claude-evolve 1.3.38 → 1.3.40

package/bin/claude-evolve-status

@@ -0,0 +1,220 @@
+#!/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
+
+# Validate configuration
+if ! validate_config; then
+  echo "[ERROR] Configuration validation failed" >&2
+  exit 1
+fi
+
+# Function to show help
+show_help() {
+  cat <<EOF
+claude-evolve status - Show evolution progress and current leader
+
+USAGE:
+  claude-evolve status [OPTIONS]
+
+OPTIONS:
+  --brief     Show only summary stats (no per-generation breakdown)
+  --winner    Show only the current best performer
+  --help      Show this help message
+
+DESCRIPTION:
+  Displays evolution progress by generation including:
+  - Candidate counts by status (pending, complete, failed, running)
+  - Current best performer across all generations
+  - Generation-by-generation breakdown
+
+EXAMPLES:
+  claude-evolve status          # Full status report
+  claude-evolve status --brief  # Just totals and winner
+  claude-evolve status --winner # Just the best performer
+EOF
+}
+
+# Parse arguments
+SHOW_BRIEF=false
+SHOW_WINNER_ONLY=false
+
+while [[ $# -gt 0 ]]; do
+  case $1 in
+  --brief)
+    SHOW_BRIEF=true
+    shift
+    ;;
+  --winner)
+    SHOW_WINNER_ONLY=true
+    shift
+    ;;
+  --help)
+    show_help
+    exit 0
+    ;;
+  *)
+    echo "[ERROR] Unknown option: $1" >&2
+    exit 1
+    ;;
+  esac
+done
+
+# Check if CSV exists
+if [[ ! -f "$FULL_CSV_PATH" ]]; then
+  echo "[ERROR] Evolution CSV not found: $FULL_CSV_PATH" >&2
+  echo "Run 'claude-evolve setup' first or navigate to the correct directory" >&2
+  exit 1
+fi
+
+# Determine what we're evolving based on paths
+EVOLUTION_CONTEXT=""
+if [[ -n "$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 "$ALGORITHM_PATH" ]]; then
+  # Get parent directory name or algorithm file name
+  if [[ -f "$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
+
+# Main status reporting using Python
+"$PYTHON_CMD" -c "
+import csv
+import sys
+
+csv_file = '$FULL_CSV_PATH'
+show_brief = '$SHOW_BRIEF' == 'true'
+show_winner_only = '$SHOW_WINNER_ONLY' == 'true'
+evolution_context = '$EVOLUTION_CONTEXT'
+
+try:
+    with open(csv_file, 'r') as f:
+        reader = csv.reader(f)
+        rows = list(reader)
+    
+    if len(rows) <= 1:
+        print('No evolution candidates found')
+        sys.exit(0)
+    
+    header = rows[0]
+    
+    # Collect all candidates with scores and statuses
+    all_candidates = []
+    stats_by_gen = {}
+    total_stats = {'pending': 0, 'complete': 0, 'failed': 0, 'running': 0}
+    
+    for row in rows[1:]:
+        if len(row) >= 1 and row[0]:  # Must have an ID
+            candidate_id = row[0]
+            
+            # Extract generation (e.g., 'gen03' from 'gen03-001')
+            if '-' in candidate_id:
+                gen = candidate_id.split('-')[0]
+                
+                # Get status and performance
+                status = row[4] if len(row) > 4 and row[4] else 'pending'
+                performance = row[3] if len(row) > 3 and row[3] else ''
+                
+                # Track by generation
+                if gen not in stats_by_gen:
+                    stats_by_gen[gen] = {'pending': 0, 'complete': 0, 'failed': 0, 'running': 0}
+                
+                if status in stats_by_gen[gen]:
+                    stats_by_gen[gen][status] += 1
+                    total_stats[status] += 1
+                else:
+                    stats_by_gen[gen]['pending'] += 1
+                    total_stats['pending'] += 1
+                
+                # Collect for winner analysis (only completed with valid scores)
+                if status == 'complete' and performance:
+                    try:
+                        score = float(performance)
+                        description = row[2] if len(row) > 2 else 'No description'
+                        all_candidates.append((candidate_id, description, score))
+                    except ValueError:
+                        pass
+    
+    # Find the winner
+    winner = None
+    if all_candidates:
+        winner = max(all_candidates, key=lambda x: x[2])
+    
+    # Show winner only
+    if show_winner_only:
+        if winner:
+            print(f'🏆 CURRENT LEADER: {winner[0]} (score: {winner[2]:.4f})')
+            print(f'   {winner[1]}')
+        else:
+            print('No completed candidates found')
+        sys.exit(0)
+    
+    # Show header
+    print(f'🧬 Evolution Status Report - {evolution_context}')
+    print('=' * 50)
+    
+    # Show overall stats
+    total_candidates = sum(total_stats.values())
+    if total_candidates > 0:
+        print(f'📊 OVERALL: {total_candidates} total candidates')
+        print(f'   • {total_stats[\"pending\"]} pending')
+        print(f'   • {total_stats[\"complete\"]} complete') 
+        print(f'   • {total_stats[\"failed\"]} failed')
+        print(f'   • {total_stats[\"running\"]} running')
+        print()
+    
+    # Show current winner
+    if winner:
+        print(f'🏆 CURRENT LEADER: {winner[0]} (score: {winner[2]:.4f})')
+        print(f'   {winner[1]}')
+        print()
+    else:
+        print('🏆 CURRENT LEADER: None (no completed candidates)')
+        print()
+    
+    # Show per-generation breakdown (unless brief mode)
+    if not show_brief and stats_by_gen:
+        print('📈 BY GENERATION:')
+        for gen in sorted(stats_by_gen.keys()):
+            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
+            
+            status_str = f'{data[\"pending\"]}p {data[\"complete\"]}c {data[\"failed\"]}f {data[\"running\"]}r'
+            
+            if gen_best:
+                print(f'   {gen}: {total} total ({status_str}) - best: {gen_best[0]} ({gen_best[2]:.4f})')
+            else:
+                print(f'   {gen}: {total} total ({status_str}) - best: none')
+
+except Exception as e:
+    print(f'Error reading evolution status: {e}')
+    sys.exit(1)
+"

