aidevops 3.13.58 → 3.13.59

package/VERSION

@@ -1 +1 @@
-3.13.58
+3.13.59

package/aidevops.sh

@@ -5,7 +5,7 @@
 # AI DevOps Framework CLI
 # Usage: aidevops <command> [options]
 #
-# Version: 3.13.58
+# Version: 3.13.59
 
 set -euo pipefail
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "aidevops",
-  "version": "3.13.58",
+  "version": "3.13.59",
   "description": "AI DevOps Framework - AI-assisted development workflows, code quality, and deployment automation",
   "type": "module",
   "bin": {

package/setup-modules/agent-deploy.sh

@@ -57,8 +57,13 @@ _restart_pulse_if_running() {
 		fi
 	fi
 
-	# No auto-restart — start it manually
-	local pulse_script="${HOME}/.aidevops/agents/scripts/pulse-wrapper.sh"
+	# No auto-restart — start it manually. Prefer the canonical repo source so a
+	# background launch never depends on a path inside ~/.aidevops/agents/scripts/
+	# that may be in the middle of a future deploy swap (must-not-be-wiped-during-deploy).
+	local pulse_script="${INSTALL_DIR:-.}/.agents/scripts/pulse-wrapper.sh"
+	if [[ ! -x "$pulse_script" ]]; then
+		pulse_script="${HOME}/.aidevops/agents/scripts/pulse-wrapper.sh"
+	fi
 	if [[ -x "$pulse_script" ]]; then
 		nohup "$pulse_script" >>"${HOME}/.aidevops/logs/pulse-wrapper.log" 2>&1 &
 		print_success "Pulse started manually (PID $!)"
@@ -307,6 +312,76 @@ _set_script_permissions_and_report() {
 	return 0
 }
 
+# _count_deployed_agent_files target_dir
+# Prints the number of files in the deployed agents tree. Non-numeric output is
+# normalised to 0 so deploy verification never compares an empty string.
+_count_deployed_agent_files() {
+	local target_dir="$1"
+	local file_count="0"
+	if [[ -d "$target_dir" ]]; then
+		file_count=$(find "$target_dir" -type f 2>/dev/null | wc -l | tr -d '[:space:]')
+	fi
+	[[ "$file_count" =~ ^[0-9]+$ ]] || file_count=0
+	printf '%s\n' "$file_count"
+	return 0
+}
+
+# _restore_latest_agents_backup target_dir
+# Restores the newest agents backup after a failed deploy verification. Backups
+# are created by create_backup_with_rotation as ~/.aidevops/agents-backups/<ts>/agents.
+_restore_latest_agents_backup() {
+	local target_dir="$1"
+	local backup_base="$HOME/.aidevops/agents-backups"
+	local latest_backup=""
+
+	if [[ ! -d "$backup_base" ]]; then
+		print_warning "No agents backup directory found for restore: $backup_base"
+		return 1
+	fi
+
+	latest_backup=$(find "$backup_base" -maxdepth 2 -type d -name agents 2>/dev/null | sort | tail -n 1)
+	if [[ -z "$latest_backup" || ! -d "$latest_backup" ]]; then
+		print_warning "No restorable agents backup found under $backup_base"
+		return 1
+	fi
+
+	print_warning "Restoring agents from latest backup: $latest_backup"
+	rm -rf "$target_dir"
+	mkdir -p "$(dirname "$target_dir")"
+	if cp -a "$latest_backup" "$target_dir"; then
+		print_success "Restored agents directory from backup"
+		return 0
+	fi
+
+	print_error "Failed to restore agents directory from backup: $latest_backup"
+	return 1
+}
+
+# _verify_deployed_agents_tree target_dir
+# Verifies the deployed agents tree is plausibly complete before .deployed-sha is
+# written. This catches empty/partial deploys that would otherwise suppress the
+# next auto-update retry by stamping the new SHA.
+_verify_deployed_agents_tree() {
+	local target_dir="$1"
+	local min_files="${AIDEVOPS_AGENT_DEPLOY_MIN_FILES:-100}"
+	local file_count="0"
+
+	[[ "$min_files" =~ ^[0-9]+$ ]] || min_files=100
+
+	if [[ ! -d "$target_dir/scripts" ]]; then
+		print_error "Deploy verification failed: $target_dir/scripts missing after swap"
+		return 1
+	fi
+
+	file_count=$(_count_deployed_agent_files "$target_dir")
+	if [[ "$file_count" -lt "$min_files" ]]; then
+		print_error "Deploy verification failed: $target_dir has $file_count files (< $min_files)"
+		return 1
+	fi
+
+	return 0
+}
+
 # _install_opencode_plugin_deps target_dir
 # Installs node_modules for the opencode-aidevops plugin.
 # GH#17829: @bufbuild/protobuf was missing; GH#17891: only symlink on first run.
@@ -651,12 +726,14 @@ deploy_aidevops_agents() {
 	fi
 
