diff --git a/backend/apps/agents/browser/browser_schema.py b/backend/apps/agents/browser/browser_schema.py index 255caadf..e9825228 100644 --- a/backend/apps/agents/browser/browser_schema.py +++ b/backend/apps/agents/browser/browser_schema.py @@ -535,6 +535,16 @@ SYSTEM_PROMPT = ( "ONLY when the page clearly contradicts your plan. Verbose per-turn prose is the single " "biggest thing that slows you down, so once the plan exists, keep execution turns short.\n\n" + "## Jump straight to deep URLs\n" + "The fastest navigation is a URL you construct yourself. Most sites expose search " + "and entities as URL patterns: a LinkedIn people search is " + "linkedin.com/search/results/people/?keywords=NAME, a Google search is " + "google.com/search?q=..., a YouTube search is youtube.com/results?search_query=..., " + "an Amazon search is amazon.com/s?k=.... When the task names a thing to find, " + "BrowserNavigate DIRECTLY to the site's search-results URL for it instead of loading " + "the homepage and driving its search UI; one navigate replaces three clicks and two " + "waits. Fall back to the UI only when you don't know the site's URL pattern.\n\n" + "## Required output structure: ReportProgress before every action\n" "Before ANY action tool (BrowserClick, BrowserType, BrowserNavigate, " "BrowserPressKey, BrowserScroll, BrowserEvaluate, BrowserClickIndex, " @@ -595,7 +605,10 @@ SYSTEM_PROMPT = ( "accessibility tree sees roles and names that the raw DOM doesn't, even on sites " "like Tinder, Instagram, and TikTok that use unlabeled