claude-evolve 1.5.4 → 1.5.8

package/bin/claude-evolve-autostatus

@@ -38,6 +38,7 @@ else
   load_config
 fi
 
+
 # Run the Python autostatus script
 exec "$PYTHON_CMD" -c '
 import os
@@ -118,9 +119,27 @@ class AutoStatus:
                 "working_dir": os.path.dirname(self.csv_path)
             }
         
+        # Helpers
+        def parse_candidate_id(cid):
+            """Return (gen_num, seq_num) for ids like gen123-045; fall back to large numbers."""
+            try:
+                left, right = cid.split("-", 1)
+                gen_num = int(left[3:]) if left.startswith("gen") else 10**9
+                seq_num = int(right)
+                return gen_num, seq_num
+            except Exception:
+                return 10**9, 10**9
+
+        def is_earlier(cid_a, cid_b):
+            """True if cid_a is earlier than cid_b by generation, then sequence."""
+            ga, sa = parse_candidate_id(cid_a)
+            gb, sb = parse_candidate_id(cid_b)
+            return (ga, sa) < (gb, sb)
+
+
         # Process candidates by generation
-        all_candidates = []
         stats_by_gen = {}
+        leader = None  # Track overall leader with earliest-wins tie behavior
         
         for row in rows[1:]:  # Skip header
             if len(row) >= 1 and row[0]:  # Must have an ID
@@ -157,20 +176,24 @@ class AutoStatus:
                             description = row[2] if len(row) > 2 else "No description"
                             candidate_info = (candidate_id, description, score)
                             stats_by_gen[gen]["candidates"].append(candidate_info)
-                            all_candidates.append(candidate_info)
+
+                            # Update overall leader: highest raw score; ties -> earliest ID
+                            if leader is None or score > leader[2] or (score == leader[2] and is_earlier(candidate_id, leader[0])):
+                                leader = candidate_info
+
+                            # Update generation best: highest raw score; ties -> earliest ID
+                            if "best" not in stats_by_gen[gen]:
+                                stats_by_gen[gen]["best"] = candidate_info
+                            else:
+                                best_id, _, best_score = stats_by_gen[gen]["best"]
+                                if score > best_score or (score == best_score and is_earlier(candidate_id, best_id)):
+                                    stats_by_gen[gen]["best"] = candidate_info
                         except ValueError:
                             pass
         
-        # Find the overall leader
-        leader = None
-        if all_candidates:
-            leader = max(all_candidates, key=lambda x: x[2])
-        
-        # Find best performer in each generation
+        # Ensure every generation has a best field
         for gen in stats_by_gen:
-            if stats_by_gen[gen]["candidates"]:
-                stats_by_gen[gen]["best"] = max(stats_by_gen[gen]["candidates"], key=lambda x: x[2])
-            else:
+            if "best" not in stats_by_gen[gen]:
                 stats_by_gen[gen]["best"] = None
         
         return {
@@ -245,9 +268,17 @@ class AutoStatus:
             self.display.move_cursor(row, 1)
             print("-" * min(self.display.cols, len(header_fmt)))
             row += 1
+            # Sort generations numerically by extracting the number after "gen"
+            def gen_sort_key(gen_str):
+                """Extract numeric value from generation string for sorting."""
+                if gen_str.startswith("gen"):
+                    try:
+                        return int(gen_str[3:])
+                    except ValueError:
+                        return 999999  # Put non-numeric at end
+                return 999999
             
-            # Sort generations numerically by extracting the number after 'gen'
-            sorted_gens = sorted(generations.keys(), key=lambda g: int(g[3:]) if g.startswith("gen") and g[3:].isdigit() else 0)
+            sorted_gens = sorted(generations.keys(), key=gen_sort_key)
             
             # Calculate how many generations we can show
             available_rows = self.display.rows - row - 1  # Leave room at bottom
@@ -344,4 +375,4 @@ class AutoStatus:
 csv_path = "'"$FULL_CSV_PATH"'"
 auto_status = AutoStatus(csv_path)
 auto_status.run()
-'
+'

package/bin/claude-evolve-ideate

@@ -18,6 +18,17 @@ else
   load_config
 fi
 
+# Setup logging to file
+if [[ -n "${FULL_EVOLUTION_DIR:-}" ]]; then
+  LOG_DIR="$FULL_EVOLUTION_DIR/logs"
+  mkdir -p "$LOG_DIR"
+  LOG_FILE="$LOG_DIR/ideate-$$-$(date +%Y%m%d-%H%M%S).log"
+  
+  # Log to both terminal and file with timestamps
+  exec > >(while IFS= read -r line; do echo "$(date '+%Y-%m-%d %H:%M:%S'): $line"; done | tee -a "$LOG_FILE") 2>&1
+  echo "[IDEATE-$$] Logging to: $LOG_FILE"
+fi
+
 # Helper function to call AI with limit check
 call_ai_with_limit_check() {
   local prompt="$1"
@@ -62,6 +73,7 @@ call_claude_with_limit_check() {
 }
 
 # Robust AI calling with fallbacks across all available models
+# Returns 0 on success and echoes the successful model name to stdout
 call_ai_for_ideation() {
   local prompt="$1"
   local generation="${2:-01}"
@@ -119,8 +131,6 @@ call_ai_for_ideation() {
     ai_output=$(call_ai_model_configured "$model" "$prompt")
     local ai_exit_code=$?
     
