claude-evolve 1.7.3 → 1.7.4

package/bin/claude-evolve-ideate

@@ -893,6 +893,8 @@ Current evolution context:
 - Algorithm: algorithm.py (base algorithm)
 - Brief: $(head -5 "$FULL_BRIEF_PATH" 2>/dev/null | head -c 500 || echo "No brief file found")
 
+IMPORTANT: DO NOT read algorithm.py or any evolution_*.py files. Focus on creative ideation based on the brief and CSV context only. Reading code files wastes tokens and time.
+
 CRITICAL INSTRUCTIONS:
 1. Use the Read tool to examine the current CSV file
    IMPORTANT: If the CSV file is large (>200 lines), read it in chunks using the offset and limit parameters to avoid context overload
@@ -994,10 +996,19 @@ IMPORTANT: You MUST use one of these exact parent IDs: $valid_parent_ids
 Successful algorithms to tune:
 $top_performers
 
-OPTIONAL: If needed for context, you may read parent algorithm source files at: evolution_<PARENT_ID>.py
-However, you can often generate good parameter tuning ideas just from the descriptions and performance scores above.
-IMPORTANT: If you read .py files, read them in chunks using offset and limit parameters to avoid context overload
-Example: Read(file_path='evolution_gen01-001.py', offset=0, limit=100) then Read(offset=100, limit=100), etc.
+IMPORTANT: Generate parameter tuning ideas based primarily on the descriptions and scores above.
+
+ONLY read parent source files (evolution_<PARENT_ID>.py) if:
+- The description is too vague to identify specific parameters
+- You need to verify actual parameter names/values
+- Reading is absolutely necessary for meaningful tuning
+
+If you must read source files:
+- Read in small chunks (offset/limit) to minimize token usage
+- Focus only on finding parameter definitions at the top of the file
+- Do NOT read the entire implementation
+
+Most of the time, you can infer parameters from descriptions like "RSI with threshold 30" or "MA period 20".
 
 CRITICAL INSTRUCTIONS:
 1. Use the Read tool to examine the current CSV file
@@ -1082,10 +1093,11 @@ IMPORTANT: You MUST use one of these exact parent IDs: $valid_parent_ids
 Successful algorithms to modify structurally:
 $top_performers
 
-OPTIONAL: If needed for context, you may read parent algorithm source files at: evolution_<PARENT_ID>.py
-However, you can often generate good structural ideas just from the descriptions and performance scores above.
-IMPORTANT: If you read .py files, read them in chunks using offset and limit parameters to avoid context overload
-Example: Read(file_path='evolution_gen01-001.py', offset=0, limit=100) then Read(offset=100, limit=100), etc.
+IMPORTANT: DO NOT read evolution_*.py files. Generate structural ideas based ONLY on:
+- The algorithm descriptions above
+- The performance scores
+- Your knowledge of common algorithmic structures and patterns
+Reading code files wastes tokens and time. Focus on high-level architectural ideas based on the descriptions.
 
 CRITICAL INSTRUCTIONS:
 1. Use the Read tool to examine the current CSV file
@@ -1170,10 +1182,11 @@ IMPORTANT: You MUST use ONLY these exact parent IDs: $valid_parent_ids
 Top performers to combine (reference at least 2 in each idea):
 $top_performers
 
-OPTIONAL: If needed for context, you may read parent algorithm source files at: evolution_<PARENT_ID>.py
-However, you can often generate good hybrid ideas just from the descriptions and performance scores above.
-IMPORTANT: If you read .py files, read them in chunks using offset and limit parameters to avoid context overload
-Example: Read(file_path='evolution_gen01-001.py', offset=0, limit=100) then Read(offset=100, limit=100), etc.
+IMPORTANT: DO NOT read evolution_*.py files. Generate crossover ideas based ONLY on:
+- The algorithm descriptions above (which describe their key features)
+- The performance scores
+- Your knowledge of how different algorithmic approaches can be combined
+Reading code files wastes tokens and time. Focus on combining the described features creatively.
 
 CRITICAL INSTRUCTIONS:
 1. Use the Read tool to examine the current CSV file
@@ -1260,9 +1273,7 @@ Current evolution context:
 - Algorithm: algorithm.py (base algorithm)
 - Brief: $(head -10 "$FULL_BRIEF_PATH" 2>/dev/null | head -c 1000 || echo "No brief file found")
 
-IMPORTANT: DO NOT read all evolution files - that uses too many tokens. Just generate creative ideas based on the brief.
-IMPORTANT: If you read .py or .csv files, read them in chunks using offset and limit parameters to avoid context overload
-Example: Read(file_path='temp-csv-123.csv', offset=0, limit=100) then Read(offset=100, limit=100), etc."
+IMPORTANT: DO NOT read algorithm.py or any evolution_*.py files - that uses too many tokens and is unnecessary for ideation. Just generate creative ideas based on the brief and top performers listed above. Focus your creativity on the problem space, not the implementation details."
 
   if [[ -n $top_performers ]]; then
     prompt+="

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-evolve",
-  "version": "1.7.3",
+  "version": "1.7.4",
   "bin": {
     "claude-evolve": "./bin/claude-evolve",
     "claude-evolve-main": "./bin/claude-evolve-main",