Menu
Core principles for designing effective subagents. Use when asked to "design a subagent", "structure a subagent", "subagent best practices". Provides essential guidance with references for deep dives.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
This skill should be used when the user asks to "set up Notion MCP", "explain OAuth flow", "understand Notion MCP", "how to authenticate with Notion", or wants to learn about the Notion MCP setup process. Provides understanding of what Notion MCP is and why OAuth-based setup is required.
This skill should be used when the user asks to "what Notion MCP tools are available", "how to use read_page", "Notion MCP tool reference", "when to use query_database", "Notion MCP capabilities", or needs a guide to what operations are possible with Notion MCP. Provides reference to all available tools and when to use each.
Technical reference for implementing subagents. Use when asked about frontmatter fields, system prompt structure, or subagent file format. Provides essential specs with references for details.
Baseado na classificação ocupacional SOC
| name | Subagent Design Patterns |
| description | Core principles for designing effective subagents. Use when asked to "design a subagent", "structure a subagent", "subagent best practices". Provides essential guidance with references for deep dives. |
| version | 0.2.0 |
A subagent is an autonomous AI assistant with custom configuration, operating in its own context window.
| Use Main Conversation | Use Subagents |
|---|---|
| Iterative refinement | Verbose output (logs, data) |
| Shared context across phases | Tool restrictions needed |
| Frequent user feedback | Self-contained tasks |
| Low latency required | Parallel/background work |
Key trade-off: Subagents isolate context but start fresh. Use for self-contained tasks with summary outputs.
For detailed analysis, see references/when-to-use.md.
Write descriptions with:
Poor: description: Helps with debugging
Strong: description: Database query performance specialist. Use proactively when analyzing slow queries or explaining execution plans. Not for schema design.
See references/role-clarity.md for complete guide with examples.
Grant minimum necessary tools:
| Pattern | Tools | Use For |
|---|---|---|
| Read-only | ["Read", "Grep", "Glob"] | Reviewers, analyzers |
| Generation | ["Read", "Write", "Bash"] | Creators, scaffolders |
| Modification | ["Read", "Edit", "Bash"] | Fixers, debuggers |
See references/tool-patterns.md for complete patterns and examples.
| Model | Best For | Cost |
|---|---|---|
haiku | Data analysis, validation | ~90% cheaper |
sonnet | Code tasks (recommended default) | Balanced |
opus | Complex reasoning | Highest |
inherit | Match parent (most common) | Varies |
default - Standard prompts (testing)acceptEdits - Auto-accept modifications (trusted agents)dontAsk - Auto-deny unpermitted (automation)| Pattern | Role | Tools | Model |
|---|---|---|---|
| Reviewer | Code/doc analysis | Read-only | Sonnet |
| Analyzer | Data/log processing | Read+Bash | Haiku |
| Generator | Create artifacts | Read+Write | Sonnet |
| Fixer | Debug/fix issues | Read+Edit+Bash | Sonnet |
| Validator | Check configs | Read+Bash | Haiku |
For deeper guidance:
references/role-clarity.md - Writing powerful descriptionsreferences/tool-patterns.md - Tool access with examplesreferences/parallel-execution.md - Background/parallel design