-    echo "[AI] $model completed with exit code $ai_exit_code" >&2
-    
     # Check if the file was modified - this is ALL that matters
     if [[ -f "$temp_csv_file" ]]; then
       local new_csv_count
@@ -140,6 +150,8 @@ call_ai_for_ideation() {
           echo "[WARN] CSV format validation failed, using original" >&2
         fi
         
+        # Echo the successful model name for caller to capture
+        echo "$model"
         return 0
       else
         echo "[INFO] CSV unchanged after $model (exit code: $ai_exit_code)" >&2
@@ -210,7 +222,7 @@ fi
 
 # Ensure CSV exists
 if [[ ! -f "$FULL_CSV_PATH" ]]; then
-  echo "id,basedOnId,description,performance,status" >"$FULL_CSV_PATH"
+  echo "id,basedOnId,description,performance,status,idea-LLM,run-LLM" >"$FULL_CSV_PATH"
 fi
 
 # Validate strategy configuration
@@ -306,6 +318,7 @@ validate_direct_csv_modification() {
   local temp_csv="$1"
   local expected_count="$2"
   local idea_type="$3"
+  local ai_model="${4:-}"  # AI model that generated the ideas
   
   # Check if the file was actually modified
   if [[ ! -f "$temp_csv" ]]; then
@@ -369,6 +382,22 @@ validate_direct_csv_modification() {
   # Clean up temp file
   rm -f "$temp_csv"
   
+  # Update idea-LLM field for newly added rows if model is known
+  if [[ -n "$ai_model" ]]; then
+    echo "[INFO] Recording that $ai_model generated the ideas" >&2
+    # Get the IDs of the newly added rows (skip header line and strip quotes)
+    local new_ids
+    new_ids=$(tail -n $added_count "$FULL_CSV_PATH" | grep -v "^id," | cut -d',' -f1 | tr -d '"')
+    
+    # Update each new row with the model that generated it
+    for id in $new_ids; do
+      if [[ -n "$id" && "$id" != "id" ]]; then
+        echo "[DEBUG] Updating $id with idea-LLM = $ai_model" >&2
+        "$PYTHON_CMD" "$SCRIPT_DIR/../lib/evolution_csv.py" "$FULL_CSV_PATH" field "$id" "idea-LLM" "$ai_model" || echo "[WARN] Failed to update $id" >&2
+      fi
+    done
+  fi
+  
   # Release the lock
   release_csv_lock
   
@@ -462,6 +491,22 @@ validate_and_apply_csv_modification_old() {
   # Clean up temp file
   rm -f "$temp_csv"
   
+  # Update idea-LLM field for newly added rows if model is known
+  if [[ -n "$ai_model" ]]; then
+    echo "[INFO] Recording that $ai_model generated the ideas" >&2
+    # Get the IDs of the newly added rows (skip header line and strip quotes)
+    local new_ids
+    new_ids=$(tail -n $added_count "$FULL_CSV_PATH" | grep -v "^id," | cut -d',' -f1 | tr -d '"')
+    
+    # Update each new row with the model that generated it
+    for id in $new_ids; do
+      if [[ -n "$id" && "$id" != "id" ]]; then
+        echo "[DEBUG] Updating $id with idea-LLM = $ai_model" >&2
+        "$PYTHON_CMD" "$SCRIPT_DIR/../lib/evolution_csv.py" "$FULL_CSV_PATH" field "$id" "idea-LLM" "$ai_model" || echo "[WARN] Failed to update $id" >&2
+      fi
+    done
+  fi
+  
   # Release the lock
   release_csv_lock
   
@@ -937,7 +982,7 @@ CRITICAL: Do NOT use any git commands (git add, git commit, git reset, etc.). On
   echo "[DEBUG] AI response: $ai_response" >&2
   
   # Validate that the CSV file was actually modified
-  if ! validate_direct_csv_modification "$temp_csv" "$count" "novel"; then
+  if ! validate_direct_csv_modification "$temp_csv" "$count" "novel" "$ai_response"; then
     rm -f "$temp_csv"
     return 1
   fi
@@ -1024,7 +1069,7 @@ CRITICAL: Do NOT use any git commands (git add, git commit, git reset, etc.). On
   echo "[DEBUG] AI response: $ai_response" >&2
   
   # Validate that the CSV file was actually modified
-  if ! validate_direct_csv_modification "$temp_csv" "$count" "hill-climbing"; then
+  if ! validate_direct_csv_modification "$temp_csv" "$count" "hill-climbing" "$ai_response"; then
     rm -f "$temp_csv"
     return 1
   fi
@@ -1111,7 +1156,7 @@ CRITICAL: Do NOT use any git commands (git add, git commit, git reset, etc.). On
   echo "[DEBUG] AI response: $ai_response" >&2
   
   # Validate that the CSV file was actually modified
-  if ! validate_direct_csv_modification "$temp_csv" "$count" "structural"; then
+  if ! validate_direct_csv_modification "$temp_csv" "$count" "structural" "$ai_response"; then
     rm -f "$temp_csv"
     return 1
   fi
@@ -1198,7 +1243,7 @@ CRITICAL: Do NOT use any git commands (git add, git commit, git reset, etc.). On
   echo "[DEBUG] AI response: $ai_response" >&2
   
