Update agents/models/prompts.py

agents/models/prompts.py

@@ -5,6 +5,7 @@
 PROMPT_IMG_TO_FEN = """
 You are an expert image to FEN generation assistant with highest attention to detail.
 Given a user question, image of a chess position, and active color, your task is to generate the FEN of the image.
+If the user question mentions not to move certain pieces, remove them from the FEN.
 
 **Active Color**: {active_color}
 
@@ -18,7 +19,17 @@ Given a user question, image of a chess position, and active color, your task is
 - Provide all portions of the FEN, including active color, castling rights, en passant, and move counters.
 - Return ONLY the FEN, nothing else.
 
-**Example Output:** rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
+**Examples:**
+
+**FEN**: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
+
+**FEN**: r1bq1rk1/pp1n1ppp/2p1pn2/3p4/3P4/2N1PN2/PPQ2PPP/R1B2RK1 w KQ - 4 12
+
+**FEN**: r1bqkb1r/pppp1ppp/2n2n2/4p3/2B1P3/8/PPPP1PPP/RNBQK1NR b KQkq - 0 4
+
+**FEN**: rnbqkbnr/ppp1pppp/8/8/3pP3/8/PPP2PPP/RNBQKBNR b KQkq d3 0 2
+
+**FEN**: 8/6p1/5K2/8/8/8/7k/8 w - - 0 1
 
 **User Question:** {question}
 
@@ -42,7 +53,13 @@ Only provide your answer, no explanation.
 - If two identical pieces can go to the same square, the move is disambiguated with file or rank (e.g., Nbd2 or R1e2).
 - En passant is recorded like a standard capture (e.g., exd6 e.p.).
 
-**Example Output:** Qxb1, Ra8#
+**Examples:**
+
+**Algebraic Chess Notation:** Qxb1, Ra8#
+
+**Algebraic Chess Notation:** R1e2
+
+**Algebraic Chess Notation:** Nf3+
 
 **User Question:** {question}