@mind-harness/mh-browser 0.2.1 → 0.2.2

package/extension/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "Mind Harness",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "permissions": ["tabGroups"],
   "background": { "service_worker": "sw.js" }
 }

package/lib/index.js

@@ -2964,7 +2964,12 @@ async function launchBrowser(options) {
 			...options.headless ? {} : {
 				viewport: null,
 				ignoreDefaultArgs: ["--disable-extensions"],
-				args: ["--enable-unsafe-extension-debugging", "--keep-alive-for-test"]
+				args: [
+					"--enable-unsafe-extension-debugging",
+					"--keep-alive-for-test",
+					"--disable-blink-features=AutomationControlled",
+					"--test-type"
+				]
 			},
 			...options.browser === "chromium" ? {} : { channel: options.browser },
 			...options.executablePath !== void 0 ? { executablePath: options.executablePath } : {}
@@ -4107,6 +4112,12 @@ function handoverQuestion(reason) {
 }
 /** The option that hands the browser back. */
 const HAND_BACK = "交回";
+/** The handover card when the browser has no window: the person cannot act in it. */
+function stepQuestion(reason) {
+	return `需要你来：${reason}（浏览器没有窗口，这里做不了；写下怎么继续）`;
+}
+/** The option on that card when the person has nothing to add. */
+const REPLIED = "继续";
 //#endregion
 //#region lib/types/service/extension.js
 const PACKAGE = "@mind-harness/mh-browser";
@@ -6723,18 +6734,30 @@ function browserDownloadTool(ctx, reader) {
 }
 //#endregion
 //#region lib/types/tools/handover.js
-const HANDOVER = [
+const THEIRS = "a check that they are human, a login form, a Mind Harness login or consent form, or any form that issues credentials. Those steps are theirs: do not click through them, work around them, or fetch the page another way. A human check can clear by itself; if it is still there after one wait, call this.";
+const HEADED = [
 	"Hand the browser to the person and wait until they hand it back.",
+	"The browser is a window on the person's screen; they can see it and act in it.",
 	"The person can take over a tab at any time; that stops the action in progress on that tab, and a tab they have touched stays with them until they hand it back. Other tabs keep working.",
-	"Call it instead of filling or submitting a login form, a Mind Harness login or consent form, or any form that issues credentials, and whenever only the person can continue."
+	`Call it when a site asks for the person themselves: ${THEIRS} A handover usually takes the person seconds.`,
+	"Call it too whenever only the person can continue."
+].join(" ");
+const HEADLESS = [
+	"Tell the person about a step only they can take, and wait for their reply.",
+	"The browser has no window: the person cannot see it or act in it, so the step cannot be finished in this browser. Their reply says how to go on.",
+	`Call it when a site asks for the person themselves: ${THEIRS}`,
+	"Call it too whenever only the person can continue."
 ].join(" ");
 const REASON = "The sentence the person sees on the handover card, unchanged.";
-const TAB = "Tab id of the human page to show the person, copied from a page element's tab attribute. Omit it to hand over without naming a tab; that does not pick the latest result. A page with an Agent View has no tab until you open it with snapshot.";
+const TAB = {
+	headed: "Tab id of the human page to show the person, copied from a page element's tab attribute. Omit it to hand over without naming a tab; that does not pick the latest result. A page with an Agent View has no tab until you open it with snapshot.",
+	headless: "Tab id of the page the step is on, copied from a page element's tab attribute; the result carries that page as it is after the reply. Omit it to name no tab; that does not pick the latest result. A page with an Agent View has no tab until you open it with snapshot."
+};
 const ABORT = Symbol("abort");
-function browserHandoverTool(ctx, reader) {
+function browserHandoverTool(ctx, reader, headed = true) {
 	return defineTool({
 		name: "browser_handover",
-		description: HANDOVER,
+		description: headed ? HEADED : HEADLESS,
 		parameters: {
 			reason: {
 				type: "string",
@@ -6743,7 +6766,7 @@ function browserHandoverTool(ctx, reader) {
 			},
 			tab: {
 				type: "string",
-				description: TAB
+				description: TAB[headed ? "headed" : "headless"]
 			}
 		},
 		output: {
@@ -6785,7 +6808,7 @@ function browserHandoverTool(ctx, reader) {
 					reason: args.reason,
 					...args.tab === void 0 ? {} : { tab: args.tab }
 				}, exec.signal);
-				const answer = await askUntilAnswered(ctx, args.reason, exec);
+				const answer = await askUntilAnswered(ctx, args.reason, exec, headed);
 				if (answer === ABORT || exec.signal.aborted) return quiet();
 				const note = answer.answers[0]?.custom;
 				const back = await ctx.mhBrowser.handBack(session.id, note);
@@ -6793,11 +6816,7 @@ function browserHandoverTool(ctx, reader) {
 				const pages = page === void 0 ? [] : [page];
 				const block = tabBlock(ctx, reader, session);
 				return {
-					text: `${[
-						...note === void 0 || note === "" ? [] : [note],
-						`${String(back.operations)} operations`,
-						takeoverText(back)
-					].join("\n")}\n${renderOutcome(pages)}${block.suffix}`,
+					text: `${[...note === void 0 || note === "" ? [] : [note], ...headed ? [`${String(back.operations)} operations`, takeoverText(back)] : []].join("\n")}\n${renderOutcome(pages)}${block.suffix}`,
 					meta: JSON.parse(JSON.stringify({
 						...readingsOf(pages, "handover"),
 						closed: block.closed,
@@ -6816,14 +6835,14 @@ function browserHandoverTool(ctx, reader) {
 	});
 }
 /** Closing the card is not an answer. Ask the same question again. */
-async function askUntilAnswered(ctx, reason, exec) {
+async function askUntilAnswered(ctx, reason, exec, headed) {
 	const stopped = aborted(exec.signal);
 	for (;;) try {
 		return await Promise.race([ctx.userQuestions.ask({
 			questions: [{
 				id: "handback",
-				question: handoverQuestion(reason),
-				options: [{ label: HAND_BACK }]
+				question: headed ? handoverQuestion(reason) : stepQuestion(reason),
+				options: [{ label: headed ? HAND_BACK : REPLIED }]
 			}],
 			...exec.agent !== void 0 ? { agent: exec.agent } : {},
 			signal: exec.signal
@@ -7187,14 +7206,14 @@ function line(receipt) {
 }
 //#endregion
 //#region lib/types/tools/register.js
-function registerBrowserTools(ctx, reader) {
+function registerBrowserTools(ctx, reader, headed) {
 	followSessionTitles(ctx);
 	ctx.tools.register(browserOpenTool(ctx, reader));
 	for (const tool of pageOperationTools(ctx, reader)) ctx.tools.register(tool);
 	ctx.tools.register(browserInvokeTool(ctx, reader));
 	ctx.tools.register(browserDownloadTool(ctx, reader));
 	ctx.tools.register(browserPresenceTool(ctx, reader));
-	ctx.tools.register(browserHandoverTool(ctx, reader));
+	ctx.tools.register(browserHandoverTool(ctx, reader, headed));
 }
 //#endregion
 //#region lib/types/index.js
@@ -7221,7 +7240,7 @@ async function apply(ctx, config) {
 		inject: ["mhBrowser", ...inject],
 		async apply(inner) {
 			const reader = await registerProjections(inner);
-			registerBrowserTools(inner, reader);
+			registerBrowserTools(inner, reader, !config.headless);
 			registerPresence(inner, reader);
 		}
 	});

package/lib/types/tools/handover.d.ts

@@ -1,5 +1,5 @@
 import type { Context } from '@deepseek-ai/cordis';
 import { type ToolDefinition } from '@deepseek-ai/dsh-tools';
 import type { LogReader } from '../projections.ts';
-export declare function browserHandoverTool(ctx: Context, reader: LogReader): ToolDefinition;
+export declare function browserHandoverTool(ctx: Context, reader: LogReader, headed?: boolean): ToolDefinition;
 //# sourceMappingURL=handover.d.ts.map

package/lib/types/tools/register.d.ts

@@ -1,4 +1,4 @@
 import type { Context } from '@deepseek-ai/cordis';
 import type { LogReader } from '../projections.ts';
-export declare function registerBrowserTools(ctx: Context, reader: LogReader): void;
+export declare function registerBrowserTools(ctx: Context, reader: LogReader, headed: boolean): void;
 //# sourceMappingURL=register.d.ts.map

package/lib/types/words.d.ts

@@ -4,4 +4,8 @@ export declare const WAITING_TITLE = "\u7B49\u4F60\uFF1A";
 export declare function handoverQuestion(reason: string): string;
 /** The option that hands the browser back. */
 export declare const HAND_BACK = "\u4EA4\u56DE";
+/** The handover card when the browser has no window: the person cannot act in it. */
+export declare function stepQuestion(reason: string): string;
+/** The option on that card when the person has nothing to add. */
+export declare const REPLIED = "\u7EE7\u7EED";
 //# sourceMappingURL=words.d.ts.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mind-harness/mh-browser",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "description": "Opens pages in a real browser for the agent, keeps chosen Agent Views in context, and hands the browser to you when a person is needed.",
   "license": "SEE LICENSE IN LICENSE",
   "files": [