Agents

InfraDots agents review and implement infrastructure changes for you. They work on your Terraform/OpenTofu code and your plans, and they follow the standards your team writes down.

What agents do

Review

The review agent reviews pull requests and commits together with the plan. It applies risk-based checklists by resource type, leaves line comments on the change, and can request changes. It requests changes when fmt or the run failed.

Implement

The implement agent writes a change and opens a pull request. A follow-up request on the same pull request adds to it instead of starting over.

Auto-implement

When the workspace allows it, the review agent can implement its own findings. It opens either a single pull request, or, with the "minimal blast radius" option, splits the fixes into several smaller pull requests.

Reviewer guidance

Reviewer Guidance is org-wide guidance that your team writes and approves. Agents read it before every review. It is not learned automatically.

Skills

Organizations can install skill packages from git, following the Agent Plugins standard. Skills give agents your playbooks and checklists.

Registry lookups

Agents look up the newest provider and module versions, and a resource's reference (arguments, attributes, examples), in the public Terraform and OpenTofu registries, asking the registry that matches the workspace (OpenTofu or Terraform) first. This lets agents pin current versions and use arguments that exist.

Where you trigger them

• Web app - from your workspaces and pull requests.
• Slack - ask the InfraDots Slack bot.
• CLI - idp agent list and idp agent history <job-id> (see idp-cli).
• Your IDE - the AI in Your IDE skill lets Claude Code or Cursor drive the CLI.

Where they run

Where	Model	Read more
InfraDots cloud	Managed by InfraDots	This page
Your Kubernetes cluster	Your own model account (Anthropic, Bedrock, Vertex, Foundry)	Self-hosted Runners
Your laptop, no account	Your own model account	Review & implement locally

Getting started

Agents are off by default: turn them on for a workspace in its settings (or --agents-enabled with idp workspace create, or agents_enabled in the Terraform provider). Then they review the workspace's runs as they finish, within your plan's agent interactions. Auto-implement is set for the organization, and each workspace can override it. See Reviewer Guidance to teach the reviewer your standards.