   # Validate that the CSV file was actually modified
-  if ! validate_direct_csv_modification "$temp_csv" "$count" "crossover"; then
+  if ! validate_direct_csv_modification "$temp_csv" "$count" "crossover" "$ai_response"; then
     rm -f "$temp_csv"
     return 1
   fi
@@ -1309,7 +1354,7 @@ CRITICAL: Do NOT use any git commands (git add, git commit, git reset, etc.). On
   echo "[DEBUG] AI response: $ai_response" >&2
   
   # Validate that the CSV file was actually modified
-  if ! validate_direct_csv_modification "$temp_csv" "$TOTAL_IDEAS" "mixed"; then
+  if ! validate_direct_csv_modification "$temp_csv" "$TOTAL_IDEAS" "mixed" "$ai_response"; then
     rm -f "$temp_csv"
     return 1
   fi

package/bin/claude-evolve-migrate-llm-columns

@@ -0,0 +1,120 @@
+#!/bin/bash
+
+set -e
+
+echo "[INFO] Migrating existing CSVs to add LLM tracking columns"
+echo "=========================================================="
+
+# Get script directory  
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# Function to add LLM columns to a CSV file
+migrate_csv() {
+  local csv_file="$1"
+  
+  echo "[INFO] Processing: $csv_file"
+  
+  # Check if file exists
+  if [[ ! -f "$csv_file" ]]; then
+    echo "[WARN] File not found: $csv_file"
+    return 1
+  fi
+  
+  # Check if it has a header
+  local header
+  header=$(head -1 "$csv_file")
+  
+  if [[ ! "$header" =~ ^id, ]]; then
+    echo "[WARN] No valid CSV header found in: $csv_file"
+    return 1
+  fi
+  
+  # Check if LLM columns already exist
+  if echo "$header" | grep -q "idea-LLM" && echo "$header" | grep -q "run-LLM"; then
+    echo "[SKIP] Already has LLM columns: $csv_file"
+    return 0
+  fi
+  
+  # Create backup
+  cp "$csv_file" "${csv_file}.bak-$(date +%Y%m%d-%H%M%S)"
+  echo "[INFO] Created backup: ${csv_file}.bak-$(date +%Y%m%d-%H%M%S)"
+  
+  # Add LLM columns to header
+  local new_header="$header,idea-LLM,run-LLM"
+  
+  # Create temporary file
+  local temp_file="${csv_file}.tmp.$$"
+  
+  # Write new header
+  echo "$new_header" > "$temp_file"
+  
+  # Copy data rows (if any) and add empty LLM columns
+  if [[ $(wc -l < "$csv_file") -gt 1 ]]; then
+    tail -n +2 "$csv_file" | while IFS= read -r line; do
+      echo "$line,," >> "$temp_file"
+    done
+  fi
+  
+  # Replace original file
+  mv "$temp_file" "$csv_file"
+  
+  echo "[SUCCESS] Added LLM columns to: $csv_file"
+  return 0
+}
+
+# Parse arguments
+if [[ $# -eq 0 ]]; then
+  echo "Usage: claude-evolve-migrate-llm-columns <csv_file> [csv_file...]"
+  echo "   OR: claude-evolve-migrate-llm-columns --all"
+  echo ""
+  echo "Adds idea-LLM and run-LLM columns to existing evolution.csv files"
+  echo ""
+  echo "Options:"
+  echo "  --all    Find and migrate all evolution.csv files in parent directories"
+  exit 1
+fi
+
+if [[ "$1" == "--all" ]]; then
+  # Find all evolution.csv files
+  echo "[INFO] Searching for evolution.csv files..."
+  
+  # Look in parent directories for evolution CSVs
+  csv_files=()
+  while IFS= read -r -d '' file; do
+    csv_files+=("$file")
+  done < <(find .. -name "evolution.csv" -type f -print0 2>/dev/null)
+  
+  if [[ ${#csv_files[@]} -eq 0 ]]; then
+    echo "[INFO] No evolution.csv files found"
+    exit 0
+  fi
+  
+  echo "[INFO] Found ${#csv_files[@]} evolution.csv files"
+  
+  # Migrate each file
+  success_count=0
+  for csv_file in "${csv_files[@]}"; do
+    if migrate_csv "$csv_file"; then
+      ((success_count++))
+    fi
+  done
+  
+  echo ""
+  echo "=========================================================="
+  echo "[INFO] Migration complete: $success_count/${#csv_files[@]} files migrated"
+  
+else
+  # Migrate specific files
+  success_count=0
+  total_count=$#
+  
+  for csv_file in "$@"; do
+    if migrate_csv "$csv_file"; then
+      ((success_count++))
+    fi
+  done
+  
+  echo ""
+  echo "=========================================================="
+  echo "[INFO] Migration complete: $success_count/$total_count files migrated"
+fi

package/bin/claude-evolve-setup

@@ -37,7 +37,7 @@ done
 # Create CSV with header
 if [[ ! -f evolution/evolution.csv ]]; then
   echo "[INFO] Creating evolution.csv with header..."
-  echo "id,basedOnId,description,performance,status" >evolution/evolution.csv
+  echo "id,basedOnId,description,performance,status,idea-LLM,run-LLM" >evolution/evolution.csv
 else
   echo "[INFO] evolution.csv already exists, skipping"
 fi

package/bin/claude-evolve-status

@@ -137,9 +137,25 @@ try:
     # Collect all candidates with scores and statuses
     all_candidates = []
     stats_by_gen = {}
+    winners_by_gen = {}
     total_stats = {'pending': 0, 'complete': 0, 'failed': 0, 'running': 0}
     retry_count = 0
     
