@poncho-ai/browser 0.6.21 → 0.6.22

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/browser@0.6.21 build /home/runner/work/poncho-ai/poncho-ai/packages/browser
+> @poncho-ai/browser@0.6.22 build /home/runner/work/poncho-ai/poncho-ai/packages/browser
 > tsup src/index.ts --format esm --dts
 
 CLI Building entry: src/index.ts
@@ -7,8 +7,8 @@
 CLI tsup v8.5.1
 CLI Target: es2022
 ESM Build start
-ESM dist/index.js 46.07 KB
-ESM ⚡️ Build success in 67ms
+ESM dist/index.js 46.62 KB
+ESM ⚡️ Build success in 52ms
 DTS Build start
-DTS ⚡️ Build success in 5074ms
+DTS ⚡️ Build success in 5085ms
 DTS dist/index.d.ts 13.69 KB

package/CHANGELOG.md

@@ -1,5 +1,16 @@
 # @poncho-ai/browser
 
+## 0.6.22
+
+### Patch Changes
+
+- [`3588b19`](https://github.com/cesr/poncho-ai/commit/3588b19cf8e8fb112df3642b93e8a6aa4d4e3021) Thanks [@cesr](https://github.com/cesr)! - Steer the agent to use `browser_open` only as a last resort. The description now
+  tells it to prefer `web_fetch` for reading pages and a dedicated API/MCP
+  integration when one exists, and to reach for the browser only when those can't
+  do the job — a page web_fetch can't render, or operating a site/web app that has
+  no API and no MCP (e.g. logging in and clicking through a UI). Reinforces that
+  credentials are entered by the user in the live view, never asked for in chat.
+
 ## 0.6.21
 
 ### Patch Changes

package/dist/index.js

@@ -1079,7 +1079,7 @@ function createBrowserTools(getSession) {
   return [
     {
       name: "browser_open",
-      description: "Open a URL in a headless browser. Returns the page title. Use this to navigate to websites and web applications. To open files from the virtual filesystem, use /api/vfs/{path} (e.g. /api/vfs/downloads/report.pdf).",
+      description: "Open a URL in a real browser whose live view the user can watch and interact with. Returns the page title. This is a HEAVY, last-resort tool \u2014 prefer cheaper options first: use `web_fetch` to read page content, and use a dedicated API or MCP integration when one exists for the service. Only reach for the browser when those can't do the job: a page `web_fetch` can't render (JS-heavy/SPA), or a task that requires operating a site or web app that has no API and no MCP integration (e.g. logging in and clicking through a UI). When a site needs credentials, navigate to its login page and let the user sign in directly in the live view \u2014 never ask for passwords in chat. To open files from the virtual filesystem, use /api/vfs/{path} (e.g. /api/vfs/downloads/report.pdf).",
       inputSchema: {
         type: "object",
         properties: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/browser",
-  "version": "0.6.21",
+  "version": "0.6.22",
   "description": "Browser automation for Poncho agents, powered by agent-browser",
   "repository": {
     "type": "git",

package/src/tools.ts

@@ -10,7 +10,10 @@ export function createBrowserTools(
     {
       name: "browser_open",
       description:
-        "Open a URL in a headless browser. Returns the page title. Use this to navigate to websites and web applications. " +
+        "Open a URL in a real browser whose live view the user can watch and interact with. Returns the page title. " +
+        "This is a HEAVY, last-resort tool — prefer cheaper options first: use `web_fetch` to read page content, and use a dedicated API or MCP integration when one exists for the service. " +
+        "Only reach for the browser when those can't do the job: a page `web_fetch` can't render (JS-heavy/SPA), or a task that requires operating a site or web app that has no API and no MCP integration (e.g. logging in and clicking through a UI). " +
+        "When a site needs credentials, navigate to its login page and let the user sign in directly in the live view — never ask for passwords in chat. " +
         "To open files from the virtual filesystem, use /api/vfs/{path} (e.g. /api/vfs/downloads/report.pdf).",
       inputSchema: {
         type: "object",