dsh-ab-wechat-scrape 0.4.6 → 0.4.8

package/README.md

@@ -275,6 +275,20 @@ Arguments and result are data-dependent and resent until compaction; each excerp
 
 None beyond the ordinary request prefix.
 
+### Auto-invocation
+
+#### What the model sees
+
+When a reader's inbound message names one or more `mp.weixin.qq.com` links, the plugin's `agent/pre-step` listener detects them (through the pure `detectWechatLinks` over the free text) and, before the model's step, folds in a priming instruction that names the links and asks for `wechat_scrape`. The tool then fires on the model's next action rather than only after the model first notices the link. The detection is self-aware: it acts only when `wechat_scrape` is registered in the assembling scope, and only on the inbound user message — never on a tool result or on its own prime — so it adds no work when no link is present and never double-fires.
+
+#### Token effect
+
+Negligible: the listener scans only the last user message for links each step; when none are found it returns the step unchanged. When links are found, the prime adds one short instruction whose text is the link list.
+
+#### KV Cache effect
+
+None beyond the ordinary request prefix; the prime is appended after the cached prefix.
+
 -----
 
 <a id="known-limitations-and-deferred-work"></a>
@@ -291,6 +305,7 @@ None beyond the ordinary request prefix.
 - The minimum-body rule is a character count, not a judgement: a genuine article shorter than minBodyChars is reported instead of saved, and a long page of boilerplate that survived filtering is saved.
 - A page that carries an article title but no body is treated as an article shell and rendered, not as a list page. A list page that carries such a title needs `list: true`.
 - A withdrawal marker is read from the page head only, and only on a page with no body element; an article whose body was emptied by filtering is caught by the minimum-body rule rather than by that marker.
+- Auto-invocation recognizes only links the host check accepts (`mp.weixin.qq.com`, article or collection pages). A link the detector cannot isolate from surrounding prose — for example one split across a line break or embedded in an image — is not detected; the layer-1 routing section still covers those cases by asking the model to call the tool.
 - A block page is reported, not circumvented: when the browser render is refused as well, an expired or absent cookie still fails the call, and the remedy is a deployment cookie plus a lower request rate.
 - The cookie jar lives for one plugin instance and is never persisted, so each process starts as a fresh visitor; the deployment's configured cookie is what carries an authenticated session across restarts.
 - The playwright driver depends on a browser the deployment already has (channel msedge by default, or an explicit executable path); it does not download one.

package/lib/html.d.ts

@@ -74,6 +74,20 @@ export declare function isMpArticleUrl(url: string): boolean;
  * @returns true for an http(s) mp.weixin.qq.com /s link.
  */
 export declare function isMpArticleLink(url: string): boolean;
+/**
+ * Find every mp.weixin.qq.com link named in free text, in first-seen order and
+ * without duplicates. This is the tool's auto-invocation trigger: a reader
+ * pastes or names a WeChat link and the plugin should retrieve it without the
+ * model having to notice the link first. The scan stops at any character that is
+ * not part of a URL — crucially including CJK characters, because Chinese text
+ * has no word spacing and would otherwise swallow the words after a link into
+ * the URL. Trailing punctuation that is still not a URL character (ASCII or
+ * full-width) is then trimmed before the host check, and only links the host
+ * check accepts reach the result.
+ * @param text - the inbound message text to scan.
+ * @returns the distinct mp.weixin.qq.com links found, in first-seen order.
+ */
+export declare function detectWechatLinks(text: string): string[];
 /**
  * The block-page marker a page carries.
  * @param html - complete page HTML.

package/lib/html.js

@@ -271,6 +271,28 @@ export function isMpArticleLink(url) {
     const path = new URL(url).pathname;
     return path === '/s' || path.startsWith('/s/');
 }
+/**
+ * Find every mp.weixin.qq.com link named in free text, in first-seen order and
+ * without duplicates. This is the tool's auto-invocation trigger: a reader
+ * pastes or names a WeChat link and the plugin should retrieve it without the
+ * model having to notice the link first. The scan stops at any character that is
+ * not part of a URL — crucially including CJK characters, because Chinese text
+ * has no word spacing and would otherwise swallow the words after a link into
+ * the URL. Trailing punctuation that is still not a URL character (ASCII or
+ * full-width) is then trimmed before the host check, and only links the host
+ * check accepts reach the result.
+ * @param text - the inbound message text to scan.
+ * @returns the distinct mp.weixin.qq.com links found, in first-seen order.
+ */
+export function detectWechatLinks(text) {
+    const found = [];
+    for (const token of text.match(/https?:\/\/[A-Za-z0-9\-._~:/?#[\]@!$&*+,;=%]+/gu) ?? []) {
+        const url = token.replace(/[.,;:!?')\]。，；：！？、）]+$/u, '');
+        if (isMpArticleUrl(url) && !found.includes(url))
+            found.push(url);
+    }
+    return found;
+}
 /**
  * The block-page marker a page carries.
  * @param html - complete page HTML.

