Bash Shell Scripting

Bash & Shell Scripting

Core Principles

• DRY: Don't Repeat Yourself
• KISS: Keep It Simple
• Fail Fast: Exit on errors immediately
• Zero Warnings: Must pass shellcheck

Quick Reference

set -euo pipefail           # Strict mode (fail-fast)
set -uo pipefail            # Controlled mode (explicit error handling)
set -Euo pipefail           # Strict + ERR trap propagation
command -v cmd >/dev/null   # Check if command exists (portable)
trap 'cleanup' EXIT         # Always cleanup
flock -n 200 || exit 1      # Prevent concurrent runs
readonly VAR="value"        # Immutable constant
local var="value"           # Function-local variable

Core Standards

| Aspect | Standard | |--------|----------| | Shebang | #!/usr/bin/env bash or #!/bin/bash | | Safety Mode | set -euo pipefail (strict) or set -uo pipefail (controlled) | | Linting | Must pass shellcheck with 0 errors/warnings | | Formatting | Must pass shfmt -i 2 -ci -sr -bn; prefer ~100-character lines | | Extension | .sh for scripts | | Variables | Uppercase constants/globals, lowercase locals, descriptive names, braces + quotes ("${VAR}") |

Generation Contract

For any non-trivial script (more than 10-15 lines, more than one command phase, argument parsing, dependency checks, config resolution, build/deploy/verify steps), generate this structure:

1. Header block.
2. Commented debug toggle (# set -x), disabled by default.
3. Strict or controlled mode.
4. Readonly constants and timestamped LOGFILE.
5. logmsg, die, and optional debug helper.
6. require_command for dependencies.
7. Helper functions with lowercase locals.
8. parse_args.
9. main().
10. main "$@" as the final line.

Reject generated Bash that skips this structure unless the script is intentionally tiny and linear.

Error Handling Modes

Strict Mode (Fail-Fast)

set -euo pipefail  # Exit immediately on any error

Use for: Simple linear scripts, dependency installation, straightforward validation.

Controlled Mode (Explicit)

set -uo pipefail  # No -e: handle errors explicitly

Use for: Diagnostics, cleanup operations, commands where failure is expected.

Strict + ERR Trap

set -Euo pipefail

trap 'echo "ERROR in ${FUNCNAME[0]:-main} at line $LINENO"' ERR

Use for: Production scripts with comprehensive error handling.

Script Template

#!/usr/bin/env bash
#
# Script Name         : .sh
#
# Purpose             : 
#
# Dependencies        : 
#
# Script Usage        : ./.sh [options] 
#
#                       
#
##----------------------------------------------------------------------------------------##
# Turn debug on or off
# set -x
##----------------------------------------------------------------------------------------##

set -euo pipefail

readonly SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
readonly DTTM="$(date -u +"%Y%m%d_%H%M%S")"
readonly SCRIPT_NAME="$(basename "${0}" .sh)"
readonly LOGFILE="${SCRIPT_NAME}_${DTTM}.log"

cleanup() {
  rm -f "${TEMP_FILE:-}" 2>/dev/null || true
}
trap cleanup EXIT

logmsg() {
  local timestamp
  timestamp="$(date -u +"%Y-%m-%dT%H:%M:%SZ")"
  printf "%s: %s\n" "${timestamp}" "$*" | tee -a "${LOGFILE}" >&2
}

die() {
  logmsg "ERROR: $*"
  exit 1
}

debug() {
  [[ "${DEBUG:-0}" == "1" ]] && logmsg "DEBUG: $*"
}

require_command() {
  local command_name="$1"
  command -v "${command_name}" >/dev/null 2>&1 || die "Missing command: ${command_name}"
}

main() {
  local arg="${1:-}"
  [[ -z "${arg}" ]] && die "Usage: ${SCRIPT_NAME} "

  require_command jq
  logmsg "Processing: ${arg}"
  # Main logic here
}

main "$@"

Minimal dependency helper:

require_command() {
  local command_name="$1"
  command -v "${command_name}" >/dev/null 2>&1 || die "Missing command: ${command_name}"
}