+    def parse_candidate_id(cid):
+        try:
+            left, right = cid.split('-', 1)
+            gen_num = int(left[3:]) if left.startswith('gen') else 10**9
+            seq_num = int(right)
+            return gen_num, seq_num
+        except Exception:
+            return 10**9, 10**9
+
+    def is_earlier(a, b):
+        ga, sa = parse_candidate_id(a)
+        gb, sb = parse_candidate_id(b)
+        return (ga, sa) < (gb, sb)
+
+
     for row in rows[1:]:
         if len(row) >= 1 and row[0]:  # Must have an ID
             candidate_id = row[0]
@@ -176,13 +192,26 @@ try:
                         score = float(performance)
                         description = row[2] if len(row) > 2 else 'No description'
                         all_candidates.append((candidate_id, description, score))
+
+                        # Track per-generation best with raw score; ties -> earlier ID
+                        if gen not in winners_by_gen:
+                            winners_by_gen[gen] = (candidate_id, description, score)
+                        else:
+                            cid, _, best_score = winners_by_gen[gen]
+                            if score > best_score or (score == best_score and is_earlier(candidate_id, cid)):
+                                winners_by_gen[gen] = (candidate_id, description, score)
                     except ValueError:
                         pass
     
-    # Find the winner
+    # Find the winner using raw score; ties -> earliest ID
     winner = None
-    if all_candidates:
-        winner = max(all_candidates, key=lambda x: x[2])
+    for cid, desc, sc in all_candidates:
+        if winner is None:
+            winner = (cid, desc, sc)
+        else:
+            wc = winner[2]
+            if sc > wc or (sc == wc and is_earlier(cid, winner[0])):
+                winner = (cid, desc, sc)
     
     
     # Show winner only
@@ -250,9 +279,8 @@ try:
             data = stats_by_gen[gen]
             total = sum(data.values())
             
-            # Find best performer in this generation
-            gen_candidates = [c for c in all_candidates if c[0].startswith(gen + '-')]
-            gen_best = max(gen_candidates, key=lambda x: x[2]) if gen_candidates else None
+            # Find best performer in this generation (using precomputed winners_by_gen)
+            gen_best = winners_by_gen.get(gen)
             
             status_str = f'{data[\"pending\"]}p {data[\"complete\"]}c {data[\"failed\"]}f {data[\"running\"]}r'
             
@@ -269,4 +297,4 @@ try:
 except Exception as e:
     print(f'Error reading evolution status: {e}')
     sys.exit(1)
-"
+"

package/bin/claude-evolve-worker

@@ -7,6 +7,17 @@ source "$SCRIPT_DIR/../lib/config.sh"
 source "$SCRIPT_DIR/../lib/csv-lock.sh"
 source "$SCRIPT_DIR/../lib/ai-cli.sh"
 
+# Setup logging to file
+if [[ -n "${FULL_EVOLUTION_DIR:-}" ]]; then
+  LOG_DIR="$FULL_EVOLUTION_DIR/logs"
+  mkdir -p "$LOG_DIR"
+  LOG_FILE="$LOG_DIR/worker-$$-$(date +%Y%m%d-%H%M%S).log"
+  
+  # Log to both terminal and file with timestamps
+  exec > >(while IFS= read -r line; do echo "$(date '+%Y-%m-%d %H:%M:%S'): $line"; done | tee -a "$LOG_FILE") 2>&1
+  echo "[WORKER-$$] Logging to: $LOG_FILE"
+fi
+
 # Track current candidate for cleanup
 CURRENT_CANDIDATE_ID=""
 TERMINATION_SIGNAL=""
@@ -62,6 +73,17 @@ call_ai_for_evolution() {
   local prompt="$1"
   local candidate_id="$2"
   
+  # Get target file path from worker context
+  local target_file="$FULL_OUTPUT_DIR/evolution_${candidate_id}.py"
+  
+  # Capture file state before AI call
+  local file_hash_before=""
+  local file_mtime_before=""
+  if [[ -f "$target_file" ]]; then
+    file_hash_before=$(shasum -a 256 "$target_file" 2>/dev/null | cut -d' ' -f1)
+    file_mtime_before=$(stat -f %m "$target_file" 2>/dev/null || stat -c %Y "$target_file" 2>/dev/null)
+  fi
+  
   # Extract generation and ID numbers for round-robin calculation
   local gen_num=0
   local id_num=0
@@ -85,14 +107,32 @@ call_ai_for_evolution() {
     exit 3
   fi
   
