claude-evolve 1.11.12 → 1.11.13

package/bin/claude-evolve-check

@@ -2,8 +2,8 @@
 # claude-evolve-check - Health check for AI model configurations
 # Tests all configured AI models to verify they're working before starting evolution runs
 #
-# AIDEV-NOTE: Claude CLI hangs when called from subshells (command substitution).
-# We use temp files instead of $() to avoid this issue.
+# AIDEV-NOTE: Claude CLI has issues with timeout command and subshells.
+# We test each model by running it directly and checking the result file.
 
 set -e
 
@@ -22,15 +22,16 @@ LIB_DIR="$(dirname "$SCRIPT_DIR")/lib"
 source "$LIB_DIR/config.sh"
 load_config
 
-# Test timeout in seconds (short to fail fast)
+# Test timeout in seconds
 TEST_TIMEOUT=30
 
 # Simple test prompt
 TEST_PROMPT="Say hello in exactly 3 words."
 
-# Temp file for output (avoids subshell issues with claude CLI)
+# Temp files
 TEMP_OUTPUT=$(mktemp)
-trap "rm -f $TEMP_OUTPUT" EXIT
+TEMP_PID=$(mktemp)
+trap "rm -f $TEMP_OUTPUT $TEMP_PID" EXIT
 
 echo -e "${CYAN}🔍 Claude Evolve - AI Model Health Check${NC}"
 echo "============================================"
@@ -39,12 +40,11 @@ echo
 # Collect all unique models into a simple list
 ALL_MODELS=""
 for model in $LLM_RUN $LLM_RUN_FALLBACK $LLM_IDEATE $LLM_IDEATE_FALLBACK; do
-  # Check if model already in list
   if ! echo "$ALL_MODELS" | grep -q -w "$model"; then
     ALL_MODELS="$ALL_MODELS $model"
   fi
 done
-ALL_MODELS=$(echo "$ALL_MODELS" | xargs)  # Trim whitespace
+ALL_MODELS=$(echo "$ALL_MODELS" | xargs)
 
 echo "Models to test: $ALL_MODELS"
 echo
@@ -55,78 +55,147 @@ FAILED=0
 FAILED_MODELS=""
 TIMEOUT_MODELS=""
 
-# Test each model - inline commands to avoid subshell issues
-for model in $ALL_MODELS; do
-  echo -n "Testing $model... "
-
-  START_TIME=$(date +%s)
+# Test a single model by writing a tiny test script and running it
+test_model() {
+  local model="$1"
+  local prompt="$2"
+  local outfile="$3"
+  local max_wait="$4"
 
-  # Run directly, write to temp file to avoid subshell
-  set +e
+  # Write a self-contained test script
+  local test_script=$(mktemp)
   case "$model" in
     opus|sonnet|haiku)
-      timeout -k 5 $TEST_TIMEOUT claude --dangerously-skip-permissions --mcp-config '' --model "$model" -p "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << SCRIPT
+#!/usr/bin/env bash
+exec claude --dangerously-skip-permissions --mcp-config '' --model $model -p "\$1"
+SCRIPT
       ;;
-    opus-think|sonnet-think)
-      base_model="${model%-think}"
-      timeout -k 5 $TEST_TIMEOUT claude --dangerously-skip-permissions --mcp-config '' --model "$base_model" -p "ultrathink
+    opus-think)
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec claude --dangerously-skip-permissions --mcp-config '' --model opus -p "ultrathink
 
