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
code
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:
- •Diagnosis - Core type (Skill / Agent) + whether Command wrapper is needed
- •Spec Template - Filled template from
references/templates/ - •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 |