@agentrhq/webcmd 0.1.2 → 0.2.0

package/skills/webcmd-adapter-author/references/site-memory.md

@@ -1,127 +1,109 @@
 # Site Memory
 
-站点记忆分两层：**in-repo 种子**（skill 自带的已知站点公共知识）+ **本地工作目录**（每台机器跑过的站点累积产物）。
+Site memory prevents every adapter run from starting cold. It has two layers:
 
----
+1. In-repo public seeds under `references/site-memory/<site>.md`, when a seed exists.
+2. Local working memory under `~/.webcmd/sites/<site>/`.
 
-## 两层结构
+Local memory is the main write target. Do not write private cookies, tokens, or user data into the repo.
 
-```
-skills/webcmd-adapter-author/references/site-memory/<site>.md
-    — 公共种子。手写 + PR 审核进入。多 agent 共享的第一批起点。
-    — 已铺：eastmoney / xueqiu / bilibili / tonghuashun
-
-~/.webcmd/sites/<site>/
-    — 本地累积。agent 跑 adapter 过程里自动写入，跨 session 复用。
-    — 不进 git，不进 PR。
-```
-
-用法：开头先读本地，命中 **不跳写 adapter**，仍要跑 Step 5 endpoint 验证 + Step 7 字段抽查（memory 可能过期或站点换版）；没命中读 in-repo；都没有走完整 recon。
-
----
-
-## Layer 1 — In-repo 种子（`references/site-memory/<site>.md`）
-
-每个覆盖站点一个 `.md`，结构固定：
-
-```markdown
-# <site>
-
-## 域名
-主 API / 备 API / 登录 / 静态资源
-
-## 默认鉴权
-`Strategy.XXX` + 必需 cookie/header + 获取方式
+## Directory Layout
 
-## 已知 endpoint（选最常用的 5-10 条）
-- `GET <url>` — 返回 X，分页参数 Y
-- ...
-
-## 字段（指向 `field-conventions.md` 的某一节）
-
-## 坑 / 陷阱
-- fltt=2 必传
-- 单位是"万"不是"元"
-- ...
-
-## 可参考的 adapter
-`clis/<site>/<name>.js` × N
-```
-
-审核门槛高，里面写的东西必须是"多数人都会踩到"的共识。一次性试错、站点局部怪癖放 Layer 2。
-
----
-
-## Layer 2 — 本地工作目录（`~/.webcmd/sites/<site>/`）
-
-agent 每跑一次相关 adapter 就可以自动写/读：
-
-```
+```text
 ~/.webcmd/sites/<site>/
-  notes.md               — 累积笔记（时间戳 + 写入人 + 发现）
-  endpoints.json         — 已验证的 endpoint 目录
-  field-map.json         — 字段代号 → 含义（key 为字段代号，value 为 {meaning, verified_at, source}）
-  verify/                — `webcmd browser verify` 期望值（值级校验锚点，每个 adapter 一份）
+  endpoints.json
+  field-map.json
+  notes.md
+  verify/
     <cmd>.json
-  fixtures/              — 完整响应样本（给字段对比 / 离线 replay；**调试时的原始 dump 也只能落在这里或 /tmp/**）
-    <cmd>-<ts>.json
-  last-probe.log         — 最近一次侦察输出（下次接着用）
+  fixtures/
+    <cmd>-<YYYYMMDDHHMM>.json
 ```
 
-`verify/` vs `fixtures/` 别混：
-- `verify/<cmd>.json` 是**结构期望**（rowCount / columns / types / patterns / notEmpty），每 adapter 一份、会被 verify 读。
-- `fixtures/<cmd>-<ts>.json` 是**原始响应样本**，给人 / 下一个 agent 做字段比对用，verify 不会读。
+## `endpoints.json`
 
-### `endpoints.json` 格式（schema 锁死）
-
-key = endpoint 的短名（`clist` / `kline` / `search` 等），不要用全 URL 当 key。
+Short endpoint name as key:
 
