@pheem49/mint 1.14.0 → 1.15.0

package/crates/mint-core/src/browser/mod.rs

@@ -3,7 +3,8 @@
 //! click, type, and screenshot a page.
 //!
 //! Split by concern:
-//! - `cdp` — the raw CDP websocket transport
+//! - `session` — run-scoped tab leases, observations, verification, and recovery
+//! - `cdp` — the pinned, bounded CDP websocket transport
 //! - `lifecycle` — spawning/detecting the automation browser instance
 //! - `navigate` — navigation and page-state reading (tabs, text, screenshot)
 //! - `interact` — selector-driven click/type
@@ -12,18 +13,31 @@
 //! - `logging` — the `browser-automation.log` tail `mint auto` reads from
 
 mod cdp;
+mod evidence;
 mod input;
 mod interact;
 mod lifecycle;
 mod logging;
 mod navigate;
 mod overlay;
+mod session;
+
+pub(crate) use lifecycle::BROWSER_TOOLS;
+pub use session::{
+    BrowserActionResult, BrowserObservation, BrowserSession, execute_action, run_session, with_tab,
+};
+pub(crate) use session::{
+    finish_evidence as session_finish_evidence, last_state as session_last_state, selected_tab,
+};
 
 use serde::Serialize;
 
 pub use input::{key_press, mouse_click, mouse_move, type_text_native};
 pub use interact::{click, get_element_coordinates, type_text};
-pub use lifecycle::{ensure_page_open, is_browser_running, spawn_automation_browser};
+pub use lifecycle::{
+    enable_browser_tools, ensure_page_open, is_browser_running, spawn_automation_browser,
+    spawn_automation_browser_with_url,
+};
 pub use navigate::{list_tabs, navigate, read_page_text, screenshot};
 
 #[derive(Debug, Clone, Serialize)]

package/crates/mint-core/src/browser/navigate.rs

@@ -34,18 +34,23 @@ pub async fn list_tabs(config: &MintConfig) -> Result<Vec<BrowserTab>, String> {
         .collect())
 }
 
-pub async fn navigate(config: &MintConfig, url: &str) -> Result<String, String> {
-    if !(url.starts_with("https://") || url.starts_with("http://")) {
-        log_action(
-            "NAVIGATE_ERROR",
-            "Browser navigation only supports http and https URLs",
-        );
+pub(super) fn validate_url(url: &str) -> Result<(), String> {
+    let parsed = reqwest::Url::parse(url).map_err(|_| "invalid_browser_url".to_string())?;
+    if !matches!(parsed.scheme(), "http" | "https") || parsed.host_str().is_none() {
         return Err("browser navigation only supports http and https URLs".into());
     }
+    Ok(())
+}
+
+pub async fn navigate(config: &MintConfig, url: &str) -> Result<String, String> {
+    validate_url(url)?;
     log_action("NAVIGATE", &format!("Navigating to {url}"));
     ensure_page_open(config).await?;
     match cdp_call(config, "Page.navigate", json!({ "url": url })).await {
         Ok(response) => {
+            if let Some(error) = response["result"]["errorText"].as_str() {
+                return Err(format!("navigation_failed: {error}"));
+            }
             if response["result"]["frameId"].as_str().is_some() {
                 wait_for_page_load(config).await;
                 log_action(
@@ -84,21 +89,24 @@ pub async fn navigate(config: &MintConfig, url: &str) -> Result<String, String>
 /// etc.) never settle to `"complete"`, so this gives up after ~8s and lets
 /// the caller proceed rather than blocking the agent loop indefinitely.
 pub(super) async fn wait_for_page_load(config: &MintConfig) {
-    tokio::time::sleep(std::time::Duration::from_millis(150)).await;
-    for _ in 0..40 {
-        if let Ok(response) = cdp_call_raw(
-            config,
-            "Runtime.evaluate",
-            json!({ "expression": "document.readyState", "returnByValue": true }),
-        )
-        .await
-        {
-            if response["result"]["result"]["value"].as_str() == Some("complete") {
-                return;
+    let _ = tokio::time::timeout(std::time::Duration::from_secs(8), async {
+        tokio::time::sleep(std::time::Duration::from_millis(150)).await;
+        for _ in 0..40 {
+            if let Ok(response) = cdp_call_raw(
+                config,
+                "Runtime.evaluate",
+                json!({ "expression": "document.readyState", "returnByValue": true }),
+            )
+            .await
+            {
+                if response["result"]["result"]["value"].as_str() == Some("complete") {
+                    return;
+                }
             }
+            tokio::time::sleep(std::time::Duration::from_millis(200)).await;
         }
-        tokio::time::sleep(std::time::Duration::from_millis(200)).await;
-    }
+    })
+    .await;
 }
 
