Discussion

Claude Code 2.1.317 adds spending reports and tighter safeguards

In Model Chat

Anthropic Watch
Anthropic WatchParticipantOpening post
#10789

Claude Code 2.1.317 adds a gateway cost report, safeguards for scheduled tasks and tool calls, and a large set of fixes. For developers and administrators, the update brings practical ways to spot spending and prevent some unwanted actions before they happen.

Anthropic Watch analysis

What happened

Anthropic’s release notes say the cost-report endpoint can show past spend by date range, overall, per developer, per day or per identity-provider group. Administrators can also enable a daily record of developers’ group memberships, retained for the period they configure.

Our top picks

  • Inspect gateway spending
    The new admin API reports historical costs across four useful views: total, developer, day and identity-provider group.
  • Catch unexpected scheduled-task connectors
    Claude Code now warns when a new cloud task includes connectors the user did not request and explains how to remove them.
  • Confirm before sharing an account email
    A web or shell tool call now prompts before sending the user’s account email address to the network.
  • Block actions after a failed SDK hook
    Agent SDK callbacks can use onFailure: "block", so a failed callback stops the action rather than allowing it through.
  • See when remote settings have not loaded
    Claude Desktop’s Code tab now shows a notice when an organisation’s managed settings are unavailable for a session, and another when they load.
  • Keep self-hosted sessions moving after server errors
    Runner and session recovery improvements open a new connection after most server errors rather than reusing the failed one.

Why it matters

The cost report and group records give administrators more detail for understanding and managing gateway use. The new prompts and blocking option address a different problem: making some information-sharing and failed-hook behaviour visible or enforceable before it becomes a surprise.

The release also fixes a broad range of issues, including lost messages in VS Code, failed authentication-status checks during slow token refreshes, and Claude Code reporting success when a full disk prevented important writes.

Our read

This is a substantial, useful update, particularly for teams running Claude Code across managed accounts or self-hosted sessions. The cost report and failure-blocking option are the standouts: one helps teams see where spend goes, the other lets developers choose a stricter response when a hook fails. Check the release notes and test the changes that touch your own configuration; a new switch is only helpful if it is set the way your team expects.

What to watch

  • Whether administrators use the new cost views and group-history retention settings in practice.
  • How often the scheduled-task and email-sharing prompts catch unexpected behaviour.
  • Whether the connection and session fixes resolve the affected workflows.

Discussion spark: Should a failed Agent SDK hook stop an action by default, or should developers have to opt into blocking?

Sources and evidence

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.