package/lib/index.d.ts

@@ -17,6 +17,14 @@
  */
 import type { Context } from '@deepseek-ai/cordis';
 import z from '@deepseek-ai/schemastery';
+import type { ContextFormed } from '@deepseek-ai/dsh-llm';
+declare module '@deepseek-ai/dsh-llm' {
+    interface MessageSourceMap {
+        'wechat-scrape': {
+            kind: 'wechat-scrape';
+        } & ContextFormed;
+    }
+}
 import type FileSystem from '@deepseek-ai/dsh-fs';
 import type { SandboxExecutionPolicy } from '@deepseek-ai/dsh-sandbox';
 import { type CleanPolicy } from './clean.ts';

package/lib/index.js

@@ -1,6 +1,7 @@
 import { isAbsolute, join, resolve } from "node:path";
 import z from "@deepseek-ai/schemastery";
 import { defineTool } from "@deepseek-ai/dsh-tools";
+import { createUserMessage } from "@deepseek-ai/dsh-llm";
 import { FsError } from "@deepseek-ai/dsh-fs";
 import { sandboxDenialMarker } from "@deepseek-ai/dsh-sandbox";
 //#region lib/html.js
@@ -250,6 +251,27 @@ function isMpArticleLink(url) {
 	return path === "/s" || path.startsWith("/s/");
 }
 /**
+* Find every mp.weixin.qq.com link named in free text, in first-seen order and
+* without duplicates. This is the tool's auto-invocation trigger: a reader
+* pastes or names a WeChat link and the plugin should retrieve it without the
+* model having to notice the link first. The scan stops at any character that is
+* not part of a URL — crucially including CJK characters, because Chinese text
+* has no word spacing and would otherwise swallow the words after a link into
+* the URL. Trailing punctuation that is still not a URL character (ASCII or
+* full-width) is then trimmed before the host check, and only links the host
+* check accepts reach the result.
+* @param text - the inbound message text to scan.
+* @returns the distinct mp.weixin.qq.com links found, in first-seen order.
+*/
+function detectWechatLinks(text) {
+	const found = [];
+	for (const token of text.match(/https?:\/\/[A-Za-z0-9\-._~:/?#[\]@!$&*+,;=%]+/gu) ?? []) {
+		const url = token.replace(/[.,;:!?')\]。，；：！？、）]+$/u, "");
+		if (isMpArticleUrl(url) && !found.includes(url)) found.push(url);
+	}
+	return found;
+}
+/**
 * The block-page marker a page carries.
 * @param html - complete page HTML.
 * @returns the marker found in the page head, or an empty string for a normal page.
@@ -2561,6 +2583,32 @@ function apply(ctx, config) {
 			rawInput: args.url ?? (args.urls ?? []).join(", ")
 		})
 	}));
+	ctx.on("agent/pre-step", async ({ signal }, next) => {
+		const decision = await next();
+		if (decision.kind === "reject" || signal.aborted) return decision;
+		const lastUser = [...decision.messages].reverse().find((message) => message.source.kind === "user");
+		if (lastUser === void 0) return decision;
+		const links = detectWechatLinks(lastUser.content.map((block) => block.type === "text" ? block.text : "").filter((part) => part !== "").join("\n"));
+		if (links.length === 0) return decision;
+		if (ctx.tools.get("wechat_scrape") === void 0) return decision;
+		return {
+			...decision,
+			messages: [createUserMessage({
+				content: [{
+					type: "text",
+					text: "检测到以下 wechat_scrape 触发链接，请调用 wechat_scrape 处理：\n" + links.join("\n")
+				}],
+				source: {
+					kind: name,
+					form: "snapshot",
+					sections: [{
+						name: SECTION_NAME,
+						text: links.join("\n")
+					}]
+				}
+			}), ...decision.messages]
+		};
+	});
 }
 //#endregion
 export { BODY_FORMATS, Config, SECTION_NAME, SECTION_ORDER, SECTION_TEXT, TOOL_NAME, albumPageUrl, apply, articleDocument, cleanPolicy, clip, collectArticle, inject, name, normalizeUrl, parseListPage, recordedArticle, renderArticle, renderScrape, requestUrls, resolveOutputDir, scrape };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-ab-wechat-scrape",
-  "version": "0.4.6",
+  "version": "0.4.8",
   "type": "module",
   "main": "lib/index.js",
   "types": "lib/index.d.ts",
@@ -66,6 +66,7 @@
   "peerDependencies": {
     "@deepseek-ai/cordis": "^4.0.4",
     "@deepseek-ai/dsh-fs": "^0.1.7-rc.1",
+    "@deepseek-ai/dsh-llm": "^0.1.7-rc.1",
     "@deepseek-ai/dsh-sandbox": "^0.1.7-rc.1",
     "@deepseek-ai/dsh-tools": "^0.1.7-rc.1"
   },
@@ -81,6 +82,7 @@
     "@deepseek-ai/dsh-sandbox-policy": "^0.1.7-rc.1",
     "@deepseek-ai/dsh-system-prompt": "^0.1.7-rc.1",
     "@deepseek-ai/dsh-tools": "^0.1.7-rc.1",
+    "@deepseek-ai/dsh-llm": "^0.1.7-rc.1",
     "@types/node": "^22.10.2",
     "prettier": "^3.3.0",
     "publint": "^0.2.0",