claude-evolve 1.3.39 → 1.3.40

package/README.md

@@ -55,6 +55,7 @@ claude-evolve setup  # Initialize evolution workspace
 claude-evolve ideate # Generate new algorithm ideas
 claude-evolve run    # Execute evolution candidates
 claude-evolve analyze # Analyze evolution results
+claude-evolve edit   # Manage candidate statuses by generation
 claude-evolve config # Manage configuration settings
 ```
 
@@ -90,6 +91,18 @@ Analyzes evolution progress and generates insights:
 - Best-performing algorithm variants
 - Suggestions for future evolution directions
 
+#### claude-evolve-edit
+Manages candidate statuses by generation for re-evaluation workflows:
+- Mark generations as failed, complete, or pending
+- Reset entire generations (delete files and clear scores)
+- Essential for re-running evaluations when algorithms or evaluators change
+
+```bash
+claude-evolve edit gen03 failed    # Mark all gen03 as failed
+claude-evolve edit all pending     # Mark everything as pending for re-run
+claude-evolve edit gen02 reboot    # Full reset of gen02 (delete files + clear data)
+```
+
 #### claude-evolve-config
 Manages configuration settings:
 - View current configuration
@@ -203,6 +216,30 @@ your-project/
 └── (your main project files)
 ```
 
+## Evolution CSV Format
+
+The evolution.csv file tracks all candidates and their results. The core columns are:
+
+**Required columns (positions 1-5):**
+1. **id** - Unique identifier for each candidate (e.g., gen01-001, gen02-015)
+2. **basedOnId** - Parent algorithm this was derived from (empty for novel ideas)
+3. **description** - What changes this variant implements
+4. **performance** - Score from evaluator (column 4) - this drives evolution selection
+5. **status** - Current state: empty/"pending", "running", "complete", "failed", "timeout"
+
+**Additional columns:**
+- Any other metrics your evaluator outputs (fitness, sharpe, total_return, yearly_return, max_drawdown, volatility, etc.)
+- Error messages for failed runs
+- Execution time
+
+**Key behaviors:**
+- The system uses column 4 (performance) for evolution selection, regardless of its name
+- Column 5 (status) determines what needs to be run
+- Empty or "pending" status means the candidate is ready to run
+- You can reset a candidate to pending by deleting all fields after the description (columns 4+)
+- Additional columns are automatically added when your evaluator returns JSON with extra fields
+- Rows with fewer than 5 fields are treated as pending candidates
+
 ## Evaluator Output Format
 
 Your evaluator must output a performance score to stdout. The system looks for either `performance` or `score` fields. Four formats are supported:
@@ -275,8 +312,9 @@ print(score)  # Simple number to stdout
 Edit `evolution/config.yaml` to customize:
 
 ```yaml
-# Working directory for evolution files
-evolution_dir: "evolution"
+# NOTE: The evolution directory is automatically inferred from this config file's location.
+# For example, if this file is at /path/to/my-experiment/config.yaml,
+# then the evolution directory will be /path/to/my-experiment/
 
 # Algorithm and evaluator file paths
 algorithm_file: "algorithm.py"

package/bin/claude-evolve-analyze

@@ -77,7 +77,30 @@ if [[ ! -f $csv_file ]]; then
   exit 1
 fi
 
-echo "=== Evolution Analysis Summary ==="
+# Determine what we're evolving based on paths
+EVOLUTION_CONTEXT=""
+if [[ -n "$EVOLUTION_DIR" ]]; then
+  # Get the evolution directory name (e.g., "evolution-atr" -> "ATR")
+  EVOLUTION_NAME=$(basename "$EVOLUTION_DIR")
+  EVOLUTION_CONTEXT="${EVOLUTION_NAME#evolution-}"