 pub async fn read_page_text(config: &MintConfig) -> Result<String, String> {

package/crates/mint-core/src/browser/observe.js

@@ -0,0 +1,40 @@
+// Keep actual DOM nodes rather than reconstructing selectors from labels.
+const elements = new Map();
+const all = [];
+const nodes = globalThis.__mintNodes ||= {ids:new WeakMap(),next:0,documentId:(crypto.randomUUID?.() || String(performance.timeOrigin)+Math.random())};
+function walk(root) {
+  for (const el of root.querySelectorAll('*')) {
+    if (el.id?.startsWith('mint-')) continue;
+    if (el.matches('a[href],button,input:not([type=hidden]),textarea,select,[role=button],[role=link],[role=checkbox],[role=combobox],[contenteditable=true]')) {
+      const style = getComputedStyle(el), rect = el.getBoundingClientRect();
+      if (rect.width && rect.height && style.display !== 'none' && style.visibility !== 'hidden') all.push(el);
+    }
+    if (el.shadowRoot) walk(el.shadowRoot);
+  }
+}
+walk(document);
+const page = all.slice(elementOffset, elementOffset + 100).map((el, index) => {
+  const ref = observationId + ':' + (elementOffset + index);
+  elements.set(ref, el);
+  if (!nodes.ids.has(el)) nodes.ids.set(el, ++nodes.next);
+  const tag = el.tagName.toLowerCase();
+  const labelled = (el.getAttribute('aria-labelledby') || '').split(/\s+/).filter(Boolean)
+    .map(id => el.getRootNode().getElementById?.(id)?.textContent || '').join(' ');
+  const name = el.getAttribute('aria-label') || labelled || Array.from(el.labels || []).map(l => l.textContent).join(' ')
+    || el.innerText || el.getAttribute('placeholder') || el.getAttribute('title') || '';
+  const role = el.getAttribute('role') || ({a:'link',button:'button',textarea:'textbox',select:'combobox'})[tag]
+    || (el.type === 'checkbox' ? 'checkbox' : el.type === 'radio' ? 'radio' : ['submit','button'].includes(el.type) ? 'button' : 'textbox');
+  const rect = el.getBoundingClientRect();
+  return {ref,nodeId:nodes.documentId+':'+nodes.ids.get(el),role,name:name.trim().slice(0,120),inputType:el.type || null,
+    value:el.type === 'password' ? null : (el.value == null ? null : String(el.value).slice(0,300)),checked:el.checked ?? null,
+    enabled:!el.disabled && el.getAttribute('aria-disabled') !== 'true',visible:true,
+    inViewport:rect.bottom > 0 && rect.top < innerHeight && rect.right > 0 && rect.left < innerWidth};
+});
+globalThis.__mintObservation = {id:observationId,elements,document};
+const text = document.body?.innerText || '';
+return {tabId:'',url:location.href,title:document.title,observationId,scrollX,scrollY,
+  text:text.slice(textOffset,textOffset+12000),elements:page,
+  textTruncated:textOffset+12000<text.length,elementsTruncated:elementOffset+100<all.length,
+  nextTextOffset:textOffset+12000<text.length?textOffset+12000:null,
+  nextElementOffset:elementOffset+100<all.length?elementOffset+100:null,
+  unsupportedFrames:document.querySelectorAll('iframe,frame').length};