zan-browser 3.0.58 → 3.0.59

package/dist/agent/prompt.js

@@ -44,7 +44,7 @@ At every step, reason about what you have and what gets you closest to the data:
 - If PROGRESS SUMMARY shows 3+ visited domains and best signal is null or below 30, pause and reason in your assessment: is the problem the sources or the seed value? Use web_search to verify the seed value exists as a real-world entity right now. If it doesn't, switch to a more common, stable example of the same type — note the change in your plan — and continue searching with the new value.
 - Parameterizability check: when you find an endpoint via fetch_url, consider whether the seed value appears in the URL. If it doesn't, the user won't be able to construct that URL for other inputs — prefer endpoints where the seed value is visible in the path or query params.
 - Before switching domains or declaring impossible, consider the page you're already on: if the browser URL contains the seed value and the page shows relevant data, you may already have what you need — try scrape or extract_dom on the current page, or eval_js to check for embedded app state (e.g. __NEXT_DATA__, __NUXT__). These are often faster paths than hunting for a hidden API.
-- When you navigate to a URL that contains the seed value in the path, exhaust that page before moving on. Try these three extraction methods in order: (1) read_network_logs for XHRs with score ≥ 50 that contain the seed value in the URL — if found, verify with fetch_url; (2) scrape the page if it shows relevant data in the DOM; (3) eval_js to check for embedded app state (window.__NEXT_DATA__, window.__NUXT__, etc.). You may skip a method only if you have clear evidence it won't work (e.g. no XHRs fired at all, page is blank, no JS framework detected). Do not leave a seed-matching page without trying all viable methods.
+- When you navigate to a URL that contains the seed value in the path, exhaust that page before moving on. First, observe the page. If observe() shows elements that are clearly the data you're looking for (lists of items, tables, cards, repeated structures matching the goal) — use scrape immediately as your next action. Visible data in the DOM is direct evidence; do not look for APIs first. Only if scrape fails or returns empty, continue with: (1) read_network_logs for XHRs with score ≥ 50 that contain the seed value in the URL — if found, verify with fetch_url; (2) eval_js to check for embedded app state (window.__NEXT_DATA__, window.__NUXT__, etc.). You may skip a method only if you have clear evidence it won't work. Do not leave a seed-matching page without trying all viable methods.
 
 ─── Response Format ───────────────────────────────────────────────────────────
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "zan-browser",
-  "version": "3.0.58",
+  "version": "3.0.59",
   "description": "AI-powered cloud browser library with observe-first, screenshot-as-fallback pattern",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",