+  EVOLUTION_CONTEXT=$(echo "$EVOLUTION_CONTEXT" | tr '[:lower:]' '[:upper:]')
+fi
+
+# If we can't determine from evolution dir, try from algorithm path
+if [[ -z "$EVOLUTION_CONTEXT" && -n "$ALGORITHM_FILE" ]]; then
+  # Get algorithm file name
+  if [[ -f "$FULL_ALGORITHM_PATH" ]]; then
+    ALGO_NAME=$(basename "$FULL_ALGORITHM_PATH" .py)
+    EVOLUTION_CONTEXT="$ALGO_NAME"
+  fi
+fi
+
+# Default if we still can't determine
+if [[ -z "$EVOLUTION_CONTEXT" ]]; then
+  EVOLUTION_CONTEXT="Algorithm"
+fi
+
+echo "=== Evolution Analysis Summary - $EVOLUTION_CONTEXT ==="
 echo
 
 # Count totals (pure shell)
@@ -402,6 +425,7 @@ with open('$csv_file', 'r') as f:
     
     max_perf = 0
     max_id = ''
+    max_desc = ''
     max_order = 0
     completed_order = 0
     
@@ -415,12 +439,16 @@ with open('$csv_file', 'r') as f:
                         max_perf = perf_val
                         max_order = completed_order
                         max_id = row[0]
+                        max_desc = row[2] if len(row) > 2 else ''
             except ValueError:
                 pass
 
 print(f'max_perf={max_perf}')
 print(f'max_row={max_order}')
 print(f'max_id=\"{max_id}\"')
+# Escape special characters in description for shell
+desc_escaped = max_desc.replace('\\\\', '\\\\\\\\').replace('\"', '\\\\\"').replace('\$', '\\\\\$').replace('\`', '\\\\\`')
+print(f'max_desc=\"{desc_escaped}\"')
 ")"
 
   # Create generation averages file and track max generation
@@ -544,7 +572,7 @@ set multiplot layout 2,1 margins 0.08,0.82,0.15,0.95 spacing 0.1,0.15
 
 #=================== TOP PLOT: Performance Over Time ===================
 # AIDEV-NOTE: Removed x-axis to eliminate tick overlap and formatting issues
-set title "Algorithm Evolution Performance Over Time" font ",14"
+set title "$EVOLUTION_CONTEXT Algorithm Evolution Performance Over Time" font ",14"
 unset xlabel
 set ylabel "Performance Score"
 set grid y  # Only show horizontal grid lines
@@ -578,10 +606,11 @@ plot "$gen_avg_file" using 1:3 with boxes linecolor rgb "#4CAF50" notitle
 
 unset multiplot
 
-# Add winner label at bottom
-set terminal png size 1200,830
+# Add winner label and description at bottom
+set terminal png size 1200,850
 set output "$output_file"
-set label "Best Overall: $max_id (Score: $max_perf)" at screen 0.5, 0.05 center font ",12"
+set label "Best Overall: $max_id (Score: $max_perf)" at screen 0.5, 0.07 center font ",12"
+set label "$max_desc" at screen 0.5, 0.04 center font ",10" textcolor rgb "#666666"
 replot
 EOF
   else

package/bin/claude-evolve-cleanup

@@ -0,0 +1,297 @@
+#!/bin/bash
+
+set -e
+
+# Load configuration
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=../lib/config.sh
+source "$SCRIPT_DIR/../lib/config.sh"
+
+# Use CLAUDE_EVOLVE_CONFIG if set, otherwise default
+if [[ -n ${CLAUDE_EVOLVE_CONFIG:-} ]]; then
+  load_config "$CLAUDE_EVOLVE_CONFIG"
+else
+  load_config
+fi
+
+# Function to show help
+show_help() {
+  cat <<EOF
+claude-evolve cleanup - Clean up unchanged algorithms and their descendants
+
+USAGE:
+  claude-evolve cleanup [OPTIONS]
+
+OPTIONS:
+  --dry-run    Show what would be done without making changes
+  --force      Actually perform the cleanup (required for real changes)
+  --help       Show this help message
+
+DESCRIPTION:
+  This tool finds algorithm files that are identical to their parent and:
+  1. Deletes the unchanged .py files
+  2. Resets those candidates to pending status in CSV
+  3. Finds and cleans up any descendants that inherited from the bad copies
+  
+  Use --dry-run first to see what would be affected.
