# Build reliable Claude Code agents

Plugin agents can trigger poorly or receive excessive tool access. This skill provides templates and guidance for focused agents with clear behavior.

## Install

```bash
npx skillstore add anthropics/agent-development
```

## Metadata

- Status: approved
- Slug: agent-development
- Version: 0.1.0
- Author version: 0.1.0
- Skillstore revision: r2
- Version status: valid
- Tree hash: 497a0b25b113dffd131c29c9e9f69fe0d76def5e01f30bf01f41844ceadb464c
- Author: anthropics
- GitHub username: anthropics
- License: MIT
- Repository: https://github.com/anthropics/claude-code/tree/main/plugins/plugin-dev/skills/agent-development
- Ref: 80999bf530a7874d7bedf8ce202001ecb4c4f5e0
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: filesystem, external\_commands
- Quality score: 84
- Quality tier: gold
- Public page: https://skillstore.pages.dev/skills/agent-development
- Manifest: https://skillstore.pages.dev/api/skills/agent-development/manifest

## Capabilities

- Explains the Markdown and YAML structure for plugin agent files.
- Defines names, descriptions, models, colors, and optional tool restrictions.
- Provides triggering examples for agent descriptions.
- Offers system prompt structure and quality guidance.
- Shows manual and AI-assisted agent creation workflows.
- Includes a shell validator for agent file structure.

## Use Cases

- Create a first plugin agent: Build a focused agent file with required frontmatter and a structured prompt.
- Improve agent triggering: Add concrete descriptions and examples so the correct agent is selected for user requests.
- Review agent permissions: Choose the smallest practical tool set for analysis, generation, or testing work.

## Prompt Templates

### Create a simple review agent

```
Help me create a Claude Code agent that reviews TypeScript files. Include a name, triggering description, focused tools, and a short system prompt.
```

### Add triggering examples

```
Improve this agent description with three concrete triggering examples. Explain when the agent should not run: [paste description].
```

### Apply least privilege

```
Recommend the minimum Claude Code tools for an agent that analyzes code and reports findings without editing files. Explain each choice.
```

### Design a specialized agent

```
Design a plugin agent for [domain]. Define boundaries, a repeatable process, edge cases, output expectations, and safe tool permissions.
```

## Limitations

- It provides guidance and templates, not a running agent implementation.
- It does not test agent behavior in every Claude Code version.
- The validator checks basic structure and does not perform a full security review.
- Tool availability depends on the host environment and plugin configuration.

## Best Practices

- Use specific triggering conditions and several realistic examples.
- Grant only the tools required for the agent task.
- Define responsibilities, process steps, edge cases, and output expectations.

## Anti Patterns

- Do not use vague descriptions that overlap with unrelated agents.
- Do not grant full tool access when read-only tools are sufficient.
- Do not omit testing for triggering behavior and edge cases.

## Security Audit

- Audited at: 2026-07-18T10:00:52.369\+00:00
- Summary: All 82 static findings are false positives from Markdown backticks, instructional prose, or the local validator shell syntax. The validator parses a supplied file with quoted values and does not evaluate its contents. No prompt injection, data exfiltration, or malicious intent was found in the reviewed materials.

## Stats

- Views: 544
- Downloads: 32
- Favorites: 0
- Popularity score: 0