codeharness 0.18.1 → 0.19.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeharness",
-  "version": "0.18.1",
+  "version": "0.19.2",
   "type": "module",
   "description": "CLI for codeharness — makes autonomous coding agents produce software that actually works",
   "bin": {

package/ralph/ralph.sh

@@ -560,6 +560,12 @@ execute_iteration() {
         loop_start_sha=$(git rev-parse HEAD 2>/dev/null || echo "")
     fi
 
+    # Snapshot sprint-state.json before iteration (for timeout delta capture)
+    local state_snapshot_path="ralph/.state-snapshot.json"
+    if [[ -f "sprint-state.json" ]]; then
+        cp "sprint-state.json" "$state_snapshot_path" 2>/dev/null || true
+    fi
+
     log_status "LOOP" "Iteration $iteration — Task: ${task_id:-'(reading from prompt)'}"
     local timeout_seconds=$((ITERATION_TIMEOUT_MINUTES * 60))
 
@@ -686,12 +692,36 @@ execute_iteration() {
         return 0
     elif [[ $exit_code -eq 124 ]]; then
         log_status "WARN" "Iteration timed out after ${ITERATION_TIMEOUT_MINUTES}m"
+
+        # Capture timeout report
+        if command -v npx &>/dev/null; then
+            log_status "INFO" "Capturing timeout report..."
+            npx codeharness timeout-report \
+                --story "${task_id:-unknown}" \
+                --iteration "$iteration" \
+                --duration "$ITERATION_TIMEOUT_MINUTES" \
+                --output-file "$output_file" \
+                --state-snapshot "$state_snapshot_path" 2>/dev/null && \
+                log_status "INFO" "Timeout report saved" || \
+                log_status "WARN" "Failed to capture timeout report"
+        fi
+
+        # Verify report file exists with non-zero content
+        local report_file="ralph/logs/timeout-report-${iteration}-${task_id:-unknown}.md"
+        if [[ -s "$report_file" ]]; then
+            log_status "INFO" "Timeout report verified: $report_file"
+        fi
+
         return 1
     else
         # Check for API limit
         if grep -qi "5.*hour.*limit\|limit.*reached.*try.*back\|usage.*limit.*reached" "$output_file" 2>/dev/null; then
             log_status "ERROR" "Claude API usage limit reached"
             return 2
+        # Check for transient API errors (500, 529, overloaded) — don't count against story
+        elif grep -qi "Internal server error\|api_error\|overloaded\|529\|503" "$output_file" 2>/dev/null; then
+            log_status "WARN" "Transient API error (not story's fault) — will retry"
+            return 4
         else
             log_status "ERROR" "$(driver_display_name) execution failed (exit code: $exit_code)"
             return 1
@@ -997,6 +1027,12 @@ main() {
                 update_status "$loop_count" "$(cat "$CALL_COUNT_FILE" 2>/dev/null || echo "0")" "circuit_breaker" "halted"
                 break
                 ;;
+            4)
+                # Transient API error — retry after brief pause, don't count against story
+                consecutive_failures=0  # reset — this isn't the story's fault
+                log_status "INFO" "Transient API error — retrying in 30s (not counting against story)"
+                sleep 30
+                ;;
             *)
                 # Failure (timeout or crash) — increment retry for the story that was being worked on
                 consecutive_failures=$((consecutive_failures + 1))