID，用 `text` 命令查看完整内容；参数支持 `t1` 或 `1`
 - `b1`, `b2`, ... — 被分页的长 `list` / `table` 块 ID，用 `block` 命令继续查看后续分页；参数支持 `b1` 或 `1`
 - `--next` / `--prev` 按当前滚动位置相对翻页
+- `--all` 会按逻辑页自动滚动、等待懒加载渲染并聚合成单页阅读视图；输出固定为 `current=1 total=1`
+- `--settle <毫秒>` 仅配合 `page --all` 使用，控制每次滚动后的固定等待时间，默认 `500`
 - `--fresh` 跳过缓存，强制从浏览器获取最新快照
 - `--version` 显示构建时注入的版本号；若未注入则显示 `unknown`
 - `open` 默认会在创建会话后直接输出当前页；用 `--quiet` 只看会话信息，用 `--wait 0` 可跳过打开后的稳定等待
 - `open` / `close` / `list` / `search` / `wait` / `plugin` / `view` 全部支持 `--json`
 - `page` / `search` / `block` / `view` 支持 `--verbose`：主要用于拿完整 JSON 细节；不带时，`--json` 默认返回更紧凑的数据
+- `page --all` 更偏全量阅读视图，方便 AI/用户一次拿完整内容；如果后续要点击或输入，建议回到普通 `page` / `page -p N` 重新获取当前页元素 ID
 - `click` / `type` 的 `<目标>` 既可以是带前缀 ID（如 `e1`）、数字 ID（如 `1` 对应 `e1`），也可以是当前页交互元素的文本查询；查询会匹配按钮文本、链接文本、输入框 placeholder/value 等
 - `click` / `type` 默认会输出更新后的整页 XML；可用 `--quiet` 只看成功结果，用 `--json` 获取结构化摘要
 - `wait` 默认等待页面稳定并返回最新页面；`--for <文本>` 会轮询最新快照，直到页面里出现匹配该文本的元素

SKILL.md

@@ -15,6 +15,7 @@ description: "Drives a real browser session via browser-cli: open stateful sessi
 - Prefer `--json` when another tool or agent will consume the result.
 - Use `--verbose` with `page`, `search`, `block`, or `view` when you need full JSON detail instead of the default compact form.
 - Use `--fresh` when the cache may be stale or the page is highly dynamic.
+- Use `page --all` when you need a full reading pass over a long or lazy-loaded page; add `--settle <ms>` if the site needs more time after each auto-scroll.
 - `open`, `click`, `type`, and `wait` return the current page by default; use `--quiet` when you only need a compact success result.
 - Use `click --new-session` only for link elements with `href`. It opens the destination in a new session and keeps the source session unchanged.
 
@@ -94,6 +95,8 @@ browser-cli page s123
 browser-cli page s123 -p 2
 browser-cli page s123 --next
 browser-cli page s123 --prev
+browser-cli page s123 --all
+browser-cli page s123 --all --settle 1000 --json
 browser-cli page s123 --fresh
 ```
 
@@ -175,6 +178,7 @@ browser-cli type s123 3 "hello world" --quiet
 - Long text may be truncated in-page and assigned `t1`, `t2`, etc.; `text` accepts either `t1` or `1`.
 - Large lists or tables may be exposed as block IDs `b1`, `b2`, etc.; `block` accepts either `b1` or `1`.
 - Pagination is viewport-based. `page -p N` reads a logical page slice without scrolling the browser manually.
+- `page --all` auto-scrolls and merges all logical pages into a single reading-oriented result. Refresh with a normal `page` before acting on IDs from a live page.
 
 ## Plugin usage