Refine system prompt guidelines for publication retrieval by specifying extraction of relevant information from the fetch_publications tool

Author: KacemMathlouthi

backend/app/utils/system_prompt.py

@@ -15,5 +15,5 @@
 * When the user query is about: "Describe parameters", "Describe species", "Describe reactions", or "What Applications are used?" — specifically in the context of model analysis: Make sure to use the `get_vcml_file` tool to retrieve the VCML file for the biomodel. This file contains detailed information about the model's structure and behavior, which is essential for providing accurate descriptions of parameters, species, reactions, and applications. Use also the "fetch_biomodels" tool to gather additional context about the biomodel, and Try when asked these questions to focus on the asked aspects,  Do not provide general summaries, model structure, or unrelated metadata unless explicitly requested. Keep the focus tightly on the requested element and be as technically precise as possible. Elaborate as much as you can on the requested aspect, providing detailed descriptions and explanations based on the VCML content.
 * When using mathematical expressions, wrap them properly: use `$expression$` for inline math (e.g., $k_{on}$, $\text{mmol}\cdot\text{ml}^{-1}$) and `$$expression$$` for display math blocks. Always use `\text{}` for text within math mode (e.g., $\text{Sos (Inactive)}$, $\text{concentration}$).
 * Format all units, chemical names, reaction rates, and numerical expressions using math mode to ensure proper rendering. Example: "The rate is $5.2 \times 10^{-3} \text{ mmol}\cdot\text{ml}^{-1}\cdot\text{min}^{-1}$".
-* If asked for publications, research papers, pubmed articles, etc. use the `fetch_publications` tool. After fetching, analyze results for relevance, filter by user's specific needs, format publication links using markdown `[Title](URL)`, provide context (date, authors, description), and clearly communicate if no relevant publications are found.
+* If asked for publications, research papers, pubmed articles, etc. use the `fetch_publications` tool. After fetching, extract the relevant information, filter by user's specific needs, format publication links using markdown `[Title](URL)`, provide context (date, authors, description), and clearly communicate if no relevant publications are found.
 """