agent.yaml is a declarative definition for any AI agent -- model, tools, guardrails, governance -- validated against a schema before it ever runs.
name: code-reviewer
version: 1.0.0
model:
provider: anthropic
name: claude-sonnet-4-20250514
fallback: gpt-4o
tools:
- type: mcp
server: github
capabilities: [read_pull_request, post_review]
guardrails:
max_tokens: 4096
require_approval: false
allowed_actions: [comment, suggest]
governance:
owner: platform-team
review_required: true
audit_log: enabledDeclare your model with automatic fallbacks. Switch providers without rewriting.
First-class MCP support. Connect to any tool server with declared capabilities.
Behavioral rules enforced at the definition layer. Not an afterthought.
Ownership, review gates, and audit trails built into every agent definition.
The spec, the CLI, and the hub work as one — so your team moves faster without breaking things.
One YAML file. Validated, versioned, portable.
Init, validate, run, import. Nothing else.
Pull curated agents. Push your own.
Curated agent definitions you can pull and run. Built to production standards.
Automated PR reviews with inline suggestions and security checks.
$ automagent pull code-reviewer Multi-source research with citations, summaries, and fact-checking.
$ automagent pull research-assistant Long-form content generation with tone control and brand guardrails.
$ automagent pull content-writer SQL generation, data visualization, and insight extraction from any source.
$ automagent pull data-analyst Install the CLI. Run init. You're building on an open standard.
npm install -g automagent