claude-evolve 1.3.25 → 1.3.27

package/README.md

@@ -203,6 +203,32 @@ your-project/
 └── (your main project files)
 ```
 
+## Evaluator Output Format
+
+Your evaluator must output a performance score to stdout. Three formats are supported:
+
+### 1. Plain Number (Simplest)
+Just output a single floating-point number:
+```
+1.077506371224117
+```
+
+### 2. JSON with "score" field
+```json
+{"score": 0.95}
+```
+
+### 3. JSON with "performance" field
+```json
+{"performance": 1.234}
+```
+
+**Important notes:**
+- Higher scores indicate better performance
+- A score of 0 indicates complete failure
+- Non-zero exit codes indicate evaluation errors
+- Any additional output (warnings, logs) should go to stderr, not stdout
+
 ## Environment Variables for Evaluators
 
 When your evaluator.py runs, it has access to the `EXPERIMENT_ID` environment variable containing the current experiment's ID (e.g., `gen07-001`). This allows evaluators to:
@@ -221,7 +247,10 @@ experiment_id = os.environ.get('EXPERIMENT_ID', 'unknown')
 
 # Use it for logging or file naming
 output_file = f"results_{experiment_id}.json"
-print(f"Evaluating experiment: {experiment_id}")
+print(f"Evaluating experiment: {experiment_id}", file=sys.stderr)  # Use stderr for logs!
+
+# Output just the score
+print(score)  # Simple number to stdout
 ```
 
 ## Configuration

package/bin/claude-evolve-ideate

@@ -78,7 +78,6 @@ call_claude_with_limit_check() {
 
 # Parse arguments
 use_strategies=true
-no_ai=false
 
 while [[ $# -gt 0 ]]; do
   case $1 in
@@ -87,11 +86,10 @@ while [[ $# -gt 0 ]]; do
 claude-evolve ideate - Generate new algorithm ideas using evolutionary strategies
 
 USAGE:
-  claude-evolve ideate [--legacy N] [--no-ai]
+  claude-evolve ideate [--legacy N]
 
 OPTIONS:
   --legacy N  Use legacy mode with N ideas (ignores strategy config)
-  --no-ai     Use manual entry mode instead of AI generation
   --help      Show this help message
 
 DESCRIPTION:
@@ -116,10 +114,6 @@ EOF
       exit 1
     fi
     ;;
-  --no-ai)
-    no_ai=true
-    shift
-    ;;
   *)
     echo "[ERROR] Unknown option: $1" >&2
     exit 1
@@ -195,21 +189,6 @@ get_next_id() {
   printf "gen%s-%03d" "$generation" $((max_id + 1))
 }
 
-# Add idea to CSV manually (fallback for manual mode)
-add_idea_manual() {
-  local description="$1"
-  local based_on_id="$2"
-  local generation="$3"
-  local id
-  id=$(get_next_id "$generation")
-
-  # Escape quotes in description
-  local escaped_desc="${description//\"/\"\"}"
-
-  # Append to CSV
-  echo "${id},${based_on_id},\"${escaped_desc}\",," >>"$FULL_CSV_PATH"
-  echo "[INFO] Added idea: $description"
-}
 
 # Get top performers for parent selection
 get_top_performers() {
@@ -246,47 +225,12 @@ with open('$FULL_CSV_PATH', 'r') as f:
 "
 }
 
-# Manual entry mode
-ideate_manual() {
-  local ideas_added=0
-
-  for ((i = 1; i <= TOTAL_IDEAS; i++)); do
-    if [[ $TOTAL_IDEAS -eq 1 ]]; then
-      read -r -p "Enter algorithm idea (or empty to skip): " description
-    else
-      read -r -p "Enter algorithm idea $i/$TOTAL_IDEAS (or empty to skip): " description
-    fi
-
-    if [[ -z $description ]]; then
-      echo "[INFO] Empty description, skipping idea"
-      continue
-    fi
-
-    add_idea_manual "$description" "" "$CURRENT_GENERATION"
-    ((ideas_added++))
-
-    if [[ $i -lt $TOTAL_IDEAS ]]; then
-      read -r -p "Add another idea? (y/N) " continue_adding
-      if [[ $continue_adding != "y" && $continue_adding != "Y" ]]; then
-        break
-      fi
-    fi
-  done
-
-  echo "[INFO] Added $ideas_added idea(s) to $EVOLUTION_CSV"
-}
 
 # Generate ideas using AI with multi-strategy approach
 ideate_ai_strategies() {
-  # Check for AI CLI (codex or claude)
-  if ! command -v codex >/dev/null 2>&1 && ! command -v claude >/dev/null 2>&1; then
-    echo "[WARN] No AI CLI found (codex or claude). Falling back to manual entry."
-    return 1
-  fi
-
   if [[ ! -f "$FULL_BRIEF_PATH" ]]; then
-    echo "[WARN] $BRIEF_FILE not found. Falling back to manual entry."
-    return 1
+    echo "[ERROR] $BRIEF_FILE not found. Run 'claude-evolve setup' first." >&2
+    exit 1
   fi
 
