# AI Command Rules > Use this skill whenever Claude needs to write, suggest, or execute shell commands, scripts, or terminal instructions — especially in AI coding contexts (Claude Code, Cursor, Copilot, agentic pipelines). Triggers include bash/shell scripts, SSH remote commands, Docker/container operations, Git commands, Python/virtual environment commands, PowerShell, Windows CMD, and any cross-platform CLI tasks. Must be consulted whenever the user mentions multiple environments (Windows/Linux/Mac, sandbox, venv, PowerShell, WSL, Docker), asks Claude to run or execute commands or write a script, or when Claude is about to emit a command block. Prevents syntax errors, wrong-shell assumptions, and commands that break silently across platforms. - Skill: `tomevault-io/ai-command-rules` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/ai-command-rules` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/ai-command-rules/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/ai-command-rules --- # AI Command Rules A skill for writing precise, environment-aware shell commands and scripts that work correctly the first time across Windows, Linux, macOS, sandboxes, virtual environments, and containers. ## Core Principle **Always detect the environment before writing a command.** Never assume. A command that works on Linux bash will silently fail or behave differently on Windows CMD, PowerShell, or macOS zsh. When in doubt, ask or emit environment-adaptive commands. --- ## Step 1: Detect the Environment Before writing any command, identify: | Signal | How to detect | |---|---| | OS | User mentions, file paths (C:\\ vs /), or ask explicitly | | Shell | bash / zsh / fish / PowerShell / CMD / sh | | Runtime context | Local terminal, Docker container, CI/CD, sandbox, SSH session | | Python env | venv, conda, pipenv, poetry, system Python, Docker | | Privileges | Root/sudo available? Admin on Windows? | **If uncertain**, emit a detection snippet first: ```bash # Universal OS+shell detector (paste in terminal first) echo "OS: $(uname -s 2>/dev/null || echo Windows)" echo "Shell: $SHELL" echo "Python: $(python --version 2>&1 || python3 --version 2>&1)" ``` --- ## Step 2: Environment-Specific Rules Read the relevant reference file before emitting commands: - **Linux / macOS / bash / zsh** → `references/unix.md` - **Windows CMD / PowerShell / WSL** → `references/windows.md` - **Docker / containers / sandboxes** → `references/containers.md` - **Python / virtual environments** → `references/python-env.md` - **SSH / remote execution** → `references/ssh-remote.md` - **Git** → `references/git.md` For cross-platform tasks, read all relevant files. --- ## Step 3: Command Writing Checklist Before emitting any command block, verify: ### ✅ Shell Compatibility - [ ] Correct quoting style for target shell (single vs double vs backtick) - [ ] Path separator correct (`/` vs `\`) - [ ] Line continuation correct (`\` bash vs `` ` `` PowerShell) - [ ] Variable syntax correct (`$VAR` bash vs `$env:VAR` PowerShell vs `%VAR%` CMD) - [ ] Logical operators correct (`&&` / `||` bash vs `-and` / `-or` PowerShell) ### ✅ Error Handling - [ ] Long scripts use `set -euo pipefail` (bash) or `$ErrorActionPreference = 'Stop'` (PS) - [ ] Destructive commands (`rm`, `del`, `DROP`) have a dry-run or confirmation step - [ ] Commands that may fail silently have explicit exit code checks ### ✅ Permissions & Privileges - [ ] `sudo` / `su` used only when needed; never prefix everything with sudo - [ ] Windows: flag if Admin PowerShell is required - [ ] Docker: flag if `--privileged` or volume mounts need host permission ### ✅ Environment Isolation - [ ] Python commands use the correct interpreter (`python` vs `python3` vs `./venv/bin/python`) - [ ] npm/node commands run inside the correct project directory - [ ] Docker commands target the correct container/image name ### ✅ Cross-Platform Output - [ ] If user may run on multiple OSes, provide alternatives (see formats below) --- ## Step 4: Output Formats ### Single environment — emit one block with a label: ```` ```bash # Linux / macOS pip3 install -r requirements.txt ``` ```` ### Cross-platform — emit tabbed alternatives: ```` **Linux / macOS (bash/zsh)** ```bash python3 -m venv .venv && source .venv/bin/activate ``` **Windows (PowerShell)** ```powershell python -m venv .venv; .\.venv\Scripts\Activate.ps1 ``` **Windows (CMD)** ```cmd python -m venv .venv && .venv\Scripts\activate.bat ``` ```` ### Agentic / AI coding context — always add safety guards: ```` ```bash set -euo pipefail # Exit on error, undefined vars, pipe failures # ... rest of script ``` ```` --- ## Common Pitfalls to Avoid | ❌ Wrong | ✅ Right | Why | |---|---|---| | `python script.py` | `python3 script.py` (Unix) | macOS/Linux default `python` may be 2.x | | `pip install X` | `pip install X --break-system-packages` (system Python) | PEP 668 restriction | | `rm -rf dist/` | `rm -rf ./dist/` | Protects against accidental root deletion | | `cd /app && run.sh` | `cd /app && ./run.sh` | PATH may not include `.` | | `source activate` | `conda activate myenv` | conda >= 4.6 deprecated old syntax | | `docker run image cmd` | `docker run --rm image cmd` | Avoid orphaned containers | | `ssh user@host cmd` | `ssh -o StrictHostKeyChecking=no user@host 'cmd'` | Non-interactive SSH needs explicit options | | `&&` in PowerShell | `;` or `-and` | `&&` not supported in PS < 7 | | `%VAR%` in PowerShell | `$env:VAR` | Different variable syntax | | `\n` in Windows paths | `/` or `\\` | Use forward slashes where possible | --- ## Sandbox / Restricted Environment Rules When Claude itself is executing commands (Claude Code, agentic pipelines, sandboxed VMs): 1. **Probe before assuming** — run `which`, `command -v`, or `Get-Command` to verify tool availability 2. **No interactive prompts** — use `-y`, `--yes`, `--non-interactive`, `-f` flags 3. **Absolute paths preferred** — don't rely on PATH being correct 4. **Temp files** — write to `/tmp` (Linux) or `$env:TEMP` (Windows), not current directory 5. **Network** — check if egress is allowed before running `curl`, `pip install`, `npm install` 6. **No `sudo` in containers** — usually running as root already; `sudo` may not exist --- ## Quick Reference: Shell Detection Snippets ```bash # Detect if running inside Docker [ -f /.dockerenv ] && echo "In Docker" || echo "Not Docker" # Detect OS in bash case "$(uname -s)" in Linux*) OS=Linux ;; Darwin*) OS=Mac ;; CYGWIN*|MINGW*|MSYS*) OS=Windows ;; esac # Detect active Python venv [ -n "$VIRTUAL_ENV" ] && echo "venv: $VIRTUAL_ENV" || echo "No venv active" # Detect conda env [ -n "$CONDA_DEFAULT_ENV" ] && echo "conda: $CONDA_DEFAULT_ENV" ``` ```powershell # PowerShell: detect version $PSVersionTable.PSVersion # PowerShell: detect if running as Admin ([Security.Principal.WindowsPrincipal][Security.Principal.WindowsIdentity]::GetCurrent()).IsInRole([Security.Principal.WindowsBuiltInRole]::Administrator) ``` --- > Source: [russodope/ai-cli-kit](https://github.com/russodope/ai-cli-kit) — distributed by [TomeVault](https://tomevault.io).