Follow Discord
Sweep 10 Nov 2026 · 02:18Z Build v2.1.322 539 read Stable v2.1.311 Latest v2.1.322 Next v2.1.322 Feeds RSS JSON llms.txt llms-full.txt Unofficial
HomeReleasesv2.1.321entry #22

Claude Code v2.1.321 ·

Auto mode's safety checker behaves differently for user-delegated direct API sessions

When CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API is set, auto mode switches off a third-party checker role and always checks allowed MCP write tools

You'll notice No documentation found Improvements
JSON All of v2.1.321
You'll noticeTier: how much it should matter to you
3Useful: my rating, 1 to 5
4Signal: worth watching, 1 to 5
Auto ModeArea: what it touches
ImprovementsKind: in v2.1.321,
What probably matters to youSection of the release

Unclear What sets this variable, and what the fallback does when the classifier refuses, is not shown.

What

In auto mode, a classifier (an automatic checker) decides whether an action is safe instead of asking you. Claude Code now reads the environment variable CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API, meant for sessions using a user-delegated credential on a direct API. An environment variable is a named value set outside Claude Code before it starts. When this one is set:

  • The classifier uses its default model rather than choosing one.
  • The classifier's third-party role is switched off for the session.
  • The classifier uses its default severity thresholds, the levels at which it treats an action as risky.
  • A fallback runs when the classifier is unavailable or refuses to answer.
  • Allowed MCP tools that change things are always included in what the classifier checks. MCP tools come from separate servers that add abilities to Claude Code.

Without the variable, whether those MCP write tools are included is still decided by a server-side setting that is off by default in the code. In readings taken before this release, that setting came back on for this site's account and for an anonymous check.

Why

Sessions run with a user-delegated direct API credential get their own auto mode behaviour, including a set fallback when the classifier cannot give an answer.

Read from
Names in the bundleCLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API
Feature flag
tengu_cowork_auto_mode_include_allowed_write_mcp Off by default, switched on for this account

The shipped code defaults this off, and the flag server returned on for the one account this site reads on this version. That is the reading that makes the entry above worth a second look, and it still says nothing about your account.

This account: on · anonymous baseline: on · compiled default in v2.1.321: off

These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.

Read once, for one account on one subscription tier, against v2.1.321. It isn't a statement about your account. What a flag value here can and cannot tell you

How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtWhat sets this variable, and what the fallback does when the classifier refuses, is not shown.
The name it cites is new in this buildNew in this build: CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API

See this entry in the whole of v2.1.321 →

Feedback