claude-evolve 1.4.4 → 1.4.6

package/README.md

@@ -63,17 +63,35 @@ your-project/
 
 ## Evaluator Requirements
 
-Your `evaluator.py` must output a performance score to stdout:
+Your `evaluator.py` must output a performance score to stdout. The system supports multiple output formats:
 
+### Format 1: Simple numeric value
 ```python
-# Simple: just print a number
+# Just print a single number
 print(1.234)
+```
+
+### Format 2: JSON with performance/score field
+```python
+# JSON with 'performance' field (recommended)
+print('{"performance": 1.234, "accuracy": 0.95, "latency": 45.2}')
 
-# Advanced: JSON with metrics
-print('{"performance": 1.234, "accuracy": 0.95}')
+# OR JSON with 'score' field
+print('{"score": 1.234, "precision": 0.88, "recall": 0.92}')
+```
+
+### Format 3: SCORE: prefix (legacy)
+```python
+# For backward compatibility
+print("SCORE: 1.234")
 ```
 
-Higher scores = better performance. Score of 0 = failure.
+**Important notes:**
+- Higher scores = better performance
+- When using JSON, all fields are saved to the CSV for analysis
+- The `performance` or `score` field is required for evolution decisions
+- Return value of 0 doesn't mean failure - it's just a low score
+- Actual failures should exit with non-zero status code
 
 ## Configuration
 

package/bin/claude-evolve-worker

@@ -123,14 +123,96 @@ Important: Make meaningful changes that match the description. Don't just add co
     local eval_end=$(date +%s)
     local eval_duration=$((eval_end - eval_start))
     
-    # Extract performance score
-    local score=$(echo "$eval_output" | grep -E "^SCORE:" | cut -d: -f2 | tr -d ' ')
+    # Extract performance score - support multiple formats
+    # Try to parse the output and extract score
+    local score_and_json=$("$PYTHON_CMD" -c "
+import sys
+import json
+import re
+
+output = '''$eval_output'''
+
+# Try different formats
+score = None
+json_data = None
+
+# Format 1: Simple numeric value (just a number on a line)
+for line in output.strip().split('\n'):
+    line = line.strip()
+    if line and not line.startswith('{'):
+        try:
+            score = float(line)
+            break
+        except ValueError:
+            pass
+
+# Format 2: JSON with 'performance' or 'score' field
+if score is None:
+    for line in output.strip().split('\n'):
+        line = line.strip()
+        if line.startswith('{'):
+            try:
+                data = json.loads(line)
+                json_data = data
+                if 'performance' in data:
+                    score = float(data['performance'])
+                elif 'score' in data:
+                    score = float(data['score'])
+                break
+            except (json.JSONDecodeError, ValueError, KeyError):
+                pass
+
+# Format 3: SCORE: prefix (backward compatibility)
+if score is None:
+    match = re.search(r'^SCORE:\s*([+-]?\d*\.?\d+)', output, re.MULTILINE)
+    if match:
+        try:
+            score = float(match.group(1))
+        except ValueError:
+            pass
+
+# Output results
+if score is not None:
+    print(f'SCORE={score}')
+    if json_data:
+        print('JSON_DATA=' + json.dumps(json_data))
+else:
+    print('SCORE=NONE')
+")
     
-    if [[ -n "$score" ]]; then
+    # Parse the Python output
+    local score=""
+    local json_data=""
+    while IFS= read -r line; do
+      if [[ "$line" =~ ^SCORE=(.*)$ ]]; then
+        score="${BASH_REMATCH[1]}"
+      elif [[ "$line" =~ ^JSON_DATA=(.*)$ ]]; then
+        json_data="${BASH_REMATCH[1]}"
+      fi
+    done <<< "$score_and_json"
+    
+    if [[ "$score" != "NONE" ]] && [[ -n "$score" ]]; then
       echo "[WORKER-$$] Evaluation complete: score=$score (${eval_duration}s)"
       
       # Update CSV with result
-      "$PYTHON_CMD" -c "
+      if [[ -n "$json_data" ]]; then
+        # If we have JSON data, update all fields
+        "$PYTHON_CMD" -c "
+import sys
+import json
+sys.path.insert(0, '$SCRIPT_DIR/..')
+from lib.evolution_csv import EvolutionCSV
+
+json_data = json.loads('$json_data')
+with EvolutionCSV('$FULL_CSV_PATH') as csv:
+    csv.update_candidate_status('$candidate_id', 'complete')
+    # Update all fields from JSON
+    for key, value in json_data.items():
+        csv.update_candidate_field('$candidate_id', key, str(value))
+"
+      else
+        # Simple score only
+        "$PYTHON_CMD" -c "
 import sys
 sys.path.insert(0, '$SCRIPT_DIR/..')
 from lib.evolution_csv import EvolutionCSV
@@ -138,6 +220,7 @@ with EvolutionCSV('$FULL_CSV_PATH') as csv:
     csv.update_candidate_status('$candidate_id', 'complete')
     csv.update_candidate_performance('$candidate_id', '$score')
 "
+      fi
     else
       echo "[WORKER-$$] ERROR: No score found in evaluation output" >&2
       echo "[WORKER-$$] Output: $eval_output" >&2

package/lib/evolution_csv.py

@@ -240,6 +240,66 @@ class EvolutionCSV:
             
         return updated
         
+    def update_candidate_field(self, candidate_id: str, field_name: str, value: str) -> bool:
+        """Update a specific field for a candidate by adding it as a new column if needed."""
+        rows = self._read_csv()
+        if not rows:
+            return False
+            
+        # Check if we have a header row
+        has_header = rows and rows[0] and rows[0][0].lower() == 'id'
+        header_row = rows[0] if has_header else None
+        
+        # Find or add the field to header
+        if has_header:
+            # Normalize field names - lowercase for comparison
+            field_lower = field_name.lower()
+            field_index = None
+            
+            # Try to find existing column
+            for i, col in enumerate(header_row):
+                if col.lower() == field_lower:
+                    field_index = i
+                    break
+            
+            # If field doesn't exist, add it to header
+            if field_index is None:
+                field_index = len(header_row)
+                header_row.append(field_name)
+        else:
+            # No header - we'll use predefined positions for known fields
+            field_map = {
+                'id': 0,
+                'basedonid': 1,
+                'description': 2,
+                'performance': 3,
+                'status': 4
+            }
+            field_index = field_map.get(field_name.lower())
+            if field_index is None:
+                # Unknown field without header - can't update
+                return False
+        
+        # Update the candidate's field
+        updated = False
+        start_idx = 1 if has_header else 0
+        
+        for i in range(start_idx, len(rows)):
+            row = rows[i]
+            if self.is_valid_candidate_row(row) and row[0].strip() == candidate_id:
+                # Ensure row has enough columns
+                while len(row) <= field_index:
+                    row.append('')
+                    
+                row[field_index] = value
+                updated = True
+                break
+                
+        if updated:
+            self._write_csv(rows)
+            
+        return updated
+        
     def get_candidate_info(self, candidate_id: str) -> Optional[Dict[str, str]]:
         """Get information about a specific candidate."""
         rows = self._read_csv()

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-evolve",
-  "version": "1.4.4",
+  "version": "1.4.6",
   "bin": {
     "claude-evolve": "./bin/claude-evolve",
     "claude-evolve-main": "./bin/claude-evolve-main",