meta-llm-type

name	meta-llm-type
description	Diagnose features into the right LLM component type — Skill, Agent, or Command wrapper. Use when users ask whether something should be a skill, need to decide between command or agent, want to diagnose a feature's type, or make structure and architecture decisions for LLM components. Do NOT use for actual implementation or general coding tasks.

Structure Organizer

Organize features into the right structure: Command, Skill, or Agent for AI coding assistants.

Quick Reference

Core Types (Knowledge Layer)

Component	Trigger	Reasoning	Execution	Use When
Skill	Auto-load on keywords / direct @path	None	No execution (knowledge)	Domain expertise to share
Agent	Goal assigned	LLM decides	Dynamic, iterative	Multi-step planning required

Optional Wrapper (Access Layer)

Component	Trigger	Purpose	Use When
Command	Human /command	UI entry point + constraints over Skill/Agent	allowed-tools restriction, dangerous ops, structured $ARGUMENTS, frequent shortcut

Key insight: Command is NOT a parallel type to Skill/Agent. It is an access pattern — a UI + security wrapper placed over Skills or Agents when human entry point and platform constraints are needed.

Workflow Routing

Intent	Workflow
Analyze a feature request	workflows/analyze.md
Generate spec template	workflows/generate-spec.md

Core Resources

Resource	Purpose
Decision Tree	Primary decision logic
Criteria Matrix	Detailed comparison criteria
Boundary Cases	12 common confusions
Combination Patterns	Multi-component architectures
Templates	Spec templates for each type

Architecture Model

Knowledge Layer:  Skill (knowledge)  |  Agent (reasoning)
Access Layer:     Command (optional UI + constraints wrapper)

Common Patterns

Pattern	Structure	Use When
Skill only	Direct invocation via @path or keywords	Most cases — knowledge is self-sufficient
Agent + Skills	Executor + Knowledge	Agent needs domain expertise
Command wrapping Skill	Entry + Knowledge + Constraints	Tool restriction or frequent human shortcut needed
Command wrapping Agent	Entry → Executor	User triggers dangerous/complex work
Full Stack	Command → Agent → Skills → Tools	Critical ops needing all layers

See combination-patterns.md for details.

Output

This skill outputs:

1. Diagnosis - Core type (Skill / Agent) + whether Command wrapper is needed
2. Spec Template - Filled template from references/templates/
3. Rationale - Why this type, why not others, and why Command wrapper is or isn't needed

Note: This skill diagnoses what type to use. Implementation is handled by the calling Command.

Platform Support

Platform	Commands	Skills	Agents
Claude Code	.claude/commands/*.md	.claude/skills/*/SKILL.md	Subagent via Task
OpenCode	.opencode/commands/*.md	skills/*/SKILL.md	agents/*.md
Cursor	.cursor/commands/*.md	.cursor/rules/*.mdx	Agent mode