writing-shell

agent	engineer
allowed-tools	["Read","Bash","Grep","Glob","Edit","Write","LS"]
context	fork
description	Idiomatic shell development for POSIX sh, Bash, Zsh, Fish, hooks, CI shell steps, and scriptable CLI glue. Use when writing or changing `.sh`, `.bash`, `.zsh`, `.fish`, `.bats`, shell functions, shell pipelines, CI `run:` shell bodies, or command-runner recipes. Emphasizes portability, quoting, safe filesystem/process handling, non-TUI CLI tools, ShellCheck, shfmt, Bats, and ShellSpec. NOT for Python, Rust, TypeScript, Go, web code, or GitHub Actions workflow/job/permissions semantics; use operating-infra.
name	writing-shell
user-invocable	false

Shell Development

Scope

• Use for shell scripts, hooks, CI shell blocks, command pipelines, and local automation glue.
• In GitHub Actions, own the shell code inside run: blocks. Use operating-infra for workflow YAML structure, jobs, permissions, runners, actions, secrets, caching, concurrency, and policy. Mixed changes compose both skills.
• Do not use for cloud, Kubernetes, Terraform, host, or network operations; use operating-infra.
• Do not use for application logic that belongs in Python, Rust, Go, TypeScript, or another project language.

Read references

• Read patterns.md before non-trivial scripts or pipelines.
• Read tools.md when choosing external CLI tools or parsing structured data.
• Read testing.md before adding or changing shell tests or quality gates.

Defaults

• Follow the existing shebang, shell, style, and project tooling first.
• For new portable scripts, use POSIX sh when the logic is simple; use Bash when arrays, pipefail, regex, or richer functions are needed.
• Use Zsh or Fish only for existing files or explicit user intent. Keep Fish/Zsh config separate from portable scripts.
• Do not rely on the agent's current shell. Put the intended shell in the shebang or invoke it explicitly in tests.
• Prefer small shell scripts that call stable tools. Move complex data modeling or business logic to a real language.

Core rules

• Quote expansions unless word splitting is intentional and documented by structure.
• Avoid eval, curl | sh, parsing ls, unguarded globbing, and unsafe rm/mv/cp paths.
• Use arrays for arguments in Bash; use newline/NUL-safe loops for filenames.
• Use mktemp plus cleanup traps for temporary files and directories.
• Check required external commands when a script depends on them. Do not install tools silently.
• Account for macOS/BSD versus GNU flag differences when writing portable scripts.
• Prefer machine-readable output from tools, then parse it with structured parsers.
• Use looking-up-docs for exact external CLI flags, syntax, or version behavior; do not guess from memory.

Comments

• Comment only tricky, non-obvious, important, or unusual shell behavior.
• Use file and function comments for reusable scripts or functions when the contract is not obvious from names and usage.
• Add a short reason after any shellcheck disable directive.
• Keep comments short. Move longer rationale to docs, issue links, or design notes.
• Do not comment obvious commands or narrate each pipeline stage.
• Keep shell tests readable without comments; add one only for unobvious fixtures, environment setup, portability constraints, or regression context.

Verification

Run the project-configured shell gates. Prefer:

• formatting: shfmt
• linting: shellcheck; checkbashisms for POSIX sh
• tests: Bats for Bash-heavy projects; ShellSpec for POSIX or multi-shell behavior
• security/policy: Semgrep shell rules when the script handles secrets, downloads, deletion, or user-controlled input

If a tool is unavailable or unconfigured, state the gap and run the closest available check. If a check fails, quote the diagnostic, fix the cause, and rerun the relevant check.

Final response

Include:

• changed files
• shell target used: POSIX sh, Bash, Zsh, or Fish
• checks run and results
• checks skipped with reasons
• remaining portability or safety risks