package/bin/claude-evolve-worker

@@ -4,6 +4,20 @@
 
 set -e
 
+# Track temp file for cleanup
+temp_file=""
+
+# Cleanup function for temp files
+cleanup_temp() {
+  if [[ -n "$temp_file" && -f "$temp_file" ]]; then
+    rm -f "$temp_file"
+    echo "[WORKER-$$] Cleaned up temp file: $temp_file" >&2
+  fi
+}
+
+# Set trap to clean up temp files on exit
+trap cleanup_temp EXIT INT TERM
+
 # Load configuration
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 # shellcheck source=../lib/config.sh
@@ -47,7 +61,7 @@ done
 if [[ -z $candidate_id ]]; then
   candidate_id=$(find_next_pending_with_lock)
   if [[ -z $candidate_id ]]; then
-    echo "[INFO] No pending candidates found"
+    echo "[DEBUG] No pending candidates found" >&2
     exit 0
   fi
 else
@@ -130,27 +144,35 @@ else
   fi
 fi
 
-# Generate output file
+# Generate output file path
 if [[ $id =~ ^[0-9]+$ ]]; then
   output_file="$FULL_OUTPUT_DIR/evolution_id${id}.py"
 else
   output_file="$FULL_OUTPUT_DIR/evolution_${id}.py"
 fi
 
+# Use temp file for mutations to avoid partial/failed edits
+temp_file="${output_file}.tmp$$"
+
 # Check if processing should be skipped using common logic
 eval "$("$PYTHON_CMD" "$SCRIPT_DIR/../lib/evolution_processor.py" "$id" "$based_on_id" "$FULL_OUTPUT_DIR" "$ROOT_DIR" "$parent_file" "$output_file")"
 