 ```json
 {
-  "clist": {
-    "url": "https://push2.eastmoney.com/api/qt/clist/get",
+  "search": {
+    "url": "https://example.com/api/search",
     "method": "GET",
     "params": {
-      "required": ["fs", "fields"],
-      "optional": ["pn", "pz", "fid", "po", "fltt"]
+      "required": ["q"],
+      "optional": ["page", "sort"]
+    },
+    "response": {
+      "rowsPath": "data.items",
+      "sampleFields": ["title", "url", "score"]
     },
-    "response": "data.diff[] 数组",
-    "verified_at": "2026-04-20",
-    "notes": "fltt=2 必传"
+    "verified_at": "YYYY-MM-DD",
+    "notes": "What was checked and what can drift."
   }
 }
 ```
 
-字段说明：
+Rules:
 
-- `url` / `method`：原样存，query string 不入 `url`，都归 `params`
-- `params.required` / `params.optional`：参数名列表。**不存具体值**（值会变，记例子放 `notes`）
-- `response`：一句话写清响应形状入口（`data.diff[] 数组` / `result.data.items` / `纯数组`），而不是把整个响应贴进来
-- `verified_at`：`YYYY-MM-DD`。超过 30 天下次读到当作过期重验
-- `notes`：一两句关键坑（`fltt=2 必传` / `ms 单位 begin` 之类），不要写长文
+- Re-verify memory hits before using them.
+- Treat entries older than 30 days as stale.
+- Mark changed endpoints stale instead of deleting evidence silently.
+- Never store cookies, bearer tokens, CSRF tokens, or private user data.
 
-### `field-map.json` 格式（schema 锁死）
+## `field-map.json`
 
-key = 字段代号（`f237` / `f152`），value 三件套：
+Map source codes or unclear keys to meanings:
 
 ```json
 {
-  "f237": {
-    "meaning": "convertible premium rate (%)",
-    "verified_at": "2026-04-20",
-    "source": "field-decode-playbook sort-key comparison vs page"
+  "num_comments": {
+    "meaning": "commentCount",
+    "verified_at": "YYYY-MM-DD",
+    "source": "visible page comparison"
   }
 }
 ```
 
-- `meaning`：人话 + 单位/精度（`%` / `元` / `万元` / `× 10^f152` 等）
-- `verified_at`：`YYYY-MM-DD`
-- `source`：怎么推出来的，让下次能复查（`field-decode-playbook sort-key` / `网页标签对照` / `bundle 搜索 var pricePct =`）
-- **已存在的 key 不要默默覆盖**。有冲突时先用 `fixtures/` 里的真实样本 + 网页肉眼值再确认一遍
+Rules:
+
+- Append new mappings.
+- Do not overwrite existing keys without visible-page proof.
+- If a conflict appears, compare against the visible page and record the decision in `notes.md`.
+
+## `notes.md`
+
+Prepend a dated note for each run:
+
+```md
+## YYYY-MM-DD by <agent/user>
+
+- What changed:
+- New endpoint evidence:
+- Field decoding evidence:
+- Pitfalls:
+- Follow-up:
+```
+
+Notes should capture decisions that future agents would otherwise rediscover.
+
+## `verify/<cmd>.json`
+
+This is the `webcmd browser verify` fixture.
 
-### `verify/<cmd>.json` 格式（schema 锁死）
+It should include:
 
-每个 adapter 一份，`webcmd browser verify <site>/<cmd>` 会自动读。**没有这份 = verify 只能证"能跑"，证不出数据对**——所以是必填产物。
+- args
+- rowCount
+- columns
+- types
+- patterns
+- notEmpty
+- mustNotContain
+- mustBeTruthy
+
+Write it after the first passing run, then tighten it manually and verify again.
+
+Example:
 
