agentic-browser 2.1.0 → 2.3.0

package/AGENTS.md

@@ -108,7 +108,7 @@ AgenticBrowserCore → ControlApi → SessionManager → BrowserController (CDP)
 
 ## Environment Variables
 
-- `AGENTIC_BROWSER_LOG_DIR` — base dir for sessions/memory/events (default: `.agentic-browser`)
+- `AGENTIC_BROWSER_DIR` — base dir for sessions/memory/events (default: `.agentic-browser`)
 - `AGENTIC_BROWSER_CHROME_EXECUTABLE_PATH` — explicit Chrome path (auto-discovered if not set)
 - `AGENTIC_BROWSER_CDP_URL` — connect to an already-running Chrome via CDP (e.g. `http://127.0.0.1:9222`)
 - `AGENTIC_BROWSER_USER_PROFILE` — use real Chrome profile (`default`, `true`, or absolute path)

package/README.md

@@ -113,7 +113,7 @@ agentic-browser agent start --user-agent "MyBot/1.0"
 agentic-browser agent status
 agentic-browser agent navigate https://example.com
 agentic-browser agent click "#login"
-agentic-browser agent content --mode text
+agentic-browser agent content --mode summary
 agentic-browser agent content --mode html --selector main
 agentic-browser agent elements
 agentic-browser agent elements --roles button,link --limit 20
@@ -156,6 +156,8 @@ MCP responses are compact — `visible`, `actions`, and `tagName` are omitted to
 
 When an element lives inside an open shadow root or a same-origin iframe, discovery returns a composed locator using `>>>` to cross boundaries, for example `iframe[name="checkout"] >>> button[aria-label="Pay"]`. The same locator string works with `agent click`, `agent type`, MCP `browser_interact`, and fallback selectors.
 
+`agent content` and MCP `browser_get_content` now default to `summary`, which returns a low-token overview with `title`, `headings`, `landmarks`, `primaryActions`, `inputs`, `alerts`, and iframe metadata. Use `a11y` when you need the deeper accessibility structure, and `text` or `html` only when you need raw content.
+
 ````
 
 ## MCP Server
@@ -225,8 +227,8 @@ More `interact` actions:
 ### 4. Read Page Content
 