-# Handle copy operation
+# Handle copy operation to temp file
 if [[ "$skip_copy" == "True" ]]; then
   echo "[WORKER-$$] ⚠️  Skipping copy - $reason"
 else
-  cp "$parent_file" "$output_file"
-  echo "[WORKER-$$] Copied parent to: $output_file"
+  cp "$parent_file" "$temp_file"
+  echo "[WORKER-$$] Copied parent to temp file: $temp_file"
 fi
 
 # Handle Claude mutation based on skip flags
 if [[ "$skip_claude" == "True" ]]; then
   echo "[WORKER-$$] ⚠️  Skipping Claude processing - $reason"
+  # If we have a temp file but are skipping Claude, move it to final location
+  if [[ -f "$temp_file" ]]; then
+    mv "$temp_file" "$output_file"
+    echo "[WORKER-$$] Moved temp file to final location (no Claude processing)"
+  fi
 else
   # Check for claude CLI
   claude_cmd="${CLAUDE_CMD:-claude}"
@@ -164,7 +186,7 @@ else
   echo "[WORKER-$$] Using Claude $CLAUDE_MODEL for mutation"
   
   # Create mutation prompt
-  prompt="Edit the file $output_file to implement this specific change: $description
+  prompt="Edit the file $temp_file to implement this specific change: $description
 
 Requirements:
 - Edit the file directly (don't just provide comments or suggestions)
@@ -189,9 +211,14 @@ The file currently contains the parent algorithm. Modify it according to the des
   claude_output=$(echo "$prompt" | "$claude_cmd" --dangerously-skip-permissions --model $CLAUDE_MODEL -p 2>&1 | tee -a "$LOGFILE")
   claude_exit_code=${PIPESTATUS[1]}
   
-  # Check for rate limit
-  if echo "$claude_output" | grep -q "Claude AI usage limit reached"; then
+  # Check for rate limit (multiple possible messages)
+  if echo "$claude_output" | grep -q -E "(usage limit|rate limit|limit reached|too many requests)"; then
     echo "[ERROR] Claude API rate limit reached" >&2
+    # Clean up the temp file
+    if [[ -f "$temp_file" ]]; then
+      rm "$temp_file"
+      echo "[WORKER-$$] Cleaned up temp file due to rate limit" >&2
+    fi
     # Reset to pending so it can be retried later
     update_csv_row_with_lock "$candidate_id" "status" "pending"
     exit 2  # Special exit code for rate limit
@@ -199,9 +226,42 @@ The file currently contains the parent algorithm. Modify it according to the des
   
   if [[ $claude_exit_code -ne 0 ]]; then
     echo "[ERROR] Claude failed to mutate algorithm" >&2
+    # Clean up the temp file
+    if [[ -f "$temp_file" ]]; then
+      rm "$temp_file"
+      echo "[WORKER-$$] Cleaned up temp file due to Claude failure" >&2
+    fi
     update_csv_row_with_lock "$candidate_id" "status" "failed"
     exit 1
   fi
+  
+  # Verify that Claude actually modified the file
+  if [[ -f "$temp_file" && -f "$parent_file" ]]; then
+    if cmp -s "$temp_file" "$parent_file"; then
+      echo "" >&2
+      echo "🚨🚨🚨 RED ALERT: UNCHANGED ALGORITHM DETECTED 🚨🚨🚨" >&2
+      echo "ERROR: Temp file is IDENTICAL to parent algorithm!" >&2
+      echo "ERROR: Claude failed to make any changes" >&2
+      echo "ERROR: Marking as failed - no evaluation will run" >&2
+      echo "🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨" >&2
+      echo "" >&2
+      
+      # Clean up temp file and mark as failed
+      rm "$temp_file"
+      update_csv_row_with_lock "$candidate_id" "status" "failed"
+      exit 1
+    else
+      # Changes were made - move temp file to final location
+      mv "$temp_file" "$output_file"
+      echo "[WORKER-$$] Changes detected - moved to: $output_file"
+    fi
+  else
+    # If we can't compare, assume it's okay and move the file
+    if [[ -f "$temp_file" ]]; then
+      mv "$temp_file" "$output_file"
+      echo "[WORKER-$$] Moved temp file to: $output_file"
+    fi
+  fi
 fi
 
