Anthropic ships ant apply, letting teams manage Claude Platform resources as code

Anthropic shipped version 1.30.0 of its ant command-line tool on September 3, 2026, adding a new ant apply command that manages Claude Platform resources — agents, environments, skills, memory stores, and deployments — as version-controlled files in a repository, rather than through manual console changes.

What's new

The workflow mirrors established infrastructure-as-code tooling: a developer describes each resource — an agent definition, an environment, a skill, a memory store, or a deployment — in a file, runs ant apply, and reviews a printed execution plan before approving it. The command then creates or updates the underlying Claude Platform resources to match.

A key piece of the release is the claude-lock.json lockfile that ant apply writes after each run. Anthropic's release notes state the command "creates and updates agents, environments, skills, memory stores, and deployments from files in your repository." Committing that lockfile ensures later runs — whether on a developer's machine or in a CI pipeline — update the same set of resources instead of accidentally creating duplicates, the standard failure mode for state-drift in declarative tooling.

Anthropic published an accompanying guide, "Manage resources as code with ant apply," walking through the file format and plan-review flow in more depth.

Context

The ant CLI is Anthropic's command-line interface for the Claude Platform, sitting alongside the Claude API, the Agent SDK, and Claude Code as part of Anthropic's expanding developer-tooling surface. As organizations wire more agents, skills, and persistent memory stores into production Claude deployments, managing those pieces by hand through a web console becomes error-prone and hard to audit — the same problem that drove infrastructure teams toward tools like Terraform and Pulumi over manual cloud-console clicking a decade earlier.

ant apply is Anthropic's answer for the Claude Platform specifically: instead of a general-purpose infra tool bolted on afterward, resource definitions live as files next to the rest of an application's code, and changes flow through the same repository, review, and CI process as any other code change.

Why it matters

For teams running multiple Claude-based agents across several environments — staging, production, per-customer deployments — configuration drift is a real operational risk: an agent tuned by hand in one environment silently diverging from what's declared elsewhere. A lockfile-backed apply model closes that gap by making the repository the single source of truth and making changes reviewable before they land.

It also signals a maturing enterprise posture for the Claude Platform. Treating agents and skills as code that can be diffed, reviewed, and rolled back is a prerequisite for the kind of governance and audit trails larger enterprise customers expect before putting agentic systems into regulated or high-stakes workflows.