What Anthropic's new usage policy means for Saggar users

Anthropic has published a revised usage policy with a stated effective date of November 12, 2026. For Saggar users, the main questions are how Claude is accessed and who is responsible when an agent acts on your behalf.

Saggar's built-in Claude Code integration launches the claude command installed on your Mac. Our source review found no reason to change that launch method in response to this policy. Your account's terms and the work you ask Claude to do still apply.

How Saggar connects to Claude

Saggar opens Claude Code in a local terminal, passes your task and launch options to it, and uses local hooks and session records to follow its progress. Claude Code handles its own connection and authentication. Saggar can ask it for sign-in status, but the reviewed integration does not extract Claude subscription tokens or implement a Claude inference proxy.

That distinction matters because the new policy expressly prohibits unauthorized resale or proxy access, including services that route requests through consumer subscriptions or misrepresent the client being used. Running the installed Claude Code CLI is the architecture we reviewed; this is not a claim of Anthropic approval or a ruling on every third-party configuration.

Anthropic's Claude Code guidance separately permits end users to sign in to an unmodified Claude Code binary with their own subscription, including on platforms that run it. Its conditions for offering Claude Code in a product include the applicable commercial agreement, preserving authentication options, and leaving usage and billing with each end user. The usage policy alone does not establish that every contractual condition has been met.

A custom startup command, shell wrapper, or provider configuration can change what runs. If you use one, check its access method separately. A Saggar subscription does not include Claude access or expand your Anthropic entitlements.

Automatic approval is delegated authority

The revised policy explicitly makes people who build or deploy agents responsible for actions through tools, browsers, and connected systems.

Saggar can answer some Claude Code permission requests automatically when you enable automatic approval. Its permission hook considers shell commands and file-editing tools, checks them against the configured risky-command rules, and leaves unsupported requests or plan-mode requests to the provider's normal handling. Saggar also supports answering recognized terminal dialogs.

These checks are not a complete assessment of an action's purpose, destination, or legal consequences. A command that passes a local rule can still do something you did not intend or are not authorized to do. Enable automatic approval only for work you are prepared to delegate, and review the permissions of connected tools as well as the prompt you give the agent.

See Claude Code in Saggar for the integration's behavior and supervising agents for the controls.

Which work needs extra care?

The policy clarifies authorized security testing and introduces more specific requirements for high-risk recommendations and physical actions. General research and some internal drafting are excluded from those high-risk requirements; delivering advice or making decisions about a particular person can be different.

If your agents advise people, affect employment or access to services, or control potentially dangerous equipment, read those sections before delegating execution. An automatic permission answer is not qualified human review, and a terminal control is not an independent equipment safety limit.

What to do before November 12

For the built-in Claude Code launch, this review does not identify a required Saggar configuration change. Check custom wrappers separately, keep your Claude access within the terms of your account, and choose automatic-approval settings that match the authority you intend to delegate.

We reviewed Saggar's source on October 14, 2026. This was a focused review of the built-in launch and approval paths, not a certification of every workflow or a live audit of individual accounts.

Read Anthropic's revised usage policy and the September 2025 version for the full wording. Claude Code's legal and compliance documentation explains which service terms apply.