⚠
High Risk:This skill has significant security concerns. Review the findings below before installing.

sub-agent-patterns

⚠Caution·Scanned 2/18/2026

This skill documents sub-agent creation, configuration, orchestration, and delegation for Claude Code. It explicitly instructs executing shell commands (git, npm, cat > file << 'EOF', ./scripts/run-linter.sh), modifying agent configs (.claude/agents/, settings.json), and making external requests (https://code.claude.com/docs/en/sub-agents, WebFetch).

by jezweb·v10a1f16·42.5 KB·397 installs
Scanned from main at 10a1f16 · Transparency log ↗
$ vett add jezweb/claude-skills/sub-agent-patternsReview security findings before installing

Sub-Agent Patterns Skill

Comprehensive guide to sub-agents in Claude Code: built-in agents, custom agent creation, configuration, and delegation patterns.

Auto-Trigger Keywords

This skill should be suggested when the user mentions:

Sub-Agent Concepts:

  • "sub-agent", "sub-agents", "subagent", "subagents"
  • "parallel agents", "multiple agents"
  • "Task tool", "launch agents"
  • "delegate tasks", "delegation"

Built-in Agents:

  • "Explore agent", "explore subagent"
  • "Plan agent", "plan subagent"
  • "general-purpose agent"
  • "thoroughness level", "quick/medium/thorough"

Custom Agents:

  • ".claude/agents"
  • "~/.claude/agents"
  • "/agents command"
  • "create agent", "custom agent"
  • "--agents flag"

Configuration:

  • "agent tools", "agent model"
  • "permissionMode", "agent permissions"
  • "agent skills", "agent hooks"
  • "resumable agent", "resume agent"
  • "disable agent", "Task(Explore)"

Bulk Operations:

  • "bulk operations", "batch processing"
  • "audit all", "update all", "check all"
  • "swarm", "multi-agent"

Errors:

  • "agent context", "agent timeout"
  • "agent not responding"

What This Skill Covers

  1. Built-in Sub-Agents - Explore (Haiku, read-only), Plan (Sonnet, plan mode), General-purpose (Sonnet, all tools)
  2. Custom Agent Creation - File format, YAML frontmatter, .claude/agents/ locations
  3. Configuration Fields - name, description, tools, model, permissionMode, skills, hooks
  4. Model Selection - sonnet, opus, haiku, inherit
  5. /agents Command - Interactive agent management
  6. CLI Configuration - --agents '{...}' flag
  7. Resumable Agents - Continue previous conversations with agentId
  8. Disabling Agents - Task(AgentName) permission rules
  9. Delegation Patterns - The "repetitive + judgment" sweet spot
  10. Prompt Templates - 5-step structure for audits, updates, research
  11. Batch Sizing - 5-8 items per agent, 2-4 agents parallel
  12. Commit Strategy - Agents edit, human commits

Quick Example

Built-in Explore Agent

User: Where are errors handled?
Claude: [Invokes Explore with "medium" thoroughness]
       → Returns: src/services/process.ts:712

Custom Agent File (.claude/agents/code-reviewer.md)

---
name: code-reviewer
description: Expert code reviewer. Use proactively after code changes.
tools: Read, Grep, Glob, Bash
model: inherit
---

You are a senior code reviewer...

Delegation Prompt Template

For each [item]:
1. Read [source file]
2. Verify with [external check]
3. Check [authoritative source]
4. Evaluate/score
5. FIX issues found  ← Key instruction

Items: [list of 5-8 items]

When NOT to Use Sub-Agents

  • Single complex task (do it yourself)
  • Simple find-replace (use scripts)
  • Tasks with dependencies between items
  • Creative/subjective decisions

Production Tested

  • Skill audits: 68 skills across 10 tiers (parallel batches)
  • Description optimization: 58 skills in ~3 minutes
  • Multi-site updates: 12 sites with consistent changes

Related Skills

  • project-planning - For planning complex multi-phase work
  • project-workflow - For session management and handoffs

References