   # Get top performers
@@ -342,6 +286,16 @@ Requirements for new CSV rows:
 - Each description should be one clear sentence describing a specific algorithmic change
 - Descriptions should explore completely different approaches than existing ones
 - All new rows should have empty performance and status fields
+
+CRITICAL CSV FORMAT RULES:
+- DO NOT modify the CSV header row
+- DO NOT change the column order
+- DO NOT add extra columns or fields
+- DO NOT modify existing rows - only append new ones
+- DO NOT add extra blank lines or formatting
+- Maintain exact CSV format: id,basedOnId,description,performance,status
+- Leave performance and status fields completely empty (just commas)
+- Use proper CSV quoting only when descriptions contain commas
 - CRITICAL: You must read existing evolution files to avoid suggesting changes that:
   * Have already been tried and failed
   * Are impossible given the codebase structure
@@ -395,6 +349,16 @@ Requirements for new CSV rows:
 - Each description should be one clear sentence about parameter tuning
 - Focus on adjusting hyperparameters, thresholds, sizes, learning rates
 - All new rows should have empty performance and status fields
+
+CRITICAL CSV FORMAT RULES:
+- DO NOT modify the CSV header row
+- DO NOT change the column order
+- DO NOT add extra columns or fields
+- DO NOT modify existing rows - only append new ones
+- DO NOT add extra blank lines or formatting
+- Maintain exact CSV format: id,basedOnId,description,performance,status
+- Leave performance and status fields completely empty (just commas)
+- Use proper CSV quoting only when descriptions contain commas
 - CRITICAL: You must read the parent algorithm file to understand:
   * What parameters are actually tunable in the code
   * What changes made this algorithm successful vs its parent
@@ -448,6 +412,16 @@ Requirements for new CSV rows:
 - Each description should be one clear sentence about architectural changes
 - Keep core insights but change implementation approach
 - All new rows should have empty performance and status fields
+
+CRITICAL CSV FORMAT RULES:
+- DO NOT modify the CSV header row
+- DO NOT change the column order
+- DO NOT add extra columns or fields
+- DO NOT modify existing rows - only append new ones
+- DO NOT add extra blank lines or formatting
+- Maintain exact CSV format: id,basedOnId,description,performance,status
+- Leave performance and status fields completely empty (just commas)
+- Use proper CSV quoting only when descriptions contain commas
 - CRITICAL: You must read the parent algorithm file to understand:
   * What structural elements can be modified within the codebase constraints
   * What architectural decisions led to this algorithm's success
@@ -501,6 +475,16 @@ Requirements for new CSV rows:
 - Each description should be one clear sentence combining elements from different algorithms
 - Be specific about what elements to merge
 - All new rows should have empty performance and status fields
+
+CRITICAL CSV FORMAT RULES:
+- DO NOT modify the CSV header row
+- DO NOT change the column order
+- DO NOT add extra columns or fields
+- DO NOT modify existing rows - only append new ones
+- DO NOT add extra blank lines or formatting
+- Maintain exact CSV format: id,basedOnId,description,performance,status
+- Leave performance and status fields completely empty (just commas)
+- Use proper CSV quoting only when descriptions contain commas
 - CRITICAL: You must read the relevant algorithm files to:
   * Identify the specific improvements that made each algorithm successful
   * Understand which components are compatible for merging
@@ -523,15 +507,9 @@ Add exactly $count hybrid combination rows to the CSV file now."
 
 # Legacy AI generation mode (for backward compatibility)
 ideate_ai_legacy() {
-  # Check for AI CLI (codex or claude)
-  if ! command -v codex >/dev/null 2>&1 && ! command -v claude >/dev/null 2>&1; then
-    echo "[WARN] No AI CLI found (codex or claude). Falling back to manual entry."
-    return 1
-  fi
-
   if [[ ! -f "$FULL_BRIEF_PATH" ]]; then
-    echo "[WARN] $BRIEF_FILE not found. Falling back to manual entry."
-    return 1
+    echo "[ERROR] $BRIEF_FILE not found. Run 'claude-evolve setup' first." >&2
+    exit 1
   fi
 
   # Get top performers (pure shell)
@@ -576,6 +554,16 @@ Requirements for new CSV rows:
 - Each description should be one clear sentence describing an algorithmic approach
 - All new rows should have empty performance and status fields
 
+CRITICAL CSV FORMAT RULES:
+- DO NOT modify the CSV header row
+- DO NOT change the column order
+- DO NOT add extra columns or fields
+- DO NOT modify existing rows - only append new ones
+- DO NOT add extra blank lines or formatting
+- Maintain exact CSV format: id,basedOnId,description,performance,status
+- Leave performance and status fields completely empty (just commas)
+- Use proper CSV quoting only when descriptions contain commas
+
 Add exactly $TOTAL_IDEAS algorithm variation rows to the CSV file now."
 