-  if [[ $ai_exit_code -eq 0 ]]; then
-    echo "[WORKER-$$] AI succeeded" >&2
+  # Check if the target file was actually modified
+  local file_was_modified=false
+  if [[ -f "$target_file" ]]; then
+    local file_hash_after
+    local file_mtime_after
+    file_hash_after=$(shasum -a 256 "$target_file" 2>/dev/null | cut -d' ' -f1)
+    file_mtime_after=$(stat -f %m "$target_file" 2>/dev/null || stat -c %Y "$target_file" 2>/dev/null)
+    
+    if [[ "$file_hash_before" != "$file_hash_after" ]] || [[ "$file_mtime_before" != "$file_mtime_after" ]]; then
+      file_was_modified=true
+    fi
+  fi
+  
+  # Success if file was modified OR exit code is 0 (for cases where file validation isn't applicable)
+  if [[ "$file_was_modified" == "true" ]] || [[ $ai_exit_code -eq 0 ]]; then
+    if [[ "$file_was_modified" == "true" ]]; then
+      echo "[WORKER-$$] AI successfully modified $target_file (exit code: $ai_exit_code)" >&2
+    else
+      echo "[WORKER-$$] AI succeeded with exit code 0" >&2
+    fi
     # Output the result for the worker to use
     echo "$ai_output"
     return 0
   fi
   
-  echo "[WORKER-$$] All AI models failed" >&2
+  echo "[WORKER-$$] AI failed: exit code $ai_exit_code, no file changes detected" >&2
   return 1
 }
 
@@ -208,6 +248,18 @@ CRITICAL: Do NOT use any git commands (git add, git commit, git reset, etc.). On
       
       echo "[WORKER-$$] Evolution applied successfully"
       
+      # Record which AI model generated the code (regardless of evaluation outcome)
+      if [[ -n "${SUCCESSFUL_RUN_MODEL:-}" ]]; then
+        echo "[WORKER-$$] Recording that $SUCCESSFUL_RUN_MODEL generated the code" >&2
+        "$PYTHON_CMD" -c "
+import sys
+sys.path.insert(0, '$SCRIPT_DIR/..')
+from lib.evolution_csv import EvolutionCSV
+with EvolutionCSV('$FULL_CSV_PATH') as csv:
+    csv.update_candidate_field('$candidate_id', 'run-LLM', '$SUCCESSFUL_RUN_MODEL')
+" || echo "[WORKER-$$] Warning: Failed to record run-LLM field" >&2
+      fi
+      
       # Check if the generated Python file has syntax errors
       echo "[WORKER-$$] Checking Python syntax..." >&2
       if ! "$PYTHON_CMD" -m py_compile "$target_file" 2>&1; then

package/lib/ai-cli.sh

@@ -12,36 +12,50 @@ call_ai_model_configured() {
   local model_name="$1"
   local prompt="$2"
   
+  # Record start time
+  local start_time=$(date +%s)
+  
   # Build command directly based on model
   case "$model_name" in
     opus|sonnet)
       local ai_output
-      ai_output=$(timeout 300 claude --dangerously-skip-permissions --model "$model_name" -p "$prompt" 2>&1)
+      ai_output=$(timeout 180 claude --dangerously-skip-permissions --model "$model_name" -p "$prompt" 2>&1)
       local ai_exit_code=$?
       ;;
-    gpt-5)
+    gpt5high)
       local ai_output
-      ai_output=$(timeout 300 codex exec -m gpt-5 --dangerously-bypass-approvals-and-sandbox "$prompt" 2>&1)
+      ai_output=$(timeout 420 codex exec --profile gpt5high --dangerously-bypass-approvals-and-sandbox "$prompt" 2>&1)
       local ai_exit_code=$?
       ;;
-    o3)
+    o3high)
       local ai_output
-      ai_output=$(timeout 300 codex exec -m o3 --dangerously-bypass-approvals-and-sandbox "$prompt" 2>&1)
+      ai_output=$(timeout 500 codex exec --profile o3high --dangerously-bypass-approvals-and-sandbox "$prompt" 2>&1)
       local ai_exit_code=$?
       ;;
     codex)
       local ai_output
-      ai_output=$(timeout 300 codex exec --dangerously-bypass-approvals-and-sandbox "$prompt" 2>&1)
+      ai_output=$(timeout 420 codex exec --dangerously-bypass-approvals-and-sandbox "$prompt" 2>&1)
       local ai_exit_code=$?
       ;;
     gemini)
       # Debug: Show exact command
-      echo "[DEBUG] Running: timeout 300 gemini -y -p <prompt>" >&2
+      echo "[DEBUG] Running: timeout 1200 gemini -y -p <prompt>" >&2
       echo "[DEBUG] Working directory: $(pwd)" >&2
       echo "[DEBUG] Files in current dir:" >&2
       ls -la temp-csv-*.csv 2>&1 | head -5 >&2
       local ai_output
-      ai_output=$(timeout 300 gemini -y -p "$prompt" 2>&1)
+      # Gemini needs longer timeout as it streams output while working (20 minutes)
+      ai_output=$(timeout 1200 gemini -y -p "$prompt" 2>&1)
+      local ai_exit_code=$?
+      ;;
+    cursor-sonnet)
+      local ai_output
+      ai_output=$(timeout 180 cursor-agent sonnet -p "$prompt" 2>&1)
+      local ai_exit_code=$?
+      ;;
+    cursor-opus)
+      local ai_output
+      ai_output=$(timeout 300 cursor-agent opus -p "$prompt" 2>&1)
       local ai_exit_code=$?
       ;;
     *)
@@ -53,8 +67,12 @@ call_ai_model_configured() {
   # Debug: log model and prompt size
   echo "[DEBUG] Calling $model_name with prompt of ${#prompt} characters" >&2
   
