{"version":"2.1.321","anchor":"subagent-output-warning-now-depends-on-claude-code-user-dele","canonical_anchor":"subagent-output-warning-now-depends-on-claude-code-user-dele","heading":"Unreviewed subagent output now comes with a security warning","tier":"notice","area":"Agents","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.321\/e\/subagent-output-warning-now-depends-on-claude-code-user-dele","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.321","markdown":"### Unreviewed subagent output now comes with a security warning\n\nClaude Code now warns when a subagent's output was not reviewed, with wording that depends on CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API\n\n**Unclear.** What else setting CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API changes beyond this warning is not known.\n\n**What**\n\nA subagent is a helper agent that Claude starts to carry out part of a task. Claude Code now builds a \"SECURITY WARNING\" saying that a subagent's output is unreviewed. Its wording depends on the environment variable `CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API`, a setting you can put in your shell before starting Claude Code:\n\n- With the variable set, the warning says the safety check gave no verdict on the output.\n\n- Without it, the warning says Claude Code no longer reviews a subagent's output itself.\n\n**Why**\n\nThe warning marks subagent output that has not been checked, so it can be treated with more care. The variable is referred to in many places in this build, which suggests it switches on a distinct way of running a session.\n\n- Area: Agents\n- Names: `CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API`\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 3\/5\n- Scope: individual\n- Heads-up: no"}