Skill Shell Quality Plus

name: shell-quality-plus description: Unified shell script quality: ShellCheck linting + BATS testing. Use for CLI scripts, headless agent commands (gemini, codex, droid, claude, kilo), or any bash script. IRON RULE: Lint first, test always. Triggers on "shell script", "bash test", "shellcheck", "CLI script", "agent command". allowed-tools: [shellcheck, Write, Bash, Read, Glob]

Shell Quality Plus - Unified Shell Script Quality

"Talk is cheap. Show me the code." - Linus Torvalds

IRON RULE: Lint First, Test Always

FOR shell scripts:
  1. Run ShellCheck BEFORE executing
  2. Write BATS tests AFTER creating the script
  3. Both are REQUIRED for production scripts

NO EXCEPTIONS for CLI scripts, agent commands, or any production shell script.

When to Use This Skill

• Writing CLI scripts for headless agents (gemini, codex, droid, claude, kilo)
• Creating bash scripts with multiple options/flags
• Testing shell script behavior
• Validating scripts before execution
• Setting up shell script CI/CD pipelines
• Debugging shell script issues

Two-Phase Workflow

┌─────────────────────────────────────────────────────────────┐
│  PHASE 1: LINTING (ShellCheck)                              │
│  → Verify code quality before execution                     │
│  → Catch common pitfalls                                    │
│  → Configure for target shell (bash/sh)                     │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│  PHASE 2: TESTING (BATS)                                    │
│  → Test functional behavior                                 │
│  → Mock external commands                                   │
│  → CI/CD integration                                        │
└─────────────────────────────────────────────────────────────┘

Phase 1: ShellCheck Linting

Quick Reference: Common Error Codes

Run ShellCheck

# Basic check
shellcheck script.sh

# Check all scripts in project
find . -name "*.sh" -exec shellcheck {} \;

# With specific shell target
shellcheck --shell=bash script.sh

# Exclude specific warnings
shellcheck --exclude=SC1091,SC2086 script.sh

# Strict mode (all warnings enabled)
shellcheck --enable=all script.sh

ShellCheck Configuration (.shellcheckrc)

Create .shellcheckrc in project root:

# Target shell
shell=bash

# Enable optional checks
enable=avoid-nullary-conditions,require-variable-braces,check-unassigned-uppercase

# Disable common false positives
disable=SC1091  # Not following sourced files
disable=SC2086  # Double quote to prevent globbing (sometimes too noisy)
disable=SC2317  # Command unreachable exit code

Suppress Inline Warnings

#!/bin/bash
# shellcheck disable=SC2086
for i in $list; do
    echo "$i"
done

Phase 2: BATS Testing

Install BATS

# macOS
brew install bats-core/bats/bats

# From source
git clone https://github.com/bats-core/bats-core.git
cd bats-core
./install.sh /usr/local

Test Structure

#!/usr/bin/env bats

setup() {
    load 'test_helper/bats-support/load'
    load 'test_helper/bats-assert/load'
    export TEST_MODE=1
}

teardown() {
    rm -rf "$BATS_TEST_TMPDIR" 2>/dev/null || true
}

@test "script accepts --approval flag" {
    run ./script.sh --approval=auto
    assert_success
    assert_output --partial "approved"
}

@test "agent command executes correctly" {
    AGENT_TYPE=claude run ./script.sh --test
    assert_success
    assert_output --regexp "agent.*executed"
}

Running Tests

# Basic execution
bats test/

# Parallel execution
bats --jobs 4 test/

# Filter by name
bats --filter "approval" test/

# JUnit output for CI
bats --formatter junit --output ./reports test/

Template: CLI Script for Headless Agents

Use this template for scripts that call agents (gemini, codex, droid, claude, kilo):

#!/bin/bash
# =============================================================================
# agent-cli-template.sh - Template for headless agent CLI scripts
# =============================================================================