 # Run evaluator
@@ -244,7 +304,7 @@ if [[ $eval_exit_code -eq 0 ]]; then
   # 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
+    if [[ $(echo "$score == 0" | bc -l) == "1" ]]; then
       update_csv_row_with_lock "$candidate_id" "status" "failed"
       update_csv_row_with_lock "$candidate_id" "performance" "$score"
       echo "[WORKER-$$] ✗ Evaluation failed with score 0"
@@ -286,7 +346,7 @@ if [[ $eval_exit_code -eq 0 ]]; then
   # Fallback: Try simple grep for score/performance fields
   if score=$(echo "$eval_output" | grep -o '"score"[[:space:]]*:[[:space:]]*[0-9.]*' | cut -d: -f2 | tr -d ' '); then
     if [[ -n $score ]]; then
-      if (( $(echo "$score == 0" | bc -l) )); then
+      if [[ $(echo "$score == 0" | bc -l) == "1" ]]; then
         update_csv_row_with_lock "$candidate_id" "status" "failed"
         update_csv_row_with_lock "$candidate_id" "performance" "$score"
         echo "[WORKER-$$] ✗ Evaluation failed with score 0"
@@ -303,7 +363,7 @@ if [[ $eval_exit_code -eq 0 ]]; then
   # Try "performance" field
   if score=$(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
+      if [[ $(echo "$score == 0" | bc -l) == "1" ]]; 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"
@@ -319,6 +379,8 @@ if [[ $eval_exit_code -eq 0 ]]; then
   
   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
+  echo "[ERROR] Actual evaluator output was:" >&2
+  echo "$eval_output" >&2
   update_csv_row_with_lock "$candidate_id" "status" "failed"
   exit 1
 else

package/lib/config.sh

@@ -78,7 +78,7 @@ load_config() {
   
   # Load config if found
   if [[ -f "$config_file" ]]; then
-    echo "[INFO] Loading configuration from: $config_file"
+    echo "[DEBUG] Loading configuration from: $config_file" >&2
     # Simple YAML parsing for key: value pairs and nested structures
     local in_ideation_section=false
     local in_parallel_section=false
@@ -143,7 +143,6 @@ load_config() {
       else
         # Handle top-level keys
         case $key in
-          evolution_dir) EVOLUTION_DIR="$value" ;;
           algorithm_file) ALGORITHM_FILE="$value" ;;
           evaluator_file) EVALUATOR_FILE="$value" ;;
           brief_file) BRIEF_FILE="$value" ;;
@@ -152,6 +151,9 @@ load_config() {
           parent_selection) PARENT_SELECTION="$value" ;;
           python_cmd) PYTHON_CMD="$value" ;;
           auto_ideate) AUTO_IDEATE="$value" ;;
+          evolution_dir) 
+            echo "[WARN] evolution_dir in config is ignored - automatically inferred from config file location" >&2
+            ;;
         esac
       fi
     done < "$config_file"
@@ -163,7 +165,7 @@ load_config() {
     local config_dir=$(dirname "$config_file")
     if [[ "$config_dir" != "." && "$config_dir" != "" ]]; then
       EVOLUTION_DIR="$config_dir"
-      echo "[INFO] Using evolution directory from config path: $EVOLUTION_DIR"
+      echo "[DEBUG] Using evolution directory from config path: $EVOLUTION_DIR" >&2
     fi
   fi
   

package/lib/csv-lock.sh

@@ -110,7 +110,13 @@ release_csv_lock() {
 # Usage: read_csv_with_lock <variable_name>
 read_csv_with_lock() {
     local var_name="$1"
-    local csv_file="${EVOLUTION_DIR:-evolution}/evolution.csv"
+    
+    # Ensure we have the full CSV path set
+    if [[ -z "$FULL_CSV_PATH" ]]; then
+        echo "[ERROR] FULL_CSV_PATH not set in read_csv_with_lock" >&2
+        return 1
+    fi
+    local csv_file="$FULL_CSV_PATH"
     