 ```bash
+agentic-browser page:content <sessionId> --mode summary
 agentic-browser page:content <sessionId> --mode title
-agentic-browser page:content <sessionId> --mode text
 agentic-browser page:content <sessionId> --mode html --selector main
 ```
 
@@ -265,7 +267,7 @@ agentic-browser memory:stats
 ## Important Notes for LLMs
 
 - Exactly **one** managed session is supported at a time.
-- Session state is persisted in `.agentic-browser/`.
+- Session state is persisted in `~/.agentic-browser/` (override with `AGENTIC_BROWSER_DIR`).
 - All commands print exactly one JSON line to `stdout`.
 - The typed `agent` commands are the preferred interface for LLMs. Use low-level `command:run` only when you need raw JSON payload control.
 - Parse only `stdout` as result object and use exit code for success/failure.

package/dist/cli/index.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { r as createCliRuntime } from "../runtime-CEvs2yzE.mjs";
+import { r as createCliRuntime } from "../runtime-BCGndjhR.mjs";
 import fs from "node:fs";
 import path from "node:path";
 import crypto from "node:crypto";
@@ -7,7 +7,7 @@ import { Command } from "commander";
 //#region src/cli/commands/agent.ts
 const STATE_FILE_NAME = "agent-state.json";
 function stateFilePath(runtime) {
-	return path.join(runtime.context.config.logDir, STATE_FILE_NAME);
+	return path.join(runtime.context.config.dataDir, STATE_FILE_NAME);
 }
 function loadState(runtime) {
 	const filePath = stateFilePath(runtime);
@@ -151,7 +151,8 @@ async function agentContent(runtime, input) {
 		action: "content",
 		...await runtime.api.getContent(sessionId, {
 			mode: input.mode,
-			selector: input.selector
+			selector: input.selector,
+			maxChars: input.maxChars
 		})
 	};
 }
@@ -249,7 +250,8 @@ async function runCommand(runtime, input) {
 async function runPageContent(runtime, input) {
 	return await runtime.api.getContent(input.sessionId, {
 		mode: input.mode,
-		selector: input.selector
+		selector: input.selector,
+		maxChars: input.maxChars
 	});
 }
 //#endregion
@@ -322,11 +324,12 @@ async function main() {
 		});
 		console.log(JSON.stringify(result));
 	});
-	program.command("page:content").argument("<sessionId>").option("--mode <mode>", "title|text|html|a11y", "text").option("--selector <selector>", "optional CSS selector").action(async (sessionId, options) => {
+	program.command("page:content").argument("<sessionId>").option("--mode <mode>", "title|text|html|a11y|summary", "summary").option("--selector <selector>", "optional CSS selector").option("--max-chars <n>", "max characters for text/html/a11y modes").action(async (sessionId, options) => {
 		const result = await runPageContent(runtime, {
 			sessionId,
 			mode: options.mode,
-			selector: options.selector
+			selector: options.selector,
+			maxChars: options.maxChars ? Number.parseInt(options.maxChars, 10) : void 0
 		});
 		console.log(JSON.stringify(result));
 	});
@@ -470,8 +473,12 @@ async function main() {
 		const result = await agentTerminate(runtime);
 		console.log(JSON.stringify(result));
 	});
-	agent.command("content").option("--mode <mode>", "title|text|html|a11y", "text").option("--selector <selector>", "optional CSS selector").action(async (options) => {
-		const result = await agentContent(runtime, options);
+	agent.command("content").option("--mode <mode>", "title|text|html|a11y|summary", "summary").option("--selector <selector>", "optional CSS selector").option("--max-chars <n>", "max characters for text/html/a11y modes").action(async (options) => {
+		const result = await agentContent(runtime, {
+			mode: options.mode,
+			selector: options.selector,
+			maxChars: options.maxChars ? Number.parseInt(options.maxChars, 10) : void 0
+		});
 		console.log(JSON.stringify(result));
 	});
 	agent.command("elements").description("List interactive elements on the current page").option("--roles <roles>", "comma-separated roles filter", (v) => v.split(",")).option("--visible-only", "only visible elements", true).option("--no-visible-only", "include hidden elements").option("--limit <n>", "max elements", "50").option("--selector <selector>", "scope to CSS selector subtree").action(async (options) => {

package/dist/index.mjs

@@ -1,3 +1,3 @@
 #!/usr/bin/env node
-import { i as createMockAgenticBrowserCore, n as createAgenticBrowserCore, t as AgenticBrowserCore } from "./runtime-CEvs2yzE.mjs";
+import { i as createMockAgenticBrowserCore, n as createAgenticBrowserCore, t as AgenticBrowserCore } from "./runtime-BCGndjhR.mjs";
 export { AgenticBrowserCore, createAgenticBrowserCore, createMockAgenticBrowserCore };

package/dist/mcp/index.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { n as createAgenticBrowserCore } from "../runtime-CEvs2yzE.mjs";
+import { n as createAgenticBrowserCore } from "../runtime-BCGndjhR.mjs";
 import { z } from "zod";
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
@@ -22,10 +22,27 @@ function truncateContent(content, limit) {
 	};
 	const originalLength = content.length;
 	return {
-		content: content.slice(0, limit) + `\n\n[Truncated — showing first ${limit} of ${originalLength} characters. Use a CSS selector to scope the content.]`,
+		content: content.slice(0, limit) + `\n\n[Truncated - showing first ${limit} of ${originalLength} characters. Use a CSS selector to scope the content, or use mode="summary" for a lower-token overview.]`,
 		truncated: true
 	};
 }
+function resolveCompactedContent(result, limit) {
+	if (result.truncated) return {