Best Practices

Always Quote Variables

# ✅ GOOD
echo "${var}"
[[ -n "${var:-}" ]] && echo "set"
printf "%s\n" "${array[@]}"

# ❌ BAD
echo $var
[ -n $var ] && echo "set"

Process Text and JSON Deliberately

# Simple string changes: prefer Bash parameter expansion
normalized_path="${input_path//\/\//\/}"

# Line/column text processing: awk/sed are appropriate when necessary
awk -F',' 'NR > 1 && $3 == "active" { print $1 }' users.csv
sed -E 's/[[:space:]]+$//' input.txt > output.txt

# JSON: use jq; never parse JSON with grep/sed/string splitting
jq -e '.users[] | select(.active == true) | .id' users.json

Terminal-safe ANSI Colors

Define color codes only when writing to an interactive terminal. Do not emit ANSI escapes into log files, CI summaries, or non-terminal output.

if [[ -t 2 ]]; then
  readonly RED=$'\033[0;31m'
  readonly GREEN=$'\033[0;32m'
  readonly YELLOW=$'\033[1;33m'
  readonly NC=$'\033[0m'
else
  readonly RED=''
  readonly GREEN=''
  readonly YELLOW=''
  readonly NC=''
fi

if [[ -n "${GREEN}" ]]; then
  printf "%sStarting job%s\n" "${GREEN}" "${NC}" >&2
fi
logmsg "Starting job"

if [[ -n "${RED}" ]]; then
  printf "%sInvalid input%s\n" "${RED}" "${NC}" >&2
fi
logmsg "Invalid input"

Use Functions

process_file() {
    local file="$1"
    [[ -f "$file" ]] || return 1
    # Process file
}

Know When Bash Is the Wrong Tool

Use Bash for glue: calling CLIs, moving files, simple validation, CI wrappers, and deployment orchestration. Prefer Python/Go/Node when the script needs complex data structures, non-trivial JSON transformation, API clients with pagination/retry state, concurrency, long-lived daemons, or more than a few hundred lines of business logic.

Use main() When the Script Has Phases

Tiny one-shot scripts can stay linear, but once a script has multiple named phases, wrap execution in main(). This separates globals/functions from execution, makes the script read like a table of contents, and makes future testing/refactoring easier.

main() {
  parse_args "$@"
  require_command curl
  require_command git
  require_command npm
  resolve_env_file
  install_dependencies_if_needed
  build_assets
  load_deploy_environment
  deploy_worker_assets "$@"
  verify_worker_hostnames
}

main "$@"

Check Command Existence

command -v docker >/dev/null 2>&1 || die "docker is required"

Temporary Files

readonly TEMP_FILE="$(mktemp)"
trap 'rm -f "$TEMP_FILE"' EXIT

echo "data" > "$TEMP_FILE"

Lock Files

exec 200>"/tmp/${SCRIPT_NAME}.lock"
flock -n 200 || { echo "Already running"; exit 1; }

Performance Patterns

Avoid Subshells in Loops

# ❌ BAD - subshell, variables don't persist
count=0
cat file.txt | while read -r line; do
    count=$((count + 1))
done
echo "$count"  # Empty! (subshell issue)

# ✅ GOOD - process substitution
count=0
while read -r line; do
    count=$((count + 1))
done 

Options:
    -h, --help      Show this help
    -v, --verbose   Enable verbose mode
    -o, --output    Output file (default: stdout)
EOF
}

parse_args() {
    VERBOSE=false
    OUTPUT=""

    while [[ $# -gt 0 ]]; do
        case "$1" in
            -h|--help) show_help; exit 0 ;;
            -v|--verbose) VERBOSE=true; shift ;;
            -o|--output) OUTPUT="$2"; shift 2 ;;
            -*) die "Unknown option: $1" ;;
            *) break ;;
        esac
    done

    [[ $# -eq 0 ]] && die "Missing required argument"
    INPUT_FILE="$1"
}

Detailed References

• Shell Utilities: See [references/shell-utilities.md](references/shell-utilities.md) for curl, jq, lynx
• Makefile Patterns: See [references/makefile-patterns.md](references/makefile-patterns.md)
• CLI Design: See [references/cli-design.md](references/cli-design.md)
• Justfile Patterns: See [references/justfile.md](references/justfile.md)
• Repo Sync (rsync): See [references/rsync-repo-sync.md](references/rsync-repo-sync.md)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: d-padmanabhan
• Source: d-padmanabhan/agent-engineering-handbook
• License: MIT

Install and usage instructions live in the source repository linked above.