 ```json
 {
@@ -136,96 +118,64 @@ key = 字段代号（`f237` / `f152`），value 三件套：
       "url": "string"
     },
     "patterns": {
-      "url": "^https://www\\.1point3acres\\.com/bbs/thread-"
+      "url": "^https://www\\.example\\.com/thread-"
+    },
+    "notEmpty": ["title", "url"],
+    "mustNotContain": {
+      "title": ["breadcrumb:", "category:"]
     },
-    "notEmpty": ["title", "url"]
+    "mustBeTruthy": ["rank"]
   }
 }
 ```
 
-字段说明：
-
-- `args`：verify 调 adapter 时要带的参数。支持两种形态：
-  - **对象**：`{ "limit": 3 }` → 展开成 `--limit 3`，用于标准 named flag 适配器
-  - **数组**：`["123", "--limit", "3"]` → **原样**追加到命令后，用于 positional 主语型适配器（`<tid>` / `<url>` / `<query>`）。repo 约定"主语优先 positional"，所以这类适配器只能用数组形态
-- `expect.rowCount.{min,max}`：包含边界。稳定列表接口收紧到 `[min, max]`，动态接口给一个宽区间
-- `expect.columns`：每行必须都有这些 key（严格要求——漏了就 fail）
-- `expect.types`：支持 `|` union（`string|null`）和 `any` 通配。写多少列类型看列的稳定性；波动大的列直接 `any` 比频繁改 fixture 好
-- `expect.patterns`：正则表达式 **字符串**（注意 `\\` 转义）。`null` / `undefined` 会被跳过，不要用正则校验可空字段
-- `expect.notEmpty`：trim 后不能为空的列。这是"adapter 没吃掉核心业务字段"的最后一道保险
-- `expect.mustNotContain`：`Record<col, string[]>`。列值里不允许出现这些子串。用来挡"字段内容污染"——比如 `description` 里混进了 `address:` / `category:` 的邻居节点文字、`title` 前面粘了面包屑前缀。`notEmpty` 挡不住这种软污染
-- `expect.mustBeTruthy`：列数组。列值必须是 JS truthy。用来挡"silent `|| 0` / `|| false` 兜底"——数值列返回 0 / 空字符串 / false 都会被 `notEmpty` 放过，但业务上通常是"没抓到"
-
-### 什么时候手写 vs `--write-fixture` 自动生成
-
-- `--write-fixture` 只是种子：生成 `rowCount.min=1` / `columns` / `types`，**没有** `patterns` / `notEmpty` / `mustNotContain` / `mustBeTruthy`——纯类型 fixture 挡不住数值错位 / 字段污染 / silent fallback。
-- 拿到种子后**必须手改**，四件套一起上：
-  - `patterns`：URL / 日期 / ID 等格式列
-  - `notEmpty`：核心业务字段
-  - `mustNotContain`：描述类文本列容易被兄弟节点污染时，把禁词（`address:` / `category:` 等）列出来
-  - `mustBeTruthy`：数值 / 布尔业务列，挡 `|| 0` / `|| false`
-- adapter 是 positional 主语型（`<tid>` / `<url>` / `<query>`）时，`--write-fixture` 的 `args` 要手写成数组形态。工具不会替你决定形态。
-- 站点换版导致 fixture 过时：`--update-fixture` 覆盖。改之前**先用肉眼核对一次网页值**，别闭着眼把错的响应固化下来。
-- **规避反模式：不要为了让 verify 通过去放松 pattern**。失败的 pattern 说明 adapter 输出有问题，要收紧 adapter，不是收紧 fixture。放松 fixture 等于默认把错数据接受下来。
-
-### `notes.md` 格式
-
-```markdown
-## 2026-04-20 by webcmd-user
-写 `convertible.js` 时遇到：
-- f237 推断是溢价率（排序对比法，页面对照）
-- `fltt=2` 不加的话价格是整数 × 10^f152
-- `fs=b:MK0354` 过滤可转债
-```
-
-顶部追加新段落，老的不删。每段有日期 + 写入人。
+Field rules:
 
-### `fixtures/<cmd>-<YYYYMMDDHHMM>.json` 格式
+- `args` controls how verify invokes the adapter. Use an object such as `{ "limit": 3 }` for named flags; verify expands it to `--limit 3`.
+- Use an array such as `["1234567", "--limit", "3"]` for positional-subject adapters (`<tid>`, `<url>`, `<query>`). The array is appended exactly as written. Do not encode a positional subject as `{ "tid": "1234567", "limit": 3 }`, because that becomes `--tid 1234567 --limit 3`.
+- `expect.rowCount.{min,max}` is inclusive. Stable list APIs should use a tight range; dynamic feeds can use a wider range.
+- `expect.columns` is strict. Each row must contain every listed key.
+- `expect.types` supports `|` unions such as `string|null` and the `any` wildcard for intentionally variable fields.
+- `expect.patterns` uses regular expression strings. Remember to escape backslashes as `\\`.
+- `expect.notEmpty` trims string values and fails when core business fields are empty.
+- `expect.mustNotContain` is `Record<column, string[]>`. It blocks soft contamination such as a `description` that accidentally includes neighboring `address:` or `category:` text.
+- `expect.mustBeTruthy` lists columns whose values must be JavaScript truthy. Use it to catch silent `|| 0`, `|| false`, or empty-string fallbacks that `notEmpty` can miss on numeric or boolean business fields.
 
-一份该 endpoint 的**完整**响应样本。用途：
+Fixture workflow:
 
-- 未来字段代号再变时，拿样本和 `field-map.json` 做 regression 对比
-- 站点换版时，新响应和旧 fixture 做 diff 看哪个字段结构变了
+- `--write-fixture` is only a seed. It usually writes `rowCount.min=1`, `columns`, and `types`; it does not know the business-specific `patterns`, `notEmpty`, `mustNotContain`, or `mustBeTruthy` checks.
+- After generating the seed, tighten it manually with URL/date/ID patterns, core-field `notEmpty`, contamination guards in `mustNotContain`, truthiness guards in `mustBeTruthy`, and a realistic `rowCount`.
+- For positional-subject adapters, handwrite or correct `args` as an array because the seed cannot infer the subject shape.
+- If a site change makes the fixture stale, compare at least one visible page value before running `--update-fixture`.
+- Do not loosen fixtures just to make verify pass. A failed pattern or guard is evidence to check the adapter output first; accepting wrong data by weakening the fixture defeats the fixture.
 
-**存之前脱敏**：去掉 cookie / token / 登录态相关 header、去掉用户自己的 uid / 用户名 / 手机号 / 邮箱。
-
----
-
-## runbook 里的读/写时机
-
-```
-Step 2 开始前 → 读  ~/.webcmd/sites/<site>/
-                → 读  references/site-memory/<site>.md
-                命中后 → 不跳写 adapter，仍要跑 Step 5 (endpoint 验证) + Step 7 (字段抽查)
-                        verified_at 超 30 天 → 当作过期，按冷启动走 Step 3 → 4
-
-Step 10 verify 首轮通过后 → 写 ~/.webcmd/sites/<site>/verify/<cmd>.json
-                            - 先 `--write-fixture` 拿种子，再手改 patterns / notEmpty / rowCount
-                            - 没这份后续 verify 挡不住数据错位，**必填**
-
-Step 11 肉眼对比通过后 → 写 ~/.webcmd/sites/<site>/
-                        - endpoints.json：按 schema 追加或更新 verified_at
-                        - field-map.json：只追加新 key，已有的不默默覆盖
-                        - notes.md：顶部追加一段
-                        - fixtures/：脱敏后存一份完整响应样本（区别于 verify/）
-```
+## `fixtures/<cmd>-<YYYYMMDDHHMM>.json`
 
-**回写是 commit，不是 stash**：不过 Step 10 verify + Step 11 肉眼对比不写，防止把错的映射喂给下一轮。
+Store a sanitized response sample for field decoding and offline replay.
 
----
+Rules:
 
-## 不要写进 `~/.webcmd/sites/` 的东西
+- Remove cookies, tokens, account identifiers, private messages, emails, and private user data.
+- Keep enough response shape to decode fields later.
+- Prefer local memory for raw samples; commit repo fixtures only when they are intentional tests.
 
-- 真实账户 cookie / token — 不要保存任何鉴权凭据
-- 用户私有数据（返回体里有个人敏感字段的 → 脱敏再存 fixtures）
-- 过期超过 30 天的 last-probe.log（自动清）
+## In-Repo Seeds
 
-## 不要写进 **repo / adapter 目录** 的东西
+In-repo seeds are public knowledge only. They may contain:
 
-调试过程里的临时 dump（`.dbg-*.html` / `raw-*.json` / `sample-*` / `trace-*.txt`）**只能**落在 `~/.webcmd/sites/<site>/fixtures/` 或系统 `/tmp/`。PR diff 会把 repo 根目录和 `clis/<site>/` 下的文件一起带走——别人 review 时看到一堆调试副产物会很烦。
+- public domains
+- known endpoint shapes
+- non-secret header requirements
+- field-code conventions
+- adapter references
+- pitfalls that apply to any user
 
----
+They must not contain:
 
-## 没有 site-memory 时
+- private credentials
+- tokens
+- cookies
+- user-specific IDs
+- scraped private content
 
-新站点没对应 `.md`，也没本地目录 → 完整走 recon + discovery，跑完直接写 `~/.webcmd/sites/<site>/`，后面就有了。
+If `references/site-memory/<site>.md` is absent, proceed with local memory only.

package/skills/webcmd-adapter-author/references/site-recon.md

@@ -1,134 +1,130 @@
 # Site Recon
 
-**Layer 1：这是哪种站？** 分类完直接进 `api-discovery.md` 找 endpoint。
+**Layer 1: what kind of site is this?** Classify the site, then go directly to `api-discovery.md` to find the endpoint.
 
-本文件**只做分类**，不讲 endpoint 怎么找。
+This file only classifies sites. It does not explain how to discover endpoints.
 
----
+## One-Step Diagnosis
 
-## 一步诊断（推荐）
+Preferred command:
 
 ```bash
 webcmd browser analyze <url>
 ```
 
-返回一份 JSON：
+The command returns JSON with:
 
 ```json
 {
-  "pattern": { "pattern": "A", "reason": "3 JSON XHR responses observed", "json_responses": 3, "auth_failures": 0 },
-  "anti_bot": { "detected": false, "vendor": null, "evidence": [], "implication": "No known anti-bot signatures. Node-side fetch may work; try COOKIE first, fall back to browser-context fetch if blocked." },
-  "initial_state": { "__INITIAL_STATE__": false, "__NUXT__": false, "__NEXT_DATA__": false, "__APOLLO_STATE__": false },
-  "nearest_adapter": { "site": "xueqiu", "example_commands": ["xueqiu search", "xueqiu hot"], "reason": "2 existing adapters target this site — reuse strategy/cookie config" },
-  "recommended_next_step": "Pick the most specific JSON endpoint from `webcmd browser network` and try a bare Node fetch with cookies; escalate to browser-context fetch only if blocked."
+  "pattern": "A|B|C|D|E",
+  "anti_bot": [],
+  "api_candidates": [],
+  "nearest_adapters": [],
+  "recommended_next_step": "..."
 }
 ```
 
-`analyze` 一步把 Pattern 分类 / 反爬厂商识别 / 最近 adapter 匹配 / 下一步建议给完。直接按 `recommended_next_step` 走，多数情况不用手跑三步诊断。
+`analyze` gives Pattern classification, anti-bot signals, nearest-adapter matches, and the next step in one pass. Follow `recommended_next_step` directly in most cases.
+
+## Manual Three-Step Diagnosis
 
-## 手动三步诊断（analyze 给不出明确结论时）
+Use this only when `analyze` is ambiguous:
 
