Awesome Claude Agents

The body of the agent file (below the front-matter) serves as the system prompt. For predictable performance, structure it as a micro-specification using the following sections:

1. Mission / Role: A single sentence defining the desired outcome.
2. Workflow: A numbered list of steps the agent must follow.
3. Output Contract: The exact Markdown or JSON format the agent must return.
4. Heuristics & Checks: A bulleted list of edge cases, validations, or scoring rubrics.
5. Delegation cues (Optional): Instructions like "If X is detected, ask <other-agent>."

Style Guidelines:

• Use active voice and imperative verbs.
• Use Markdown headings up to ### for readability.
• Wrap code fences with language tags.
• Avoid external links inside prompts; keep documentation offline.

To set up an optimal AI development team for your specific codebase, navigate to your project directory and invoke the @agent-team-configurator.

This agent will:

1. Locate and update your CLAUDE.md file (preserving existing content).
2. Detect your technology stack (e.g., by inspecting package.json, go.mod, etc.).
3. Map available specialists to your stack.
4. Create an "AI Team Configuration" section in CLAUDE.md with a Task|Agent|Notes mapping table.

claude "use @agent-team-configurator and optimize my project to best use the available subagents."

When creating Claude sub-agents, follow these location and naming conventions to ensure proper precedence and discovery:

• Project-level agents: Place in .claude/agents/ within your repository. These have the highest precedence and will override user-level agents with the same name.
• User-level agents: Place in ~/.claude/agents/ for global availability across all projects.
• Filename: Use kebab-case.md (e.g., code-reviewer.md). This should mirror the name field in the front-matter.
• Version Control: Commit project-level agents to your repository to allow for PR-style reviews.

After installation, verify that the agents are correctly recognized by the Claude Code CLI by running the /agents command. You should see a list of the 24 available specialized agents.

claude /agents

Some agents in this collection can use the Context7 MCP server to fetch up-to-date documentation more efficiently than the default WebFetch fallback. Installing Context7 provides faster retrieval, better structured responses, and cached documentation access.

If you do not install it, agents will automatically fall back to using WebFetch to retrieve documentation from official websites.

claude mcp add context7 -- npx -y @upstash/context7-mcp

Every agent file must begin with a YAML front-matter block to define its identity, trigger conditions, and capabilities.

• name: A unique, intent-revealing identifier using lowercase and hyphens only.
• description: This is used by Claude's router. Write when and why the agent should run. To encourage auto-delegation, include phrases like "MUST BE USED" or "use PROACTIVELY".
• tools: A comma-separated list of allowed tools (e.g., LS, Read, Grep). To inherit all available tools, omit this field entirely.

Note: Do not mix behavioral instructions (meant for the agent) into the description block (meant for the router).

---
name: unique-agent-name          # Lowercase & hyphens only
description: MUST BE USED …      # Natural‑language trigger phrase
tools: LS, Read, Grep            # Omit to inherit every tool
---

After installing the Context7 MCP server, ensure it is correctly configured in your Claude MCP settings file. After updating the configuration, you must restart Claude Code to activate the server.

{
  "mcpServers": {
    "context7": {
      "command": "context7-mcp",
      "args": []
    }
  }
}

Once your project is configured, you can start building complex features by using the @agent-tech-lead-orchestrator. This agent coordinates multi-step development tasks and utilizes other specialized agents to complete the work.

claude "use @agent-tech-lead-orchestrator and build a user authentication system"

Use this template to quickly scaffold a new, high-impact sub-agent. Fill in the placeholders and iterate based on performance.

---
name: <agent-name>
description: MUST BE USED to <do X> whenever <condition>. Use PROACTIVELY before <event>.
tools: <tool1>, <tool2>
---

# <Title> – <Concise Role Tagline>

## Mission
One sentence.

## Workflow
1. …
2. …

## Output Format
```markdown

## Section
- field: value

Heuristics

• Bullet
• Bullet

Control agent capabilities by managing the tools field in the front-matter based on the required security and scope: