Artifact Selector

Artifact Selector

When to Use This Skill

• Designing a new workflow
• Deciding what type of artifact to create
• Analyzing requirements to suggest artifact types
• Checking for redundancy with existing artifacts

Decision Tree

Use this flowchart to select the optimal artifact type:

                    ┌──────────────────┐
                    │ What triggers    │
                    │ the action?      │
                    └────────┬─────────┘
                             │
        ┌────────────────────┼────────────────────┐
        ▼                    ▼                    ▼
   User types           Automatic            Every matching
   /command             context match        operation
        │                    │                    │
        ▼                    ▼                    ▼
   ┌─────────┐         ┌─────────┐         ┌─────────┐
   │ COMMAND │         │  SKILL  │         │  HOOK   │
   └─────────┘         └─────────┘         └─────────┘
        │                    │
        │              ┌─────┴─────┐
        │              ▼           ▼
        │         Background   Side effects
        │         knowledge    (deploy, etc)
        │              │           │
        │              ▼           ▼
        │         user-invocable  disable-model-
        │         : false         invocation: true
        │
        │
   Needs isolated    ─────────────────────────▶  ┌─────────┐
   context? Yes                                   │  AGENT  │
                                                  └─────────┘

Keyword-to-Artifact Mapping

Artifact Selection Matrix

Choose COMMAND when:

• User should explicitly trigger (/command-name)
• Has observable side effects (deploy, commit, publish)
• Needs arguments from user
• One-time discrete action

Examples:

• /deploy - Deploy to production
• /pr - Create pull request
• /test run - Run specific tests

Choose SKILL when:

• Claude should automatically apply knowledge
• Provides patterns, conventions, or domain expertise
• Background context that applies across tasks
• Reusable knowledge without side effects

Examples:

• API design patterns skill
• Security review checklist skill
• Testing best practices skill

Choose AGENT when:

• Deep analysis requiring isolated context
• Parallel independent work
• Specialized expertise (security audit, code review)
• Task shouldn't pollute main conversation

Examples:

• Security audit agent
• Test generator agent
• Documentation agent

Choose HOOK when:

• Action must run on every matching operation
• Automation (formatting, linting, logging)
• Guardrails (block dangerous commands)
• No user interaction needed

Examples:

• Auto-format after file write
• Security check before bash command
• Audit logging after modifications

Choose CLAUDE.md when:

• Information applies to all sessions
• Project architecture and conventions
• Critical paths and gotchas
• Quick reference that's always relevant

Examples:

• Project structure overview
• Build and test commands
• Key architectural decisions

Redundancy Detection

Before creating a new artifact, check:

1. Scan existing commands: .claude/commands/**/*.md
2. Scan existing skills: .claude/skills/*/SKILL.md
3. Scan existing agents: .claude/agents/*.md
4. Check settings.json: hooks already configured

Signs of Redundancy:

• Similar trigger conditions in descriptions
• Overlapping functionality
• Same files/patterns being processed
• Duplicate validation logic

When to Combine vs Separate:

• Combine if artifacts share >70% of logic
• Separate if they have distinct triggers or contexts
• Extend existing artifact if adding related functionality

Parallel vs Sequential Execution

Use Parallel Agents when:

• Tasks are independent
• No shared state needed
• Different analysis perspectives
• Speed is important

Task security-audit "Audit auth code" &
Task perf-review "Review performance" &
Task test-coverage "Check coverage"

Use Sequential when:

• Output of one feeds into another
• Shared state required
• Order matters

Token Budget Guide

Anti-Patterns

❌ Creating a command when a hook would work

If it should run automatically on every file save, use a hook.

❌ Creating an agent for simple tasks

Agents have overhead. Simple reads don't need isolated context.

❌ Putting everything in CLAUDE.md

CLAUDE.md is for ambient context, not workflows. Use skills for patterns.

❌ Duplicate logic across artifacts

Extract shared logic to a skill that multiple artifacts reference.

❌ Over-scoping tool access

Give agents/skills only the tools they need. Don't grant write access for read-only tasks.