+		content: result.content,
+		truncated: true
+	};
+	return truncateContent(result.content, limit);
+}
+function parseStructuredSummary(result) {
+	if (result.structuredContent) return result.structuredContent;
+	if (!result.content) return;
+	try {
+		const parsed = JSON.parse(result.content);
+		return parsed && typeof parsed === "object" ? parsed : void 0;
+	} catch {
+		return;
+	}
+}
 function summarizeTextContent(content) {
 	const lines = uniqueLines(content);
 	return {
@@ -68,9 +85,23 @@ function compactPageContent(result, maxChars) {
 		text: 8e3,
 		a11y: 1e4,
 		html: 4e3,
-		title: void 0
+		title: void 0,
+		summary: void 0
 	}[result.mode];
-	const truncated = truncateContent(result.content, limit);
+	if (result.mode === "summary") {
+		const structured = parseStructuredSummary(result);
+		if (structured) return {
+			mode: "summary",
+			...structured,
+			truncated: false
+		};
+		return {
+			mode: "summary",
+			content: result.content,
+			truncated: false
+		};
+	}
+	const truncated = resolveCompactedContent(result, limit);
 	if (result.mode === "title") return {
 		...result,
 		truncated: false
@@ -81,6 +112,7 @@ function compactPageContent(result, maxChars) {
 			mode: result.mode,
 			content: truncated.content,
 			truncated: truncated.truncated,
+			originalLength: result.originalLength,
 			summaryLines: summary.summaryLines,
 			previewLines: summary.previewLines
 		};
@@ -91,6 +123,7 @@ function compactPageContent(result, maxChars) {
 			mode: result.mode,
 			content: truncated.content,
 			truncated: truncated.truncated,
+			originalLength: result.originalLength,
 			headings: summary.headings,
 			landmarks: summary.landmarks,
 			actions: summary.actions,
@@ -101,7 +134,8 @@ function compactPageContent(result, maxChars) {
 	return {
 		mode: result.mode,
 		content: truncated.content,
-		truncated: truncated.truncated
+		truncated: truncated.truncated,
+		originalLength: result.originalLength
 	};
 }
 function summarizeElements(elements) {
@@ -152,6 +186,14 @@ function compactMemoryResults(results) {
 			if (Array.isArray(compactAlias.fallbackSelectors) && compactAlias.fallbackSelectors.length === 0) delete compactAlias.fallbackSelectors;
 			return compactAlias;
 		});
+		if (compact.siteProfile) {
+			const profile = compact.siteProfile;
+			compact.siteProfile = {
+				selectorPatterns: (profile.selectorPatterns ?? []).map((p) => `${p.pattern} (x${p.frequency})`),
+				cookieBanner: profile.cookieBanner,
+				visitCount: profile.visitCount
+			};
+		}
 		return compact;
 	}) };
 }
@@ -159,6 +201,13 @@ function compactMemoryResults(results) {
 //#region src/mcp/index.ts
 let core;
 let activeSessionId;
+const DEFAULT_MCP_MAX_CHARS = {
+	summary: void 0,
+	title: void 0,
+	text: 8e3,
+	a11y: 1e4,
+	html: 4e3
+};
 function getCore() {
 	if (!core) core = createAgenticBrowserCore();
 	return core;
@@ -265,26 +314,29 @@ server.tool("browser_interact", "Interact with a page element or perform browser
 		text: JSON.stringify(compact)
 	}] };
 });
-server.tool("browser_get_content", "Get the current page content. Modes: \"text\" (readable text), \"a11y\" (accessibility tree — best for understanding page structure), \"title\" (page title only), \"html\" (raw HTML). Use \"a11y\" to see the full page hierarchy with roles, names, and states. A session is auto-started if needed.", {
+server.tool("browser_get_content", "Get the current page content. Modes: \"summary\" (recommended low-token overview with headings, actions, inputs, alerts, and iframe awareness), \"text\" (readable text), \"a11y\" (accessibility tree for deeper structure), \"title\" (page title only), \"html\" (raw HTML). A session is auto-started if needed.", {
 	mode: z.enum([
 		"title",
 		"text",
 		"html",
-		"a11y"
-	]).default("text").describe("Content extraction mode"),
+		"a11y",
+		"summary"
+	]).default("summary").describe("Content extraction mode"),
 	selector: z.string().optional().describe("CSS selector to scope content (e.g. \"main\", \"#content\")"),
-	maxChars: z.number().optional().describe("Maximum characters to return (default: 12000 for text/a11y, 6000 for html, no cap for title). Use a CSS selector to scope content instead of raising this limit."),
+	maxChars: z.number().optional().describe("Maximum characters to return (default: 8000 for text, 10000 for a11y, 4000 for html, ignored for summary/title). Use a CSS selector to scope content instead of raising this limit."),
 	sessionId: z.string().optional().describe("Session ID (auto-resolved if omitted)")
 }, async ({ mode, selector, maxChars, sessionId }) => {
 	const sid = await resolveSession(sessionId);
+	const effectiveMaxChars = maxChars ?? DEFAULT_MCP_MAX_CHARS[mode];
 	const result = await getCore().getPageContent({
 		sessionId: sid,
 		mode,
-		selector
+		selector,
+		maxChars: effectiveMaxChars
 	});
 	return { content: [{
 		type: "text",
-		text: JSON.stringify(compactPageContent(result, maxChars))
+		text: JSON.stringify(compactPageContent(result))
 	}] };
 });
 server.tool("browser_get_elements", "Discover all interactive elements on the current page (buttons, links, inputs, etc.). Returns CSS selectors and fallbackSelectors you can use with browser_interact. Pass fallbackSelectors to browser_interact for automatic retry when the primary selector breaks. Actions are derived from role: link/button/custom→click, input/textarea/contenteditable→click+type+press, select→click+select, checkbox/radio→toggle. A session is auto-started if needed.", {