 ```bash
-webcmd browser open <url>
+webcmd browser open <url> --trace on --keep-tab true --window foreground
 webcmd browser wait time 2
-webcmd browser network
+webcmd browser network --format json
 ```
 
-看 `network` 输出判：
-
-| `network` 看到什么 | 站点类型 | 特征 |
-|------------------|---------|------|
-| 大量 `/api/...` JSON 请求，包含目标数据 | **A. SPA / JSON XHR** | React/Vue，数据走 fetch |
-| 有请求但都是广告 / 埋点，无目标数据 | **B. SSR / inline data** | 首屏在 HTML 里，深层再走 API |
-| 完全空 / 只有静态资源 | **C. JSONP / `<script src>` 驱动** | 老金融行情站常见 |
-| 有 API 但 401/403/签名错 | **D. Token / CSRF 鉴权型** | 在 A 基础上加鉴权 |
-| `Content-Type: text/event-stream` / WebSocket 握手 | **E. 流式** | 行情 tick / chat |
+Read `network` output this way:
 
-分不清时参考下面五节的其他信号。
+| `network` shows | Site type | Signals |
+| --- | --- | --- |
+| Many `/api/...` JSON requests containing target data | **A. SPA / JSON XHR** | React/Vue style app, data loaded through fetch/XHR |
+| Requests exist but are ads, analytics, or no target data | **B. SSR / inline data** | First screen data is in HTML, deeper pages may use API |
+| Empty except static resources | **C. JSONP / `<script src>` driven** | Data may arrive through script tags or callback-wrapped payloads |
+| API exists but returns 401/403 or signature errors | **D. Token / CSRF auth** | Pattern A plus auth headers or page-sourced tokens |
+| `Content-Type: text/event-stream` or WebSocket handshake | **E. Streaming** | Live feed, chat, or tick data |
 
-**数据是 SPA / 异步加载时，`wait time 2` 可能不够**。改用 `webcmd browser wait xhr '/api/path-fragment'` 直接等具体接口到场，比盲 `wait time 5` 更稳。
+If data is loaded asynchronously, `wait time 2` may not be enough. Prefer `webcmd browser wait xhr '/api/path-fragment'` for a specific interface over blind `wait time 5`.
 
 ---
 
-## Pattern A — SPA / JSON XHR
+## Pattern A - SPA / JSON XHR
 
-**代表**：xueqiu、linear、notion、大多数现代 SaaS
+**Examples:** GitHub, Linear, Notion, many modern SaaS apps.
 
-**信号**：
-- URL 一访问就是 `/`，后续数据都在 network tab
-- `document.querySelector('main').childElementCount` 一开始为 0，后被 JS 填充
-- `window.React / window.Vue / window.__REACT_DEVTOOLS_GLOBAL_HOOK__` 存在
+**Signals:**
 
-**下一步**：`api-discovery.md` §1（network 精读）
+- The initial URL loads shell HTML, then target data appears in network.
+- `document.querySelector('main').childElementCount` starts low and is later populated by JavaScript.
+- `window.React`, `window.Vue`, or `window.__REACT_DEVTOOLS_GLOBAL_HOOK__` exists.
 
-**注意 — Pattern A 命中不等于 strategy 选 `PAGE_FETCH`**：
-- 先看 `webcmd browser analyze` 输出的 `api_candidates[]`：`verdict=likely_data` 的条目才是真候选；`verdict=noise`（analytics / beacon / personalization）不能算 API 信号
-- booking #1680 反例：17 个 JSON XHR 看起来像 Pattern A，但全是 analytics side-channel，最终走 `DOM_STATE` / `UI_SELECTOR`
-- replay 候选 endpoint 后，按 `strategy-selection.md` 的契约模型选 strategy；`PUBLIC_API` / `COOKIE_API` 都不通才考虑 `PAGE_FETCH`
+**Next step:** `api-discovery.md` section 1, network deep read.
 
----
+**Important:** Pattern A does not automatically mean `PAGE_FETCH`.
 
-## Pattern B — SSR / inline data
+- First inspect `api_candidates[]` from `webcmd browser analyze`: only `verdict=likely_data` entries are real candidates. `verdict=noise` entries such as analytics, beacons, or personalization do not count as API signals.
+- The booking #1680 counterexample had many JSON XHRs that looked like Pattern A, but they were analytics side-channels; the final strategy was `DOM_STATE` / `UI_SELECTOR`.
+- After replaying a candidate endpoint, choose strategy through `strategy-selection.md`. Consider `PAGE_FETCH` only after `PUBLIC_API` and `COOKIE_API` fail.
 
-**代表**：bilibili 个人主页、小红书、微博、部分 Next.js / Nuxt 页
+## Pattern B - SSR / Inline Data
 
-**信号**：
-- 第一个请求（`document`）返回的 HTML 里已经含目标数据（`curl <url> | grep <某数字>`）