     if ! acquire_csv_lock; then
         return 1
@@ -130,7 +136,12 @@ read_csv_with_lock() {
 # Write CSV with lock
 # Usage: echo "content" | write_csv_with_lock
 write_csv_with_lock() {
-    local csv_file="${EVOLUTION_DIR:-evolution}/evolution.csv"
+    # Ensure we have the full CSV path set
+    if [[ -z "$FULL_CSV_PATH" ]]; then
+        echo "[ERROR] FULL_CSV_PATH not set in write_csv_with_lock" >&2
+        return 1
+    fi
+    local csv_file="$FULL_CSV_PATH"
     local temp_file="${csv_file}.tmp.$$"
     
     if ! acquire_csv_lock; then
@@ -153,7 +164,13 @@ update_csv_row_with_lock() {
     local target_id="$1"
     local field="$2"
     local value="$3"
-    local csv_file="${FULL_CSV_PATH:-${EVOLUTION_DIR:-evolution}/evolution.csv}"
+    
+    # Ensure we have the full CSV path set
+    if [[ -z "$FULL_CSV_PATH" ]]; then
+        echo "[ERROR] FULL_CSV_PATH not set in update_csv_row_with_lock" >&2
+        return 1
+    fi
+    local csv_file="$FULL_CSV_PATH"
     
     if ! acquire_csv_lock; then
         return 1
@@ -202,7 +219,12 @@ with open('${csv_file}.tmp', 'w', newline='') as f:
 # Find next pending candidate with lock
 # Usage: next_pending=$(find_next_pending_with_lock)
 find_next_pending_with_lock() {
-    local csv_file="${FULL_CSV_PATH:-${EVOLUTION_DIR:-evolution}/evolution.csv}"
+    # Ensure we have the full CSV path set
+    if [[ -z "$FULL_CSV_PATH" ]]; then
+        echo "[ERROR] FULL_CSV_PATH not set in find_next_pending_with_lock" >&2
+        return 1
+    fi
+    local csv_file="$FULL_CSV_PATH"
     
     if ! acquire_csv_lock; then
         return 1

package/lib/csv_helper.py

@@ -91,7 +91,7 @@ def main():
             performance = data.get('performance') or data.get('score', 0)
             
             # Build fields to update
-            fields = {'performance': performance, 'status': 'complete' if performance > 0 else 'failed'}
+            fields = {'performance': performance, 'status': 'complete' if performance != 0 else 'failed'}
             
             # Add all other fields from the JSON
             for key, value in data.items():

package/package.json

@@ -1,13 +1,12 @@
 {
   "name": "claude-evolve",
-  "version": "1.3.38",
+  "version": "1.3.40",
   "bin": {
     "claude-evolve": "./bin/claude-evolve",
     "claude-evolve-main": "./bin/claude-evolve-main",
     "claude-evolve-setup": "./bin/claude-evolve-setup",
     "claude-evolve-ideate": "./bin/claude-evolve-ideate",
     "claude-evolve-run": "./bin/claude-evolve-run",
-    "claude-evolve-run-parallel": "./bin/claude-evolve-run-parallel",
     "claude-evolve-worker": "./bin/claude-evolve-worker",
     "claude-evolve-analyze": "./bin/claude-evolve-analyze",
     "claude-evolve-config": "./bin/claude-evolve-config"

package/templates/config.yaml

@@ -1,19 +1,20 @@
 # claude-evolve configuration file
 # This file defines paths and settings for the evolution process
+# 
+# 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/
 
-# Working directory for evolution files
-evolution_dir: "evolution"
-
-# Algorithm and evaluator file paths (relative to evolution_dir)
+# Algorithm and evaluator file paths (relative to evolution directory)
 algorithm_file: "algorithm.py"
 evaluator_file: "evaluator.py"
 brief_file: "BRIEF.md"
 
-# CSV file for tracking evolution (relative to evolution_dir)
+# CSV file for tracking evolution (relative to evolution directory)
 evolution_csv: "evolution.csv"
 
-# Output directory for generated algorithms (relative to evolution_dir)
-# Leave empty to use evolution_dir directly
+# Output directory for generated algorithms (relative to evolution directory)
+# Leave empty to use evolution directory directly
 output_dir: ""
 
 # Parent algorithm selection strategy