custom-agents

You can configure custom AI agents by creating JSON files in local or global directories, specifying fields like name, description, tools, allowedTools, and model to centralize tool access control and context inclusion for your project workflows.

Custom AI agents are used to automate specialized project workflows like infrastructure management, development processes, and code reviews. They reduce prompt leakage and setup time by provisioning defined resources, hooks, and MCP integrations for your team.

Yes, you can use MCP integrations and hooks when creating custom agents. The configuration supports mcpServers for integrations and hooks to automate lifecycle events, alongside fields for prompts, toolAliases, and resource provisioning.

The best way to set up an infrastructure specialist agent is using the interactive wizard or CLI to generate a JSON configuration file. You define its specific tools, prompts, and MCP settings tailored to your infrastructure project requirements.

Custom agent configuration files need to be stored locally in the .swarm-ai/agents directory or globally in ~/.swarm-ai/agents. This allows teams to centralize configurations for reuse across different projects and workflows.

Setting up code-review agents does not require manual JSON editing. You can use the interactive wizard or CLI to create and tailor the agent, defining its tools, prompts, and MCP settings without manually writing the configuration file.