-- `window.__INITIAL_STATE__` / `window.__NEXT_DATA__` / `window.__NUXT__` 存在
-- 关 JS 仍能看到首屏数据
+**Examples:** Reddit post pages, YouTube watch pages, many Next.js / Nuxt pages.
 
-**下一步**：`api-discovery.md` §2（state 抽取） + §1（深层数据回到 network）
+**Signals:**
 
----
+- The first `document` response already contains target data (`curl <url> | grep <known-value>`).
+- `window.__INITIAL_STATE__`, `window.__NEXT_DATA__`, or `window.__NUXT__` exists.
+- The first screen is still visible with JavaScript disabled.
 
-## Pattern C — JSONP / `<script src>` 驱动
+**Next step:** `api-discovery.md` section 2 for state extraction, plus section 1 when deeper data returns to network.
 
-**代表**：eastmoney、tonghuashun、老一代金融站
+## Pattern C - JSONP / `<script src>` Driven
 
-**信号**：
-- `network` 空或只有 css/font
-- 页面上肯定有数据（价格、成交量等）
-- `document.querySelectorAll('script[src]')` 里有指向 `push / api / data` 域名的 src
-- 响应是 `jQuery123({...})` 这种回调包裹（JSONP）
+**Examples:** older quote pages, legacy directory pages, callback-wrapped data feeds.
 
-**下一步**：`api-discovery.md` §3（bundle / script src 搜索）
+**Signals:**
 
----
+- `network` is empty or mostly CSS/fonts.
+- The page clearly displays data such as price, count, or volume.
+- `document.querySelectorAll('script[src]')` includes URLs under `push`, `api`, or `data` style domains.
+- Response is callback-wrapped, such as `callback123({...})`.
 
-## Pattern D — Token / CSRF / Bearer
+**Next step:** `api-discovery.md` section 3, bundle / script src search.
 
-**代表**：Twitter/X、部分企业 SaaS
+## Pattern D - Token / CSRF Auth
 
-**信号**：
-- 已经是 Pattern A，但 `fetch(url, {credentials:'include'})` 返回 401/403
-- network 里请求头有 `X-Csrf-Token / Authorization: Bearer / X-Client-Id / X-Workspace-Id` 等自定义字段
-- 401 响应体带 `{"code":"AUTH_REQUIRED","csrf":"..."}` 类提示
+**Examples:** Twitter/X and some enterprise SaaS apps.
 
-**下一步**：`api-discovery.md` §4（token 来源排查） + §5（store action / intercept 降级）
+**Signals:**
 
----
+- It is otherwise Pattern A, but `fetch(url, { credentials: 'include' })` returns 401/403.
+- Network requests contain custom headers such as `X-Csrf-Token`, `Authorization: Bearer`, `X-Client-Id`, or `X-Workspace-Id`.
+- 401 responses include hints like `{"code":"AUTH_REQUIRED","csrf":"..."}`.
 
-## Pattern E — 流式
+**Next step:** `api-discovery.md` section 4 token-source investigation, then section 5 store-action / intercept fallback if needed.
 
-**代表**：行情 tick、LLM chat
+## Pattern E - Streaming
 
-**信号**：
-- `network` 里有 `101 Switching Protocols`（WebSocket 握手）
-- Response headers 含 `Content-Type: text/event-stream`
-- 请求一直 pending 不结束
+**Examples:** LLM chat streams, live feeds, real-time dashboards.
 
-**下一步**：先找**同数据的 HTTP 轮询接口**（90% 概率有）。真没有再走 intercept 收 N 条。
+**Signals:**
 
----
+- `network` contains `101 Switching Protocols`.
+- Response headers include `Content-Type: text/event-stream`.
+- The request stays pending.
+
+**Next step:** first find an HTTP polling endpoint with the same data. Most sites have one. Use intercept only when no polling endpoint exists.
 
-## 识别失败怎么办
+## If Classification Fails
 
-诊断信号互相矛盾（比如 network 非空但目标数据不在里面），按优先级硬走：
+When diagnostic signals conflict, such as non-empty network with no target data, use this priority order:
 
-1. 先当 A，试 `api-discovery.md` §1
-2. 不行当 B，试 §2
-3. 还不行当 C，试 §3
-4. 401 出现了切 D，试 §4
-5. 所有手段都试过，启动 intercept（§5）
+1. Treat as A and try `api-discovery.md` section 1.
+2. If that fails, treat as B and try section 2.
+3. If that fails, treat as C and try section 3.
+4. If 401/403 appears, switch to D and try section 4.
+5. After all other paths fail, use intercept from section 5.
 
-不要纠结分类。分类是帮忙定第一步，没命中就按顺序降级。
+Do not get stuck debating classification. Classification chooses the first move; fallback order handles misses.