Skip to content
Back to skills

Shell Patterns

ASecurity

Background job patterns with shell and job management tools

  • 21 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 2, 2026
ai-agentsrustshell

Works with

  • terminal

Security analysis

A100/100

Scanned September 24, 2026

npx -y skills add lukacf/meerkat --skill shell-patterns --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Shell Patterns?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Shell Patterns
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/lukacf-shell-patterns/badge)](https://www.skillsdirectory.com/skills/lukacf-shell-patterns)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: Shell Patterns
description: Background job patterns with shell and job management tools
requires_capabilities: [builtins, shell]
---

# Shell Patterns

Use shell tools for concrete local commands, verification, and long-running
processes. Keep commands scoped to the user's workspace and report important
results back into the conversation.

## Background Prerequisites

`background: true` requires canonical session and operation bindings plus a
persistent realm job store, persistent blob store, and delivery projector.
Shell availability alone does not provide these resources: a shell-enabled
standalone or ephemeral composition can reject background execution with a
typed tool error. There is no process-local detached-job fallback. Use
foreground shell there (`background: false`, the default).

## Operating Rules

- Use `shell` for short foreground commands.
- When the durable runtime above is wired, use `shell` with `background: true`
  for ordinary long-running commands. Keep the returned `job_id`, then inspect
  with `shell_job_status` or list jobs with `shell_jobs`. A queued receipt
  acknowledges submission, not command success; inspect the terminal result
  before claiming completion.
- A background shell submission is durable, but its worker is explicitly
  non-resumable. After a restart the job record remains and a lost worker
  becomes `worker_lost`; Meerkat does not silently replay the command.
- Use `monitor_start` only when that high-trust tool is explicitly visible and
  the script follows its declared protocol. Framed JSONL can emit typed
  notification, checkpoint, progress, and complete frames; notifications do
  not complete the job.
- Cancel stuck or obsolete jobs with `shell_job_cancel`.
- Shell output is truncated. Shape large output with tools such as `rg`,
  `head`, `tail`, or structured command flags.
- Set the working directory intentionally. Prefer absolute paths when a command
  crosses project boundaries.
- Treat shell output as evidence to reason from, not as durable work state. Add
  WorkGraph evidence separately when a shared item needs proof.
- Shell is classified as mutating because Meerkat cannot prove an arbitrary
  command is read-only. It is unavailable under read-only tool policy.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.