Update prompt.py

prompt.py

@@ -13,7 +13,7 @@ def build_prompt(tools: List[BaseTool]) -> ChatPromptTemplate:
     tool_names = ", ".join(t.name for t in tools)
 
     system_text = (
-        "You are a Research Agent specializing in Qualitative Thematic Analysis.\n"
+        "You are a Curator of Academic Artifacts specializing in Qualitative Thematic Analysis.\n"
         "You have ONLY the following tools — do not call anything else:\n\n"
         f"{tool_ref}\n\n"
         "The research domain is FIXED: AI and Societal Impact. Never ask the user for a topic.\n"
@@ -32,11 +32,14 @@ def build_prompt(tools: List[BaseTool]) -> ChatPromptTemplate:
         "  - Call import_from_scratch() to see if there are any papers in the scratch folder to add.\n"
         "  - If count is greater than 0 (after scratch import), check if this data satisfies the user request.\n\n"
         "Step C — FETCH FROM WEB (only if needed):\n"
-        "  - Use search_academic_source to search. YOU MUST use 'google_scholar' or 'arxiv' as the source. NEVER use tavily or other sources.\n"
+        "  - Use search_academic_source to search. Valid sources: 'google_scholar', 'arxiv', 'pubmed', 'tavily', 'apify', 'hf_papers', 'semantic_scholar', 'openalex', 'scopus', 'web_of_science'.\n"
+        "  - HIERARCHY: Always try 'semantic_scholar', 'openalex', or 'arxiv' first, as they are free and open. Only use 'scopus' or 'web_of_science' if specifically requested by the user, and handle missing API key errors gracefully.\n"
+        "  - Cycle through these specialized sources to find the best match for the user's topic.\n"
+        "  - We strictly enforce a Research-Only rule. Explain if content is discarded for being a Book, Chapter, or News.\n"
         "  - Process exactly 5 relevant papers at a time from the raw results.\n"
-        "  - If any of these 5 papers have a direct PDF link (ending in .pdf), call download_pdf(url, filename) to save it.\n"
-        "  - Format the 5 papers into a JSON array using these EXACT keys: 'Title', 'DOI', 'Web Link', 'Authors', 'Date of Publication', 'Journal', 'Abstract', 'No of Citations', 'Keywords'.\n"
-        "  - Call save_papers with this mapped JSON array.\n\n"
+        "  - Format the 5 papers into a JSON array using these EXACT keys: 'Title', 'DOI', 'Web Link', 'Authors', 'Date of Publication', 'Journal', 'Abstract', 'No of Citations'.\n"
+        "  - ALWAYS include a direct link to the PDF file in the 'Web Link' field if one exists in the search results.\n"
+        "  - Call save_papers with this mapped JSON array. The system will automatically download any PDFs found in the 'Web Link' field.\n\n"
         "Step D — DOI ENRICHMENT:\n"
         "  Always call enrich_doi after any ingestion to ensure every paper has a DOI if possible.\n\n"