On mixed dev projects - this will murder your mdworker_shared x worker, and make you pay a hidden I/O tax that can result in EBUSY. This will correct that.

macOS_fix_spotlight.sh

#!/usr/bin/env bash
# macOS_fix_spotlight.sh — drop .metadata_never_index in build/cache dirs
# so mds_worker stops melting your CPU walking node_modules etc.
#
# GUI alternative: System Settings → Spotlight → Search Privacy (bottom of
# the pane) lets you add directories Spotlight should ignore entirely. This
# script is the recursive, repo-aware version — it marks only the noisy
# build/cache subdirs and leaves source trees searchable.
#
# Other OSes have their own Spotlight-equivalents (not addressed by this
# script, listed here for reference):
#   Windows — Windows Search (Win/Win+S), PowerToys Run (Alt+Space),
#             Everything (voidtools) for instant filename search.
#   Linux   — GNOME Activities (Super key, Tracker-backed), KDE KRunner
#             (Alt+Space), or cross-desktop Ulauncher / Albert.
#
# Antivirus / EDR agents are the other usual suspect for CPU spikes during
# heavy file activity (git checkouts, npm/pnpm installs, terraform init,
# large builds). They scan every file the indexer or your tools touch, so
# the Spotlight fix only solves half the problem on managed laptops. If
# `mds_worker` is quiet but your fans are still loud, check what's running:
#   macOS    — CrowdStrike Falcon (`falconctl`), SentinelOne (`sentinelctl`),
#              Sophos, Defender for Endpoint, Jamf Protect, Carbon Black.
#   Windows  — Defender (real-time protection), CrowdStrike, SentinelOne,
#              Sophos, Carbon Black, Trellix/McAfee.
#   Linux    — ClamAV (`clamd`), CrowdStrike, SentinelOne, Sophos.
# Mitigation is org-policy dependent: ask IT/security to add path exclusions
# for known-noisy dirs (node_modules, .terraform, build/) — don't disable the
# agent. On a personal machine, the exclusion lives in the AV's own UI.
#
# Usage: macOS_fix_spotlight.sh [--dry-run] [--verbose] [--force] [ROOT]

set -euo pipefail

# --- args --------------------------------------------------------------------
dry_run=0
verbose=0
force=0
root=""

usage() {
  cat >&2 <<'EOF'
macOS_fix_spotlight.sh — drop .metadata_never_index in build/cache dirs
so mds_worker stops melting your CPU walking node_modules etc.

Usage: macOS_fix_spotlight.sh [--dry-run] [--verbose] [--force] [ROOT]
  ROOT defaults to $PWD. Refuses to scan / or $HOME without --force.
  -n, --dry-run    show what would be marked, don't touch anything
  -v, --verbose    print 'exists' and 'skip' lines too
      --force      allow scanning / or $HOME
  -h, --help       this message
EOF
  exit "${1:-0}"
}

while [ $# -gt 0 ]; do
  case "$1" in
    -n|--dry-run) dry_run=1 ;;
    -v|--verbose) verbose=1 ;;
    --force)      force=1 ;;
    -h|--help)    usage 0 ;;
    --)           shift; break ;;
    -*)           printf 'unknown flag: %s\n' "$1" >&2; usage 1 ;;
    *)
      if [ -n "$root" ]; then
        printf 'too many positional args\n' >&2
        usage 1
      fi
      root="$1"
      ;;
  esac
  shift
done
[ $# -gt 0 ] && [ -z "$root" ] && root="$1"

root="${root:-$PWD}"
[ -d "$root" ] || { printf 'not a directory: %s\n' "$root" >&2; exit 1; }
root="$(cd "$root" && pwd -P)"

case "$root" in
  /|"${HOME:-/__nope__}")
    if [ "$force" -ne 1 ]; then
      printf 'refusing to scan %s without --force\n' "$root" >&2
      exit 2
    fi
    ;;
esac

# --- categories --------------------------------------------------------------
# Always-safe: directory name uniquely implies a build/cache artifact.
ALWAYS=(
  node_modules .next .nuxt .svelte-kit .turbo .parcel-cache .pnpm-store
  __pycache__ .mypy_cache .pytest_cache .ruff_cache .tox .nox
  .gradle DerivedData Pods .nyc_output
  .terraform .serverless .wrangler
  _build deps
  venv .venv
)