   echo "[INFO] Generating $TOTAL_IDEAS ideas (legacy mode)..."
@@ -591,19 +579,12 @@ CURRENT_GENERATION=$(get_next_generation)
 echo "[INFO] Starting ideation for generation $CURRENT_GENERATION"
 
 # Main execution
-if [[ $no_ai == true ]]; then
-  echo "[INFO] Manual entry mode"
-  ideate_manual
-elif [[ $use_strategies == true ]]; then
+if [[ $use_strategies == true ]]; then
   echo "[INFO] Multi-strategy AI generation mode"
-  if ! ideate_ai_strategies; then
-    echo "[INFO] Falling back to manual entry"
-    ideate_manual
-  fi
+  ideate_ai_strategies
+  echo "[INFO] Ideation complete! Check $EVOLUTION_CSV for new ideas."
 else
   echo "[INFO] Legacy AI generation mode"
-  if ! ideate_ai_legacy; then
-    echo "[INFO] Falling back to manual entry" 
-    ideate_manual
-  fi
+  ideate_ai_legacy
+  echo "[INFO] Ideation complete! Check $EVOLUTION_CSV for new ideas."
 fi

package/bin/claude-evolve-worker

@@ -239,7 +239,23 @@ fi
 
 # Process results
 if [[ $eval_exit_code -eq 0 ]]; then
-  # Extract score
+  # First, check if output is just a plain number
+  if [[ $eval_output =~ ^[[:space:]]*-?[0-9]+\.?[0-9]*[[:space:]]*$ ]]; then
+    score=$(echo "$eval_output" | tr -d ' ')
+    if (( $(echo "$score == 0" | bc -l) )); then
+      update_csv_row_with_lock "$candidate_id" "status" "failed"
+      update_csv_row_with_lock "$candidate_id" "performance" "$score"
+      echo "[WORKER-$$] ✗ Evaluation failed with score 0"
+      exit 1
+    else
+      update_csv_row_with_lock "$candidate_id" "performance" "$score"
+      update_csv_row_with_lock "$candidate_id" "status" "complete"
+      echo "[WORKER-$$] ✓ Evaluation complete, score: $score"
+      exit 0
+    fi
+  fi
+  
+  # Try JSON "score" field
   if score=$(echo "$eval_output" | grep -o '"score"[[:space:]]*:[[:space:]]*[0-9.]*' | cut -d: -f2 | tr -d ' '); then
     if [[ -n $score ]]; then
       if (( $(echo "$score == 0" | bc -l) )); then
@@ -256,7 +272,7 @@ if [[ $eval_exit_code -eq 0 ]]; then
     fi
   fi
   
-  # Try "performance" field
+  # Try JSON "performance" field
   if score=$(echo "$eval_output" | grep -o '"performance"[[:space:]]*:[[:space:]]*[0-9.]*' | cut -d: -f2 | tr -d ' '); then
     if [[ -n $score ]]; then
       if (( $(echo "$score == 0" | bc -l) )); then
@@ -274,6 +290,7 @@ if [[ $eval_exit_code -eq 0 ]]; then
   fi
   
   echo "[ERROR] No score found in evaluator output" >&2
+  echo "[ERROR] Expected: plain number (e.g., 1.23) or JSON with 'score' or 'performance' field" >&2
   update_csv_row_with_lock "$candidate_id" "status" "failed"
   exit 1
 else

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-evolve",
-  "version": "1.3.25",
+  "version": "1.3.27",
   "bin": {
     "claude-evolve": "./bin/claude-evolve",
     "claude-evolve-main": "./bin/claude-evolve-main",

package/templates/evaluator.py

@@ -40,11 +40,10 @@ def evaluate_performance(algorithm_module):
     end_time = time.time()
     execution_time = end_time - start_time
     
-    return {
-        "execution_time": execution_time,
-        "score": 1.0 / execution_time if execution_time > 0 else 0,
-        "status": "success"
-    }
+    # Calculate a performance score (higher is better)
+    score = 1.0 / execution_time if execution_time > 0 else 0
+    
+    return score  # Simple: just return the number
 
 
 def main():
@@ -60,15 +59,18 @@ def main():
     
     try:
         algorithm_module = load_algorithm(algorithm_file)
-        metrics = evaluate_performance(algorithm_module)
-        print(json.dumps(metrics))
+        score = evaluate_performance(algorithm_module)
+        
+        # Option 1: Just print the number (simplest)
+        print(score)
+        
+        # Option 2: Print as JSON (if you need more structure)
+        # print(json.dumps({"score": score}))
+        
         sys.exit(0)
     except Exception as e:
-        error_result = {
-            "error": str(e),
-            "status": "failed"
-        }
-        print(json.dumps(error_result))
+        # Log errors to stderr, not stdout
+        print(f"Error: {e}", file=sys.stderr)
         sys.exit(1)