+
+EXAMPLES:
+  claude-evolve cleanup --dry-run   # Preview changes
+  claude-evolve cleanup --force     # Actually clean up
+EOF
+}
+
+# Parse arguments
+DRY_RUN=true
+FORCE=false
+
+while [[ $# -gt 0 ]]; do
+  case $1 in
+  --dry-run)
+    DRY_RUN=true
+    shift
+    ;;
+  --force)
+    FORCE=true
+    DRY_RUN=false
+    shift
+    ;;
+  --help)
+    show_help
+    exit 0
+    ;;
+  *)
+    echo "[ERROR] Unknown option: $1" >&2
+    exit 1
+    ;;
+  esac
+done
+
+if [[ $FORCE == false ]]; then
+  DRY_RUN=true
+fi
+
+# Validate configuration
+if ! validate_config; then
+  echo "[ERROR] Configuration validation failed" >&2
+  exit 1
+fi
+
+# Check if CSV exists
+if [[ ! -f "$FULL_CSV_PATH" ]]; then
+  echo "[ERROR] Evolution CSV not found: $FULL_CSV_PATH" >&2
+  exit 1
+fi
+
+echo "🧹 Claude-Evolve Duplicate Cleanup Tool"
+echo "========================================"
+echo "Evolution directory: $FULL_EVOLUTION_DIR"
+echo "CSV file: $FULL_CSV_PATH"
+echo "Mode: $(if [[ $DRY_RUN == true ]]; then echo "DRY RUN (preview only)"; else echo "FORCE (will make changes)"; fi)"
+echo ""
+
+# Use Python to analyze and clean up duplicates
+"$PYTHON_CMD" -c "
+import csv
+import os
+import sys
+import shutil
+from pathlib import Path
+
+csv_file = '$FULL_CSV_PATH'
+evolution_dir = '$FULL_EVOLUTION_DIR'
+dry_run = '$DRY_RUN' == 'true'
+algorithm_file = '$FULL_ALGORITHM_PATH'
+
+def files_identical(file1, file2):
+    \"\"\"Check if two files have identical content.\"\"\"
+    if not os.path.exists(file1) or not os.path.exists(file2):
+        return False
+    
+    try:
+        with open(file1, 'rb') as f1, open(file2, 'rb') as f2:
+            return f1.read() == f2.read()
+    except Exception:
+        return False
+
+def get_algorithm_file_path(candidate_id, base_algorithm):
+    \"\"\"Get the file path for a candidate's algorithm.\"\"\"
+    # Handle both old and new format IDs
+    if candidate_id.isdigit():
+        filename = f'evolution_id{candidate_id}.py'
+    else:
+        filename = f'evolution_{candidate_id}.py'
+    
+    return os.path.join(evolution_dir, filename)
+
+def get_parent_file_path(based_on_id, base_algorithm):
+    \"\"\"Get the file path for a parent algorithm.\"\"\"
+    if not based_on_id or based_on_id == '0' or based_on_id == '\"\"':
+        return base_algorithm
+    
+    # Handle both old and new format IDs
+    if based_on_id.isdigit():
+        filename = f'evolution_id{based_on_id}.py'
+    else:
+        filename = f'evolution_{based_on_id}.py'
+    
+    return os.path.join(evolution_dir, filename)
+
+try:
+    # Read CSV
+    with open(csv_file, 'r') as f:
+        reader = csv.reader(f)
+        rows = list(reader)
+    
+    if len(rows) <= 1:
+        print('No candidates found in CSV')
+        sys.exit(0)
+    
+    header = rows[0]
+    candidates = {}
+    
+    # Build candidate map
+    for i, row in enumerate(rows[1:], 1):
+        if len(row) >= 3:
+            candidate_id = row[0]
+            based_on_id = row[1] if len(row) > 1 else ''
+            description = row[2] if len(row) > 2 else ''
+            performance = row[3] if len(row) > 3 else ''
+            status = row[4] if len(row) > 4 else ''
+            
+            candidates[candidate_id] = {
+                'row_index': i,
+                'based_on_id': based_on_id,
+                'description': description,
+                'performance': performance,
+                'status': status,
+                'file_path': get_algorithm_file_path(candidate_id, algorithm_file)
+            }
+    
+    print(f'Found {len(candidates)} candidates to analyze')
+    print('')
+    
+    # Find unchanged candidates
+    unchanged_candidates = []
+    
+    for candidate_id, info in candidates.items():
+        if not info['based_on_id'] or info['based_on_id'] == '0' or info['based_on_id'] == '\"\"':
+            # Skip root candidates (no parent)
+            continue
+            
+        parent_file = get_parent_file_path(info['based_on_id'], algorithm_file)
+        candidate_file = info['file_path']
+        
+        if os.path.exists(candidate_file) and files_identical(candidate_file, parent_file):
+            unchanged_candidates.append(candidate_id)
+            print(f'📋 UNCHANGED: {candidate_id} is identical to parent {info[\"based_on_id\"]}')
+            print(f'   File: {os.path.basename(candidate_file)}')
+            print(f'   Description: {info[\"description\"]}')
+            print(f'   Status: {info[\"status\"]}')
+            print('')
+    
+    if not unchanged_candidates:
+        print('✅ No unchanged candidates found - all algorithms appear to be properly mutated!')
