New controls for the reply check: disableReplyCheck and CLAUDE_CODE_REPLY_GATE_POLICY

What

In Claude Tag sessions on a self-hosted runner, a reply check re-prompts Claude, at most four times per turn, when a turn ends without Claude calling its reply tool. A self-hosted runner is a machine you operate yourself that runs Claude Code sessions. A turn is one round of Claude working before it hands control back. This release adds two ways to control that check.

• disableReplyCheck is a new setting that turns the reply check off. It must be set in managed settings, the settings an administrator controls centrally, and it is ignored anywhere else. Turning it off also stops two messages the service asks to have sent to Claude. It joins the same group of settings as disableAllHooks.
• CLAUDE_CODE_REPLY_GATE_POLICY is a new environment variable, a named setting passed to a program when it starts. Claude Code reads it at launch and then removes it. It holds a policy covering things such as which tools end a turn, limits on asking, and a status nudge.
• When the policy is accepted, Claude Code logs that it was. It warns when part of the policy cannot be read, and gives an error when the value is too long or unreadable.
• In a terminal or a remote session, that warning tells the self-hosted runner to update.
• If the policy includes disableReplyCheck, Claude Code warns that it is ignored there, because only managed settings can turn the reply check off.

The settings help text also now says transcripts are kept 'on this computer': 'Days to keep transcripts on this computer before automatic cleanup'.

Why

Administrators of self-hosted runners now have a switch to stop the re-prompting. The new variable gives a way to adjust the reply check when a session starts, and it does nothing unless it is set. Turning the check off entirely is still limited to managed settings.