agent-creator

Files (7)

• README-en.md
• SKILL.md
• aws-specialist-agent.json · examples/
• code-review-agent.json · examples/
• mobile-app-agent.json · examples/
• schema.md · references/
• agent.json.template · templates/

SKILL.md (first lines)

# Agent Creator

Skill for creating valid Kiro agent JSON files (`.kiro/agents/<name>.json`).
Covers all configurable fields: tools, allowedTools, toolsSettings, resources, mcpServers, hooks, toolAliases, welcomeMessage, and model.

## Output

A single file: `.kiro/agents/<name>.json` in the user's project.

```
<project-root>/
└── .kiro/
    └── agents/
        └── <name>.json
```

## Procedure

### 1. Gather intent

Ask (or infer from context):
- What is the agent's specialty/domain?
- What tasks will it perform?
- What tools does it need? (file system, shell, AWS, Docker, databases, etc.)
- Does it need MCP servers?
- Does it need to run commands on startup (hooks)?
- Which tools require user confirmation vs. auto-allow?

Skip questions already answered by the user's request.

### 2. Draft name and description

- `name`: kebab-case, lowercase, ≤64 chars. Matches the filename without `.json`.
- `description`: one sentence, imperative voice. What the agent does + its domain.
- `prompt`: system-level instruction. Include: role, behavior constraints, key responsibilities, any safety rules (e.g., "always verify before executing").

### 3. Select tools

Choose from available Kiro tools:

| Tool | Purpose |
|------|---------|
| `fs_read` / `read` | Read files |
| `fs_write` / `write` | Write/create files |
| `execute_bash` / `shell` | Run shell commands |
| `grep` | Search file contents |
| `glob` | Match file patterns |
| `use_aws` / `aws` | AWS SDK calls |
| `@<server>` | All tools from an MCP server |
| `@<server>/<tool>` | Specific MCP tool |

For MCP servers: list them in `mcpServers` and reference as `@<server-name>` in `tools`.

### 4. Configure allowedTools

`allowedTools` = subset of `tools` that auto-execute without user confirmation.

Rules of thumb:
- Read-only tools → safe to auto-allow (`fs_read`, `grep`, `glob`, read-only MCP tools)
- Write/destructive tools → require confirmation (exclude from `allowedTools`)
- For AWS: set `"autoAllowReadonly": true` in `toolsSettings.use_aws` instead of listing each read-only API

### 5. Configure toolsSettings

Per-tool constraints. Common patterns:

```json
"toolsSettings": {
  "use_aws": { "autoAllowReadonly": true },
  "execute_bash": { "autoAllowReadonly": true },
  "write": { "allowedPaths": ["src/**", "*.yaml"] },
  "shell": { "allowedCommands": ["npm test", "git diff", "docker-compose ps"] }
}
```

### 6. Add resources

Files/URIs always injected into context when the agent runs.

| Pattern | Example |
|---------|---------|
| Local file | `"file://README.md"` |
| Glob | `"file://docs/**/*.md"` |
| Skill | `"skill://.kiro/skills/**/SKILL.md"` |
| Remote URL | `"https://docs.example.com/api"` |

Keep list focused: only files the agent needs for every interaction.

### 7. Configure mcpServers (optional)

Two server types:

**stdio (command-based):**
```json
"mcpServers": {
  "my-server": {
    "command": "my-mcp-server",
    "args": ["--flag", "value"],
    "env": { "SECRET": "$ENV_VAR" },
    "timeout": 30000
  }
}
```

**HTTP:**
```json
"mcpServers": {
  "my-server": {
    "url": "https://api.example.com/mcp",
    "type": "http"
  }
}
```

### 8. Configure hooks (optional)

Commands run automatically. Two hook points:

| Hook | When it runs |

From salomao-santos/kiro-bridge-tools, Apache-2.0.

More from salomao-santos/kiro-bridge-tools