-  # Always log basic info
-  echo "[AI] $model_name exit code: $ai_exit_code, output length: ${#ai_output} chars" >&2
+  # Calculate duration
+  local end_time=$(date +%s)
+  local duration=$((end_time - start_time))
+  
+  # Always log basic info with timing
+  echo "[AI] $model_name exit code: $ai_exit_code, output length: ${#ai_output} chars, duration: ${duration}s" >&2
   
   # Show detailed output if verbose or if there was an error
   if [[ "${VERBOSE_AI_OUTPUT:-false}" == "true" ]] || [[ $ai_exit_code -ne 0 ]]; then
@@ -105,7 +123,7 @@ clean_ai_output() {
   local model_name="$2"
   
   # Handle codex-specific output format
-  if [[ "$model_name" == "codex" || "$model_name" == "o3" || "$model_name" == "gpt-5" ]]; then
+  if [[ "$model_name" == "codex" || "$model_name" == "o3high" || "$model_name" == "gpt5high" ]]; then
     # Clean codex output - extract content between "codex" marker and "tokens used"
     if echo "$output" | grep -q "^\[.*\] codex$"; then
       # Extract content between "codex" line and "tokens used" line
@@ -196,11 +214,19 @@ call_ai_with_round_robin() {
     ai_output=$(call_ai_model_configured "$model" "$prompt")
     local ai_exit_code=$?
     
-    # Just check exit code
-    if [[ $ai_exit_code -eq 0 ]]; then
-      # Clean output if needed
-      ai_output=$(clean_ai_output "$ai_output" "$model")
-      echo "[AI] $model returned exit code 0" >&2
+    # Clean output if needed
+    ai_output=$(clean_ai_output "$ai_output" "$model")
+    
+    # Success if exit code is 0, or if it's just a timeout (124)
+    # Timeout doesn't mean the AI failed - it may have completed the task
+    if [[ $ai_exit_code -eq 0 ]] || [[ $ai_exit_code -eq 124 ]]; then
+      if [[ $ai_exit_code -eq 124 ]]; then
+        echo "[AI] $model timed out but continuing (exit code: 124)" >&2
+      else
+        echo "[AI] $model returned exit code 0" >&2
+      fi
+      # Export the successful model for tracking (used by worker)
+      export SUCCESSFUL_RUN_MODEL="$model"
       # Debug: log what AI returned on success
       if [[ "${DEBUG_AI_SUCCESS:-}" == "true" ]]; then
         echo "[AI] $model success output preview:" >&2

package/lib/config.sh

@@ -54,8 +54,8 @@ DEFAULT_MAX_RETRIES=3
 DEFAULT_MEMORY_LIMIT_MB=12288
 
 # Default LLM CLI configuration - use simple variables instead of arrays
-DEFAULT_LLM_RUN="sonnet gpt-5 sonnet gpt-5"
-DEFAULT_LLM_IDEATE="gemini gpt-5 opus"
+DEFAULT_LLM_RUN="sonnet cursor-sonnet"
+DEFAULT_LLM_IDEATE="gemini opus gpt5high o3high cursor-opus"
 
 # Load configuration from config file
 load_config() {
@@ -98,12 +98,14 @@ load_config() {
   # Set LLM CLI defaults (compatibility for older bash)
   # Initialize associative array for LLM commands
   # Use simpler approach for compatibility
-  LLM_CLI_gpt_5='codex exec -m gpt-5 --dangerously-bypass-approvals-and-sandbox "{{PROMPT}}"'
-  LLM_CLI_o3='codex exec -m o3 --dangerously-bypass-approvals-and-sandbox "{{PROMPT}}"'
+  LLM_CLI_gpt5high='codex exec --profile gpt5high --dangerously-bypass-approvals-and-sandbox "{{PROMPT}}"'
+  LLM_CLI_o3high='codex exec --profile o3high --dangerously-bypass-approvals-and-sandbox "{{PROMPT}}"'
   LLM_CLI_codex='codex exec --dangerously-bypass-approvals-and-sandbox "{{PROMPT}}"'
   LLM_CLI_gemini='gemini -y -p "{{PROMPT}}"'
   LLM_CLI_opus='claude --dangerously-skip-permissions --model opus -p "{{PROMPT}}"'
   LLM_CLI_sonnet='claude --dangerously-skip-permissions --model sonnet -p "{{PROMPT}}"'
+  LLM_CLI_cursor_sonnet='cursor-agent sonnet -p "{{PROMPT}}"'
+  LLM_CLI_cursor_opus='cursor-agent opus -p "{{PROMPT}}"'
   LLM_RUN="$DEFAULT_LLM_RUN"
   LLM_IDEATE="$DEFAULT_LLM_IDEATE"
   
@@ -322,12 +324,13 @@ show_config() {
   echo "  Memory limit: ${MEMORY_LIMIT_MB}MB"
   echo "  LLM configuration:"
   # Show LLM configurations using dynamic variable names
-  for model in gpt_5 o3 codex gemini opus sonnet; do
+  for model in gpt5high o3high codex gemini opus sonnet cursor_sonnet cursor_opus; do
     var_name="LLM_CLI_${model}"
-    if [[ -n "${!var_name}" ]]; then
+    var_value=$(eval echo "\$$var_name")
+    if [[ -n "$var_value" ]]; then
       # Convert underscore back to dash for display
       display_name=$(echo "$model" | sed 's/_/-/g')
-      echo "    $display_name: ${!var_name}"
+      echo "    $display_name: $var_value"
     fi
   done
   echo "  LLM for run: $LLM_RUN"

package/lib/evolution_csv.py

@@ -199,7 +199,7 @@ class EvolutionCSV:
         for i in range(start_idx, len(rows)):
             row = rows[i]
             
-            if self.is_valid_candidate_row(row) and row[0].strip() == candidate_id:
+            if self.is_valid_candidate_row(row) and row[0].strip().strip('"') == candidate_id.strip().strip('"'):
                 # Ensure row has at least 5 columns
                 while len(row) < 5:
                     row.append('')
@@ -227,7 +227,7 @@ class EvolutionCSV:
         for i in range(start_idx, len(rows)):
             row = rows[i]
             
-            if self.is_valid_candidate_row(row) and row[0].strip() == candidate_id:
+            if self.is_valid_candidate_row(row) and row[0].strip().strip('"') == candidate_id.strip().strip('"'):
                 # Ensure row has at least 4 columns
                 while len(row) < 4:
                     row.append('')
@@ -263,10 +263,14 @@ class EvolutionCSV:
                     field_index = i
                     break
             
-            # If field doesn't exist, add it to header
+            # If field doesn't exist, add it to header and extend all rows
             if field_index is None:
                 field_index = len(header_row)
                 header_row.append(field_name)
+                # Extend all data rows with empty values for the new column
+                for i in range(1, len(rows)):
+                    while len(rows[i]) <= field_index:
+                        rows[i].append('')
         else:
             # No header - we'll use predefined positions for known fields
             field_map = {
@@ -274,7 +278,9 @@ class EvolutionCSV:
                 'basedonid': 1,
                 'description': 2,
                 'performance': 3,
-                'status': 4
+                'status': 4,
+                'idea-llm': 5,
+                'run-llm': 6
             }
             field_index = field_map.get(field_name.lower())
             if field_index is None:
@@ -287,7 +293,10 @@ class EvolutionCSV:
         
         for i in range(start_idx, len(rows)):
             row = rows[i]
-            if self.is_valid_candidate_row(row) and row[0].strip() == candidate_id:
+            # Strip quotes from both stored ID and search ID for comparison
+            stored_id = row[0].strip().strip('"') if len(row) > 0 else ''
+            search_id = candidate_id.strip().strip('"')
+            if self.is_valid_candidate_row(row) and stored_id == search_id:
                 # Ensure row has enough columns
                 while len(row) <= field_index:
                     row.append('')
@@ -311,7 +320,7 @@ class EvolutionCSV:
         start_idx = 1 if rows and rows[0] and rows[0][0].lower() == 'id' else 0
         
         for row in rows[start_idx:]:
-            if self.is_valid_candidate_row(row) and row[0].strip() == candidate_id:
+            if self.is_valid_candidate_row(row) and row[0].strip().strip('"') == candidate_id.strip().strip('"'):
                 return {
                     'id': row[0].strip() if len(row) > 0 else '',
                     'basedOnId': row[1].strip() if len(row) > 1 else '',
@@ -344,7 +353,7 @@ class EvolutionCSV:
             
         for i in range(start_idx, len(rows)):
             row = rows[i]
-            if self.is_valid_candidate_row(row) and row[0].strip() == candidate_id:
+            if self.is_valid_candidate_row(row) and row[0].strip().strip('"') == candidate_id.strip().strip('"'):
                 deleted = True
                 # Skip this row (delete it)
                 continue
@@ -371,6 +380,7 @@ def main():
         print("  update <id> <status>    - Update candidate status")
         print("  perf <id> <performance> - Update candidate performance")
         print("  info <id>               - Get candidate info")
+        print("  field <id> <field> <val>- Update specific field")
         print("  check                   - Check if has pending work")
         sys.exit(1)
         
@@ -430,6 +440,17 @@ def main():
                 has_work = csv_ops.has_pending_work()
                 print("yes" if has_work else "no")
                 
+            elif command == 'field' and len(sys.argv) >= 5:
+                candidate_id = sys.argv[3]
+                field_name = sys.argv[4]
+                value = sys.argv[5] if len(sys.argv) >= 6 else ''
+                success = csv_ops.update_candidate_field(candidate_id, field_name, value)
+                if success:
+                    print(f"Updated {candidate_id} field {field_name} to {value}")
+                else:
+                    print(f"Failed to update {candidate_id} field {field_name}")
+                    sys.exit(1)
+                
             else:
                 print(f"Unknown command: {command}")
                 sys.exit(1)

package/lib/memory_limit_wrapper.py

@@ -58,12 +58,19 @@ def monitor_memory_usage_native(process: subprocess.Popen, limit_mb: int) -> Opt
                 