# shellcheck disable=SC2086  # Allow word splitting for arrays
# shellcheck disable=SC2154  # Variables set by CLI args

set -Eeuo pipefail  # ERR trap, exit on error, unset=error, pipefail

# -----------------------------------------------------------------------------
# CONFIGURATION
# -----------------------------------------------------------------------------
AGENT_TYPE="${AGENT_TYPE:-claude}"    # gemini/codex/droid/claude/kilo
APPROVAL_MODE="${APPROVAL_MODE:-auto}"  # auto/manual/confirm
OUTPUT_FORMAT="${OUTPUT_FORMAT:-json}"   # json/text/markdown
SCRIPT_PATH="${SCRIPT_PATH:-}"
VERBOSE="${VERBOSE:-false}"

# -----------------------------------------------------------------------------
# HELPER FUNCTIONS
# -----------------------------------------------------------------------------

log() {
    if [[ "$VERBOSE" == "true" ]]; then
        echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*"
    fi
}

error() {
    echo "ERROR: $*" >&2
    exit 1
}

# Execute command on target agent
execute_agent_command() {
    local agent="$1"
    local cmd="$2"

    case "$agent" in
        gemini)
            gemini-cli exec --script="$cmd" --approval="$APPROVAL_MODE" --format="$OUTPUT_FORMAT"
            ;;
        codex)
            codex run "$cmd" --no-confirm --output="$OUTPUT_FORMAT"
            ;;
        droid)
            droid-cli execute "$cmd" --headless --approval="$APPROVAL_MODE"
            ;;
        claude|kilo)
            claude-code "$cmd" --approval="$APPROVAL_MODE" --format="$OUTPUT_FORMAT"
            ;;
        *)
            error "Unknown agent: $agent"
            ;;
    esac
}

# Validate prerequisites
validate_prerequisites() {
    local agent="$1"
    command -v "$agent" >/dev/null 2>&1 || error "$agent not found in PATH"
}

# Show usage
usage() {
    cat <<EOF
Usage: $(basename "$0") [OPTIONS]

Execute commands on headless agents.

OPTIONS:
    --agent AGENT      Agent type: gemini, codex, droid, claude, kilo (default: claude)
    --approval MODE    Approval mode: auto, manual, confirm (default: auto)
    --format FORMAT    Output format: json, text, markdown (default: json)
    --script FILE      Script file to execute
    --verbose          Enable verbose output
    --help             Show this help

EXAMPLES:
    $(basename "$0") --agent claude --script ./my-script.sh --approval auto
    $(basename "$0") --agent codex --script ./deploy.sh --format json
    $(basename "$0") --agent droid --verbose --script ./cleanup.sh
EOF
}

# -----------------------------------------------------------------------------
# MAIN
# -----------------------------------------------------------------------------

main() {
    # Parse arguments
    while [[ $# -gt 0 ]]; do
        case "$1" in
            --agent)
                AGENT_TYPE="$2"
                shift 2
                ;;
            --approval)
                APPROVAL_MODE="$2"
                shift 2
                ;;
            --format)
                OUTPUT_FORMAT="$2"
                shift 2
                ;;
            --script)
                SCRIPT_PATH="$2"
                shift 2
                ;;
            --verbose)
                VERBOSE=true
                shift
                ;;
            --help)
                usage
                exit 0
                ;;
            *)
                error "Unknown option: $1"
                ;;
        esac
    done

    # Validate
    [[ -n "$SCRIPT_PATH" ]] || error "Script path is required (--script)"
    [[ -f "$SCRIPT_PATH" ]] || error "Script not found: $SCRIPT_PATH"

    log "Agent: $AGENT_TYPE"
    log "Approval: $APPROVAL_MODE"
    log "Format: $OUTPUT_FORMAT"
    log "Script: $SCRIPT_PATH"