+        sys.exit(0)
+    
+    print(f'Found {len(unchanged_candidates)} unchanged candidates')
+    print('')
+    
+    # Find descendants of unchanged candidates
+    def find_descendants(bad_parent_id, all_candidates, found=None):
+        if found is None:
+            found = set()
+        
+        for cand_id, info in all_candidates.items():
+            if info['based_on_id'] == bad_parent_id and cand_id not in found:
+                found.add(cand_id)
+                # Recursively find descendants of this candidate
+                find_descendants(cand_id, all_candidates, found)
+        
+        return found
+    
+    all_affected = set(unchanged_candidates)
+    
+    # Find all descendants
+    for unchanged_id in unchanged_candidates:
+        descendants = find_descendants(unchanged_id, candidates)
+        all_affected.update(descendants)
+        
+        if descendants:
+            print(f'🔗 DESCENDANTS of {unchanged_id}: {sorted(descendants)}')
+    
+    print('')
+    print(f'📊 SUMMARY:')
+    print(f'   • {len(unchanged_candidates)} unchanged candidates')
+    print(f'   • {len(all_affected) - len(unchanged_candidates)} descendants affected')
+    print(f'   • {len(all_affected)} total candidates to clean up')
+    print('')
+    
+    if dry_run:
+        print('🔍 DRY RUN - Showing what would be done:')
+        print('')
+        
+        for candidate_id in sorted(all_affected):
+            info = candidates[candidate_id]
+            action = 'DELETE FILE & RESET' if candidate_id in unchanged_candidates else 'RESET (descendant)'
+            print(f'   {action}: {candidate_id}')
+            print(f'     File: {os.path.basename(info[\"file_path\"])}')
+            print(f'     Description: {info[\"description\"]}')
+            print('')
+        
+        print('To actually perform cleanup, run with --force')
+    else:
+        print('🧹 PERFORMING CLEANUP:')
+        print('')
+        
+        # Delete files and update CSV
+        files_deleted = 0
+        rows_updated = 0
+        
+        for candidate_id in sorted(all_affected):
+            info = candidates[candidate_id]
+            
+            # Delete file if it exists (for unchanged candidates)
+            if candidate_id in unchanged_candidates and os.path.exists(info['file_path']):
+                try:
+                    os.remove(info['file_path'])
+                    files_deleted += 1
+                    print(f'   ✅ DELETED: {os.path.basename(info[\"file_path\"])}')
+                except Exception as e:
+                    print(f'   ❌ FAILED to delete {os.path.basename(info[\"file_path\"])}: {e}')
+            
+            # Reset CSV row (clear performance and status, keep description)
+            row_idx = info['row_index']
+            if len(rows[row_idx]) >= 5:
+                # Clear performance (column 3) and status (column 4), but keep first 3 columns
+                rows[row_idx] = rows[row_idx][:3] + ['', ''] + rows[row_idx][5:]
+                rows_updated += 1
+                print(f'   ✅ RESET CSV: {candidate_id} -> pending')
+        
+        # Write updated CSV
+        try:
+            with open(csv_file + '.tmp', 'w', newline='') as f:
+                writer = csv.writer(f)
+                writer.writerows(rows)
+            
+            # Atomic replace
+            os.rename(csv_file + '.tmp', csv_file)
+            print('')
+            print(f'✅ CLEANUP COMPLETE:')
+            print(f'   • {files_deleted} files deleted')
+            print(f'   • {rows_updated} CSV rows reset to pending')
+            print(f'   • CSV updated successfully')
+            
+        except Exception as e:
+            print(f'❌ FAILED to update CSV: {e}')
+            sys.exit(1)
+
+except Exception as e:
+    print(f'Error: {e}')
+    sys.exit(1)
+"
+
+echo ""
+if [[ $DRY_RUN == true ]]; then
+  echo "💡 TIP: Run with --force to actually perform the cleanup"
+fi