builders · review · first seen Nov 3, 10:32 UTC

Anthropic adds connector and email warnings to Claude Code

Claude Code v2.1.317 shipped warnings for unrequested connectors on new cloud scheduled tasks and for tool calls that would send an account email address to the network. Builders should care because both checks surface a decision at the point where a workflow could expose access or account information.

L0 · primary1 source4 claims checked4 min read

What shipped

The /schedule flow now warns when a new cloud scheduled task gets connectors the user did not ask for. The warning includes how to remove those connectors. This is a warning about the connectors assigned to a new task, not a claim that the task is automatically corrected or that every connector assignment is blocked. For a builder creating scheduled work, the useful moment is before accepting a task whose connector access differs from what was requested. [1]

Claude Code now asks before a web or shell tool call sends the user’s account email address to the network. The condition is specific: it concerns the account email address and an outbound web or shell call. The release note does not describe a general review of all information leaving those tools. Treat the prompt as a chance to inspect that particular disclosure, not as evidence that other outbound data has been checked. [1]

The same release adds onFailure: "block" for Agent SDK hook callbacks. With that option, a failed callback blocks the action rather than letting it run. This is a developer control for the failure path of a hook callback. Separately, the Claude apps gateway admin API adds GET /v1/organizations/analytics/cost for past spend over a date range, with views in total, per developer, per day or per IdP group. These additions serve different workflows from the two Claude Code warnings. [1][2]

Why it matters if you build

A cloud scheduled task can run after its creator has moved on to other work. A warning at creation gives the creator a reason to compare the task’s connectors with the request and remove ones they did not intend to include. The release note says the warning explains how to remove them; it does not say removal happens on its own. The practical change is a more visible review point, with the decision still left to the person setting up the task. [1]

An account email address can be included in a tool call that reaches the network. The new check puts a question before the described web or shell call sends it. A builder who expects that disclosure can decide whether to proceed; one who does not can reconsider the call. For Agent SDK users, onFailure: "block" addresses a separate risk: a callback that fails need not leave its associated action free to run. Neither release-note line establishes coverage beyond the condition it names. [1][2]

The cost endpoint gives gateway administrators a way to read past spend for a selected date range and examine it at the listed levels. That can inform a review of total spend or its distribution by developer, day or IdP group. The excerpt also says the gateway records each developer’s IdP groups daily when admin is set, retaining those records for admin.identity_retention_days, which the cost report uses for grouping. No pricing, budget control or prospective cost estimate is established here. [1]

How to try it today

1. In Claude Code v2.1.317, create a cloud scheduled task with /schedule and check whether its connectors match what you asked for; use the warning’s removal guidance if they do not.
2. Before approving a web or shell call that would send your account email address to the network, read the prompt and decide whether that disclosure is intended.
3. If you use Agent SDK hook callbacks, assess whether a callback failure should block its action and use onFailure: "block" where that is your intended behavior.
4. If you administer a Claude apps gateway, request past spend for a date range through GET /v1/organizations/analytics/cost and review the available groupings.

What we checked

Every claim below was walked down the source ladder. The rung is how close the source is to the thing itself; the tense is the source's own.

1. shipped L0 · primary

   Claude Code v2.1.317 adds a warning when a new cloud scheduled task gets connectors the user did not ask for.

   per GitHub releases · anthropics/claude-code (L0)
2. shipped L0 · primary

   Claude Code v2.1.317 asks before a web or shell tool call sends the user's account email address to the network.

   per GitHub releases · anthropics/claude-code (L0)
3. shipped L0 · primary

   Agent SDK hook callbacks add onFailure: "block", which blocks an action if the callback fails.

   per GitHub releases · anthropics/claude-code (L0)
4. shipped L0 · primary

   The Claude apps gateway admin API adds GET /v1/organizations/analytics/cost for reporting past spend over a date range.

   per GitHub releases · anthropics/claude-code (L0)

What is not confirmed

The release note does not establish that the connector warning blocks task creation, removes connectors automatically or covers changes to existing scheduled tasks. It says the warning applies when a new cloud scheduled task gets connectors the user did not ask for. [1]

The email check is not described as a general data-loss control. The source does not establish how it detects the account email address, whether it covers other addresses or whether it applies to tools other than web and shell. [1]

The excerpt does not specify a default failure behavior for every Agent SDK hook callback or the cost endpoint’s access requirements, response format and reporting delay. Those details need confirmation before a builder relies on them in a policy or reporting integration. [1][2]

Who should act, who should wait

Move now if you create cloud scheduled tasks, review outbound web or shell calls, maintain Agent SDK hooks, or administer gateway spend reporting: each has a concrete shipped behavior to inspect. Wait before treating the warnings as automatic enforcement or building assumptions about the cost endpoint’s response and permissions into production reporting. The supplied release note supports the named checks and controls, but not those broader guarantees. [1][2][3][4]

Sources

• v2.1.317 — GitHub releases · anthropics/claude-code L0 · primary