
creating-agents
by aiskillstore
Security-audited skills for Claude, Codex & Claude Code. One-click install, quality verified.
SKILL.md
name: creating-agents description: | Creates Claude Code subagents for task delegation. Guides through agent configuration, tool selection, and system prompts. Use this skill when the user wants to create a new agent, delegate tasks to specialists, or asks about subagents.
Creating Agents
Guides creation of Claude Code subagents for task delegation.
Quick Start
- Define agent purpose (what task does it handle?)
- Choose location (project or user level)
- Select tools (minimal set needed)
- Write system prompt
- Save to
.claude/agents/
Workflow: Create New Agent
Progress:
- [ ] Define purpose and triggers
- [ ] Choose storage location
- [ ] Select tools and model
- [ ] Write system prompt
- [ ] Create agent file
Step 1: Define Purpose
Ask user:
- What specific task should this agent handle?
- When should it be invoked? (trigger phrases)
- Should it run proactively or on-demand?
Step 2: Choose Location
| Location | Path | Use For |
|---|---|---|
| Project | .claude/agents/ | Team-shared, project-specific |
| User | ~/.claude/agents/ | Personal, cross-project |
Project agents take priority over user agents.
Step 3: Select Tools and Model
Tools - Grant minimum needed:
| Tool | Purpose |
|---|---|
| Read | Read files |
| Write | Create files |
| Edit | Modify files |
| Glob | Find files |
| Grep | Search content |
| Bash | Run commands |
| Task | Spawn subagents |
Model - Choose based on task:
| Model | Best For |
|---|---|
opus | Complex reasoning, nuanced decisions |
sonnet | General tasks (default) |
haiku | Quick lookups, simple analysis |
inherit | Use parent's model |
Step 4: Write System Prompt
Keep prompts focused:
- State the agent's role clearly
- Define scope and constraints
- Provide examples if helpful
- Avoid unnecessary detail
Step 5: Create Agent File
---
name: {agent-name}
description: {when to use - include trigger words}
tools: Read, Grep, Glob
model: sonnet
---
{System prompt here}
Save to .claude/agents/{name}.md
Agent File Format
---
name: agent-name # Required: lowercase, hyphens
description: | # Required: when to invoke
Reviews code for quality issues.
Use when user asks for code review.
tools: Read, Grep, Glob # Optional: omit to inherit all
model: sonnet # Optional: opus, sonnet, haiku, inherit
permissionMode: default # Optional: permission handling
skills: skill1, skill2 # Optional: auto-load skills
---
System prompt defining the agent's behavior.
Built-in Agents
Before creating custom agents, know what's built-in:
| Agent | Model | Tools | Purpose |
|---|---|---|---|
| general-purpose | sonnet | All | Complex multi-step tasks |
| plan | haiku | Read, Glob, Grep, Bash | Research and strategy |
| explore | haiku | Read, Glob, Grep | Fast codebase exploration |
When to create custom agents:
- Need different tool restrictions
- Want domain-specific prompts
- Need proactive invocation
When to Use Each Type
| Need | Use |
|---|---|
| Quick file search | Built-in explore |
| Research before planning | Built-in plan |
| Multi-step code changes | Built-in general-purpose |
| Code review with specific rules | Custom reviewer agent |
| Security analysis | Custom security agent |
| Domain expertise (DB, API, etc.) | Custom specialist agent |
Proactive Invocation
To make Claude automatically use your agent, include in description:
- "PROACTIVELY" or "MUST BE USED"
- Clear trigger conditions
description: |
PROACTIVELY reviews all code changes before commit.
MUST BE USED when user mentions "review" or "check code".
Templates
Use templates from templates/ directory:
- templates/reviewer.md - Code review agent
- templates/researcher.md - Read-only research
- templates/specialist.md - Domain expert
See reference.md for complete configuration details.
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
1ヶ月以内に更新
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon
