Commands & Skills
Commands
Type /skill-name in Claude Code to run.
Install package @shahboura/agents-claude, use agents-claude for install/update/uninstall, and run day-to-day workflows in claude.
Most commands accept an optional argument to scope the task: /skill-name [argument] — the argument is passed directly to the target agent.
| Command | Argument hint | Purpose |
|---|---|---|
/api-docs | [module, file, or endpoint path] | Generate API reference docs |
/architecture-decision | [decision topic or system name] | Create an ADR |
/architecture-review | [system, component, or design document] | Review design for quality attributes |
/blog-post | [topic or title] | Write a blog post |
/code-review | [file, PR, or scope — blank for current changes] | Security, perf, and style review |
/content-review | [content text, file path, or topic] | Score and critique content quality |
/create-readme | [project name or path] | Generate or improve a README |
/generate-tests | [file, class, or function name] | Generate targeted tests |
/plan-project | [feature, objective, or epic] | Multi-phase implementation plan |
/refactor-plan | [target module, file, or scope] | Safe staged refactor strategy |
/security-audit | [scope, file, component, or 'full project'] | Security-focused code review |
/1-on-1-prep | [person] [context] | Prepare for a 1-on-1 meeting |
Skills
Skills are reusable behaviors loaded on demand:
project-bootstrap- Create a minimalAGENTS.mdscaffoldagent-diagnostics- Validate agent setup and instruction coveragedocs-validation- Outline docs lint and link checks
Using Skills
Skills are defined in .claude/skills/[skill-name]/SKILL.md. Claude can invoke model-invocable skills automatically when relevant. You can always invoke user-invocable skills manually with /skill-name.
/project-bootstrap
/agent-diagnostics
For high-impact workflows, set disable-model-invocation: true to require manual invocation.
Recommended command-skill frontmatter (least privilege):
---
name: my-command
description: What it does
disable-model-invocation: true
user-invocable: true
argument-hint: "[path or scope]"
---
Note: This repository targets Claude skill schema where
context,agent, andallowed-toolsmay be unsupported depending on your Claude version. Keep skills schema-compliant first, then use agent/tool restrictions via.claude/agents/*.mdand.claude/settings.json.
Skill Selection Guardrails
- Load skills on demand for matching tasks only.
- Use one relevant skill by default; add a second only for clear cross-domain need.
- If technology/domain is ambiguous, ask for clarification before loading.
Installer Lifecycle Notes
agents-claudesupports scope-aware lifecycle operations:--update(auto-detects installed scopes)--update --all--uninstall --global--uninstall --all--status
- Uninstall is manifest-based and removes installer-managed files only.
- Backup sessions include
backup-manifest.jsonrestore hints.
Security Hardening
- Keep sensitive path denies in
.claude/settings.json. - Use PreToolUse hooks for command/path guardrails.
- Keep skill descriptions concise to minimize context overhead.
Custom Skills
Add a directory under .claude/skills/<name>/ containing SKILL.md. Use frontmatter like:
---
name: my-skill
description: What this skill does
disable-model-invocation: true
---
See Claude Skills docs for full reference.