 	# Postcondition: verify the swap actually produced a functional agents dir.
-	# _atomic_stage_and_deploy_agents returns 0 on success, but a belt-and-
-	# suspenders check here catches any future regression where the function
-	# might return early without correctly populating $target_dir (GH#22014).
-	if [[ ! -d "$target_dir/scripts" ]]; then
-		print_error "Deploy verification failed: $target_dir/scripts missing after swap"
+	# _atomic_stage_and_deploy_agents returns 0 on success, but this belt-and-
+	# suspenders check catches future regressions where the function returns early
+	# without correctly populating $target_dir (GH#22014/GH#21973). Do not write
+	# .deployed-sha unless this passes; otherwise auto-update would suppress the
+	# next retry even though agents/ is empty or partial.
+	if ! _verify_deployed_agents_tree "$target_dir"; then
 		print_error "The agents directory was not correctly deployed — setup cannot continue"
+		_restore_latest_agents_backup "$target_dir" || true
 		return 1
 	fi
 

package/setup.sh

@@ -12,7 +12,7 @@ shopt -s inherit_errexit 2>/dev/null || true
 # AI Assistant Server Access Framework Setup Script
 # Helps developers set up the framework for their infrastructure
 #
-# Version: 3.13.58
+# Version: 3.13.59
 #
 # Quick Install:
 #   npm install -g aidevops && aidevops update          (recommended)
@@ -1144,6 +1144,30 @@ _setup_lock_pid_alive() {
 	return $?
 }
 
+_setup_lock_dir_age_seconds() {
+	local lock_dir="$1"
+	local lock_mtime=""
+	local now=""
+	if ! [[ -d "$lock_dir" ]]; then
+		printf '%s\n' 0
+		return 0
+	fi
+	if type _file_mtime_epoch >/dev/null 2>&1; then
+		lock_mtime=$(_file_mtime_epoch "$lock_dir" 2>/dev/null || true)
+	elif [[ "$(uname -s 2>/dev/null || true)" == "Darwin" || "$(uname -s 2>/dev/null || true)" == "FreeBSD" ]]; then
+		lock_mtime=$(stat -f %m "$lock_dir" 2>/dev/null || true)
+	else
+		lock_mtime=$(stat -c %Y "$lock_dir" 2>/dev/null || true)
+	fi
+	now=$(date +%s 2>/dev/null || true)
+	if [[ "$lock_mtime" =~ ^[0-9]+$ && "$now" =~ ^[0-9]+$ && "$now" -ge "$lock_mtime" ]]; then
+		printf '%s\n' $((now - lock_mtime))
+		return 0
+	fi
+	printf '%s\n' 0
+	return 0
+}
+
 _setup_register_child_pid() {
 	local pid="$1"
 	[[ -n "$pid" ]] || return 0
@@ -1216,6 +1240,18 @@ _setup_acquire_noninteractive_setup_lock() {
 		if [[ -r "$lock_dir/owner.pid" ]]; then
 			owner_pid=$(tr -d '[:space:]' <"$lock_dir/owner.pid" 2>/dev/null || true)
 		fi
+		if [[ -z "$owner_pid" ]]; then
+			local _lock_age="0"
+			_lock_age=$(_setup_lock_dir_age_seconds "$lock_dir")
+			if [[ "$_lock_age" -le 300 ]]; then
+				print_error "Another setup.sh --non-interactive process is acquiring the deploy lock (lock: ${lock_dir}, age ${_lock_age}s). Exiting to avoid overlapping deployments."
+				return 75
+			fi
+			print_warning "Removing stale setup.sh --non-interactive lock with no owner at ${lock_dir} (age ${_lock_age}s)"
+			rm -rf "$lock_dir" 2>/dev/null || true
+			attempts=$((attempts + 1))
+			continue
+		fi
 		if _setup_lock_pid_alive "$owner_pid"; then
 			[[ -r "$lock_dir/command" ]] && owner_cmd=$(tr '\n' ' ' <"$lock_dir/command" 2>/dev/null || true)
 			# Build actionable diagnostics: elapsed time since lock was acquired