# Conditional: name is generic, mark only if a sibling marker file proves
# intent. Patterns are globbed against entries in the candidate's PARENT.
CONDITIONAL_NAMES=( target vendor env dist build out coverage .cache .bundle bin obj )

siblings_for() {
  case "$1" in
    target)   echo "Cargo.toml" ;;
    vendor)   echo "go.mod composer.json Gemfile" ;;
    env)      echo "pyvenv.cfg" ;;
    dist)     echo "package.json setup.py pyproject.toml" ;;
    build)    echo "package.json build.gradle pom.xml CMakeLists.txt" ;;
    out)      echo "package.json" ;;
    coverage) echo "package.json pyproject.toml Cargo.toml" ;;
    .cache)   echo "package.json Cargo.toml" ;;
    .bundle)  echo "Gemfile" ;;
    bin|obj)  echo "*.csproj *.fsproj *.vbproj" ;;
    *)        echo "" ;;
  esac
}

# --- helpers -----------------------------------------------------------------
in_list() {
  local needle="$1"; shift
  local item
  for item in "$@"; do [ "$item" = "$needle" ] && return 0; done
  return 1
}

parent_has_any() {
  local parent="$1"; shift
  local pattern
  local matches
  shopt -s nullglob
  for pattern in "$@"; do
    matches=( "$parent"/$pattern )
    if [ "${#matches[@]}" -gt 0 ]; then
      shopt -u nullglob
      return 0
    fi
  done
  shopt -u nullglob
  return 1
}

marked=0; would=0; skipped_existing=0; skipped_cond=0

process_dir() {
  local dir="$1"
  local marker="$dir/.metadata_never_index"
  if [ -e "$marker" ]; then
    [ "$verbose" -eq 1 ] && printf 'exists  %s\n' "$dir"
    skipped_existing=$((skipped_existing + 1))
    return
  fi
  if [ "$dry_run" -eq 1 ]; then
    printf 'would   %s\n' "$dir"
    would=$((would + 1))
  else
    touch "$marker"
    printf 'marked  %s\n' "$dir"
    marked=$((marked + 1))
  fi
}

# --- find --------------------------------------------------------------------
all_names=( "${ALWAYS[@]}" "${CONDITIONAL_NAMES[@]}" )

name_expr=()
for n in "${all_names[@]}"; do
  if [ "${#name_expr[@]}" -eq 0 ]; then
    name_expr+=( -name "$n" )
  else
    name_expr+=( -o -name "$n" )
  fi
done

# Prune .git so we don't walk pack files for nothing. Then prune-and-print
# any directory whose name matches our combined list.
while IFS= read -r -d '' dir; do
  base="${dir##*/}"
  if in_list "$base" "${ALWAYS[@]}"; then
    process_dir "$dir"
    continue
  fi
  parent="${dir%/*}"
  # shellcheck disable=SC2206  # intentional word-split on space-separated patterns
  patterns=( $(siblings_for "$base") )
  if [ "${#patterns[@]}" -eq 0 ] || parent_has_any "$parent" "${patterns[@]}"; then
    process_dir "$dir"
  else
    [ "$verbose" -eq 1 ] && printf 'skip    %s  (no sibling marker for %s/)\n' "$dir" "$base"
    skipped_cond=$((skipped_cond + 1))
  fi
done < <(find "$root" \
              -type d -name .git -prune -o \
              -type d \( "${name_expr[@]}" \) -prune -print0)

# --- summary -----------------------------------------------------------------
total_acted=$((marked + would))
total_seen=$((total_acted + skipped_existing + skipped_cond))
if [ "$dry_run" -eq 1 ]; then
  printf 'dry-run: would mark %d, %d already marked, %d skipped (no sibling), %d total under %s\n' \
    "$would" "$skipped_existing" "$skipped_cond" "$total_seen" "$root"
else
  printf 'marked %d, %d already marked, %d skipped (no sibling), %d total under %s\n' \
    "$marked" "$skipped_existing" "$skipped_cond" "$total_seen" "$root"
fi