    # Validate prerequisites
    validate_prerequisites "$AGENT_TYPE"

    # Execute
    log "Executing agent command..."
    execute_agent_command "$AGENT_TYPE" "$SCRIPT_PATH"

    log "Done!"
}

# Run in test mode if TEST_MODE is set (for BATS)
if [[ -n "${TEST_MODE:-}" ]]; then
    main "$@" > /dev/null 2>&1 || exit 1
else
    main "$@"
fi

Common ShellCheck Violations & Fixes

SC2086: Double Quote to Prevent Globbing

# BAD
for i in $list; do
    echo $i
done

# GOOD
for i in "$list"; do
    echo "$i"
done

# GOOD (if list is array)
for i in "${list[@]}"; do
    echo "$i"
done

SC2181: Check Exit Code Directly

# BAD
some_command
if [ $? -eq 0 ]; then
    echo "success"
fi

# GOOD
if some_command; then
    echo "success"
fi

SC2154: Referenced but Not Assigned

# BAD - $AGENT_TYPO is not set
echo "$AGENT_TYPO"

# GOOD - Check if set first
echo "${AGENT_TYPE:-default}"

SC2317: Command Unreachable Exit Code

# BAD - return after exit
exit 1
echo "unreachable"

# GOOD - remove unreachable code
exit 1

BATS Test Examples for CLI Scripts

#!/usr/bin/env bats

setup() {
    load 'test_helper/bats-support/load'
    load 'test_helper/bats-assert/load'
    export TEST_MODE=1
}

@test "script fails without --script argument" {
    run ./agent-cli-template.sh
    assert_failure
    assert_output --partial "Script path is required"
}

@test "script fails with unknown agent" {
    run ./agent-cli-template.sh --agent unknown --script test.sh
    assert_failure
    assert_output --partial "Unknown agent"
}

@test "script shows help with --help" {
    run ./agent-cli-template.sh --help
    assert_success
    assert_output --partial "Usage:"
    assert_output --partial "--agent"
    assert_output --partial "--approval"
}

@test "script accepts all valid agents" {
    for agent in claude kilo; do
        run ./agent-cli-template.sh --agent "$agent" --script test.sh
        assert_failure  # Fails because test.sh doesn't exist, but agent is valid
        refute_output --partial "Unknown agent"
    done
}

@test "script respects --verbose flag" {
    VERBOSE=true run ./agent-cli-template.sh --agent claude --script test.sh 2>&1
    assert_output --partial "Agent:"
    assert_output --partial "Approval:"
}

CI/CD Integration

GitHub Actions

name: Shell Quality

on: [push, pull_request]

jobs:
  shell-quality:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4

      - name: Install ShellCheck
        run: sudo apt-get install -y shellcheck

      - name: Install BATS
        run: |
          git clone https://github.com/bats-core/bats-core.git
          ./bats-core/install.sh /usr/local

      - name: Run ShellCheck
        run: |
          find . -name "*.sh" -exec shellcheck --exclude=SC1091 {} \;

      - name: Run BATS tests
        run: |
          bats --formatter junit --output ./reports test/

      - name: Publish test results
        uses: EnricoMi/publish-unit-test-result-action@v2
        if: always()
        with:
          files: reports/report.xml

Quick Reference

ShellCheck Commands

# Basic
shellcheck script.sh

# All scripts
find . -name "*.sh" -exec shellcheck {} \;

# With exclusions
shellcheck --exclude=SC1091,SC2086 script.sh

# JSON output (for CI parsing)
shellcheck --format=json script.sh

BATS Commands

# Run tests
bats test/

# Parallel
bats --jobs 4 test/

# Filter
bats --filter "approval" test/

# JUnit output
bats --formatter junit --output ./reports test/

After Creating a Shell Script

1. Run ShellCheck - Fix all warnings (or document why not)
2. Write BATS tests - Test main functionality
3. Run BATS - Verify tests pass
4. Integrate in CI/CD - Automatic quality checks

Remember: Lint First, Test Always. No exceptions for production scripts.