CLI Reference

agentcli is the developer CLI for the AWS Agent Platform. It covers the full agent development lifecycle: blueprint validation, prompt management, graph visualization, deployment, policy generation, evaluation, and artifact generation.

Installation

The CLI is published as the agent-cli package alongside agent-core. Install it from your configured package registry:

pip install agent-cli

For development (editable install from the repository source):

pip install -e "cli/[dev]"

Environment Variables

Variable	Purpose
`AGENT_REGISTRY_URL`	Prompt Registry HTTP API base URL (default: `http://localhost:8000`). Used when the registry is accessed over HTTP.
`AWS_DEFAULT_REGION`	AWS region for deployment operations (`agentcli deploy`)
`AWS_REGION`	AWS region for evaluation operations (`agentcli eval`)

Command Groups

Command	Description
`agentcli blueprint`	Lint and validate agent, strategy, and workflow blueprint YAML files
`agentcli prompt`	Push, get, list, diff, promote, and rollback versioned prompt templates
`agentcli graph`	Render ASCII topology diagrams from multi-agent graph blueprints
`agentcli deploy`	Build and deploy agents to Amazon Bedrock AgentCore Runtime
`agentcli policy`	Validate and generate Cedar policies from blueprint rules
`agentcli eval`	Run on-demand evaluation and check online evaluation status
`agentcli generate`	Generate deployment artifacts (runtime config, Dockerfile) from blueprints

Quick Reference

# Validate a blueprint before committing
agentcli blueprint lint agents/my-agent.yaml

# Push a new prompt version
agentcli prompt push prompts/system.jinja2 --id my-agent-system --version 1.2.0

# Deploy an agent to AgentCore Runtime
agentcli deploy agent agents/my-agent.yaml --env production

# Generate Cedar policies from blueprint rules
agentcli policy generate agents/my-agent.yaml \
  --gateway-arn "arn:aws:bedrock-agentcore:${AWS_REGION}:${AWS_ACCOUNT_ID}:gateway/my-gateway"

# Run evaluation on a session
agentcli eval run --agent-id my-agent --session-id sess-123 \
  --evaluators Builtin.Correctness,Builtin.GoalSuccessRate

# Dry-run a deploy to generate artifacts without deploying
agentcli deploy agent agents/my-agent.yaml --dry-run

Getting Help

Every command supports --help:

agentcli --help
agentcli blueprint --help
agentcli blueprint lint --help
agentcli prompt --help
agentcli eval run --help