-$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+$1"
+SCRIPT
+      ;;
+    sonnet-think)
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec claude --dangerously-skip-permissions --mcp-config '' --model sonnet -p "ultrathink
+
+$1"
+SCRIPT
       ;;
     gemini-pro)
-      timeout -k 5 $TEST_TIMEOUT gemini -y -m gemini-3-pro-preview -p "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec gemini -y -m gemini-3-pro-preview -p "$1"
+SCRIPT
       ;;
     gemini-flash|gemini-3-flash)
-      timeout -k 5 $TEST_TIMEOUT gemini -y -m gemini-2.5-flash -p "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec gemini -y -m gemini-2.5-flash -p "$1"
+SCRIPT
       ;;
     kimi-coder)
-      timeout -k 5 $TEST_TIMEOUT kimi --print -y -m kimi-for-coding -c "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec kimi --print -y -m kimi-for-coding -c "$1"
+SCRIPT
       ;;
     kimi-k2.5)
-      timeout -k 5 $TEST_TIMEOUT opencode -m openrouter/moonshotai/kimi-k2.5 run "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec opencode -m openrouter/moonshotai/kimi-k2.5 run "$1"
+SCRIPT
       ;;
     glm-5)
-      timeout -k 5 $TEST_TIMEOUT opencode -m openrouter/z-ai/glm-5 run "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec opencode -m openrouter/z-ai/glm-5 run "$1"
+SCRIPT
       ;;
     glm-5-zai)
-      timeout -k 5 $TEST_TIMEOUT opencode -m zai-coding-plan/glm-5 run "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec opencode -m zai-coding-plan/glm-5 run "$1"
+SCRIPT
       ;;
     glm-zai)
-      timeout -k 5 $TEST_TIMEOUT opencode -m zai-coding-plan/glm-4.7 run "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec opencode -m zai-coding-plan/glm-4.7 run "$1"
+SCRIPT
       ;;
     codex-oss-local)
-      timeout -k 5 $TEST_TIMEOUT codex exec --dangerously-bypass-approvals-and-sandbox --skip-git-repo-check --oss --local-provider=ollama "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec codex exec --dangerously-bypass-approvals-and-sandbox --skip-git-repo-check --oss --local-provider=ollama "$1"
+SCRIPT
       ;;
     gpt5|gpt5high)
-      timeout -k 5 $TEST_TIMEOUT codex exec --dangerously-bypass-approvals-and-sandbox "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec codex exec --dangerously-bypass-approvals-and-sandbox "$1"
+SCRIPT
       ;;
     qwen)
-      timeout -k 5 $TEST_TIMEOUT opencode -m openrouter/qwen/qwen3.5-plus-02-15 run "$TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
-      EXIT_CODE=$?
+      cat > "$test_script" << 'SCRIPT'
+#!/usr/bin/env bash
+exec opencode -m openrouter/qwen/qwen3.5-plus-02-15 run "$1"
+SCRIPT
       ;;
     *)
-      echo "Unknown model: $model" > "$TEMP_OUTPUT"
-      EXIT_CODE=1
+      echo "Unknown model: $model" > "$outfile"
+      rm -f "$test_script"
+      return 1
       ;;
   esac
+  chmod +x "$test_script"
+
+  # Run the test script in background, capture output to file
+  "$test_script" "$prompt" > "$outfile" 2>&1 &
+  local pid=$!
+
+  # Poll for completion
+  local elapsed=0
+  while kill -0 "$pid" 2>/dev/null; do
+    if [[ $elapsed -ge $max_wait ]]; then
+      kill "$pid" 2>/dev/null
+      sleep 1
+      kill -9 "$pid" 2>/dev/null
+      wait "$pid" 2>/dev/null
+      rm -f "$test_script"
+      return 124
+    fi
+    sleep 1
+    ((elapsed++))
+  done
+
+  wait "$pid"
+  local rc=$?
+  rm -f "$test_script"
+  return $rc
+}
+
+# Test each model
+for model in $ALL_MODELS; do
+  echo -n "Testing $model... "
+
+  START_TIME=$(date +%s)
+
+  set +e
+  test_model "$model" "$TEST_PROMPT" "$TEMP_OUTPUT" "$TEST_TIMEOUT"
+  EXIT_CODE=$?
   set -e
 
   END_TIME=$(date +%s)
   DURATION=$((END_TIME - START_TIME))
 
-  # Read output from temp file
-  OUTPUT=$(cat "$TEMP_OUTPUT")
+  # Read output
+  OUTPUT=$(cat "$TEMP_OUTPUT" 2>/dev/null || true)
   OUTPUT_LEN=${#OUTPUT}
 
   # Check result
@@ -134,7 +203,7 @@ $TEST_PROMPT" > "$TEMP_OUTPUT" 2>&1
     echo -e "${GREEN}✓ OK${NC} (${DURATION}s, ${OUTPUT_LEN} chars)"
     ((PASSED++))
   elif [[ $EXIT_CODE -eq 124 ]]; then
-    echo -e "${RED}✗ TIMEOUT${NC} (${TEST_TIMEOUT}s - likely permission dialog or hanging)"
+    echo -e "${RED}✗ TIMEOUT${NC} (${TEST_TIMEOUT}s)"
     TIMEOUT_MODELS="$TIMEOUT_MODELS $model"
     ((FAILED++))
   else

package/lib/config.sh

@@ -61,18 +61,19 @@ DEFAULT_WORKER_MAX_CANDIDATES=3
 # Primary: Strong models used in normal operation
 # Fallback: Cheap/backup models used only when primary tier exhausted
 #
-# Run: GLM-5 and Qwen are primary models for code generation
+# Run: Primary models for code generation
+# Sonnet 4.6: 79.6% SWE-bench, $3/$15 per M tokens
 # GLM-5: 744B MoE, $0.80/M tokens, 77.8% SWE-bench
 # Qwen: hybrid linear attention + sparse MoE, strong reasoning
-DEFAULT_LLM_RUN="glm-5 glm-5 qwen kimi-k2.5 kimi-k2.5"
-DEFAULT_LLM_RUN_FALLBACK="gemini-3-flash codex-oss-local haiku"
+DEFAULT_LLM_RUN="sonnet glm-5 glm-5 qwen kimi-k2.5"
+DEFAULT_LLM_RUN_FALLBACK="gemini-3-flash codex-oss-local"
 #
 # Ideate: Only agentic models that can edit files reliably
 # AIDEV-NOTE: Ideation REQUIRES file editing - non-agentic models (opencode run, codex) return text
 # but don't actually edit files. Only use claude/gemini CLI, cursor-agent, or zai-coding-plan models.
 # OpenRouter models (via opencode) are chat-only and CANNOT edit files for ideation.
 DEFAULT_LLM_IDEATE="opus-think sonnet-think glm-5-zai gemini-pro kimi-coder"
-DEFAULT_LLM_IDEATE_FALLBACK="sonnet glm-zai haiku"
+DEFAULT_LLM_IDEATE_FALLBACK="sonnet glm-5-zai"
 
 # Load configuration from a YAML file and update variables
 _load_yaml_config() {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-evolve",
-  "version": "1.11.12",
+  "version": "1.11.13",
   "bin": {
     "claude-evolve": "bin/claude-evolve",
     "claude-evolve-main": "bin/claude-evolve-main",