                 if memory_mb > limit_mb:
                     print(f"[MEMORY] Process exceeded {limit_mb}MB limit (using {memory_mb:.1f}MB), terminating", file=sys.stderr)
-                    # Kill the entire process group
+                    # Kill the entire process group - fix race condition
+                    try:
+                        pgid = os.getpgid(process.pid)
+                        os.killpg(pgid, signal.SIGTERM)
+                    except ProcessLookupError:
+                        return f"Memory limit exceeded: {memory_mb:.1f}MB > {limit_mb}MB"
+                    
+                    time.sleep(2)  # Give it time to cleanup
+                    
                     try:
-                        os.killpg(os.getpgid(process.pid), signal.SIGTERM)
-                        time.sleep(2)  # Give it time to cleanup
                         if process.poll() is None:
-                            os.killpg(os.getpgid(process.pid), signal.SIGKILL)
+                            pgid = os.getpgid(process.pid) 
+                            os.killpg(pgid, signal.SIGKILL)
                     except ProcessLookupError:
                         pass
                     return f"Memory limit exceeded: {memory_mb:.1f}MB > {limit_mb}MB"
@@ -92,12 +99,19 @@ def monitor_memory_usage(process: subprocess.Popen, limit_mb: int) -> Optional[s
                 
                 if memory_mb > limit_mb:
                     print(f"[MEMORY] Process exceeded {limit_mb}MB limit (using {memory_mb:.1f}MB), terminating", file=sys.stderr)
-                    # Kill the entire process group
+                    # Kill the entire process group - fix race condition
+                    try:
+                        pgid = os.getpgid(process.pid)
+                        os.killpg(pgid, signal.SIGTERM)
+                    except ProcessLookupError:
+                        return f"Memory limit exceeded: {memory_mb:.1f}MB > {limit_mb}MB"
+                    
+                    time.sleep(2)  # Give it time to cleanup
+                    
                     try:
-                        os.killpg(os.getpgid(process.pid), signal.SIGTERM)
-                        time.sleep(2)  # Give it time to cleanup
                         if process.poll() is None:
-                            os.killpg(os.getpgid(process.pid), signal.SIGKILL)
+                            pgid = os.getpgid(process.pid) 
+                            os.killpg(pgid, signal.SIGKILL)
                     except ProcessLookupError:
                         pass
                     return f"Memory limit exceeded: {memory_mb:.1f}MB > {limit_mb}MB"
@@ -112,6 +126,19 @@ def monitor_memory_usage(process: subprocess.Popen, limit_mb: int) -> Optional[s
     
     return None
 
+def validate_memory_limit(limit_mb: int) -> bool:
+    """Validate memory limit against system resources."""
+    if limit_mb <= 0:
+        return True  # 0 or negative means disabled
+    
+    # Basic sanity checks
+    if limit_mb < 10:
+        print(f"[MEMORY] Warning: Memory limit {limit_mb}MB is very small", file=sys.stderr)
+    elif limit_mb > 64000:
+        print(f"[MEMORY] Warning: Memory limit {limit_mb}MB is very large", file=sys.stderr)
+    
+    return True
+
 def main():
     if len(sys.argv) < 3:
         print("Usage: memory_limit_wrapper.py <memory_limit_mb> <command> [args...]", file=sys.stderr)
@@ -123,6 +150,9 @@ def main():
         print(f"Error: Invalid memory limit '{sys.argv[1]}' - must be integer MB", file=sys.stderr)
         sys.exit(1)
     
+    if not validate_memory_limit(memory_limit_mb):
+        sys.exit(1)
+    
     command = sys.argv[2:]
     
     if memory_limit_mb <= 0:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-evolve",
-  "version": "1.5.4",
+  "version": "1.5.8",
   "bin": {
     "claude-evolve": "./bin/claude-evolve",
     "claude-evolve-main": "./bin/claude-evolve-main",

package/templates/config.yaml

@@ -47,9 +47,11 @@ auto_ideate: true
 max_retries: 3
 
 # Memory protection configuration
-# Memory limit in MB for evaluation processes (0 = no limit)
+# Memory limit in MB for evaluation processes (0 = no limit)  
 # This prevents runaway algorithms from consuming all system memory
-memory_limit_mb: 2048
+# Default: 12GB (reasonable for ML workloads, adjust based on your system RAM)
+# Recommendation: Set to ~50-75% of available system RAM
+memory_limit_mb: 12288
 
 # Parallel execution configuration
 parallel:
@@ -70,5 +72,14 @@ llm_cli:
 
   # commented out because these change over time; if you want to fix them in a particular
   # configuration, uncomment them and set them
-  #run: sonnet
-  #ideate: gemini gpt-5 opus
+  #run: sonnet cursor-sonnet
+  #ideate: gemini opus gpt5high o3high cursor-opus
+  
+  # Available models:
+  # - sonnet: Claude 3.5 Sonnet via Claude CLI
+  # - opus: Claude 3 Opus via Claude CLI
+  # - gemini: Gemini via Gemini CLI
+  # - gpt5high: GPT-5 via Codex CLI (high reasoning)
+  # - o3high: O3 via Codex CLI (high reasoning)
+  # - cursor-sonnet: Claude 3.5 Sonnet via Cursor Agent CLI
+  # - cursor-opus: Claude 3 Opus via Cursor Agent CLI