Claude Code v2.1.319 unblocks unattended agent runs

📅 November 6, 2026 • GitHub Claude Code v2.1.319 targets unattended agent runs Anthropic released Claude Code v2.1.319, and per the release notes the update concentrates on one operational problem - keeping long agent sessions moving when nobody is watching the terminal. The changes cover background execution for MCP tasks, a new permissions setting, a latency command for API code and a gateway setting for auto mode. MCP tool calls stop blocking the session MCP tool calls that run as tasks are now sent to the background immediately, and their result is reported on completion. A running Claude Code session no longer waits on them - the agent continues with the rest of its plan while a slow database query, a long-running build check or any other task-type MCP call finishes on its own. The result arrives as a completion report instead of a pause. denyInsteadOfPrompting - approvals become policy The new permissions setting denyInsteadOfPrompting changes what happens at the moment a tool call would normally stop and wait for user confirmation. The call is denied instead, and the run continues - in the terminal, in subagents and in teammates. For scheduled or headless runs this replaces an indefinite stall with a defined outcome, and the session keeps producing until it reaches a natural stopping point. Latency optimization as a command A new /claude-api latency-optimize command finds places where your Claude API code causes waits and selects faster settings, verified against evals. Instead of manually profiling requests and guessing at parameter changes, developers get a command that proposes concrete adjustments and validates them against a measurable baseline. Auto mode behind LLM gateways Teams that route Claude Code through LLM gateways get a behavior change as well - CLAUDE_CODE_AUTO_MODE_SERVER now enables server-side checking even when CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is set. Infrastructure that deliberately keeps experimental betas off retains a working auto mode path, which removes a configuration conflict for gateway operators. IMPACT The affected audience is engineers who run Claude Code in CI pipelines, cron jobs, long autonomous coding sessions and multi-agent setups with subagents and teammates. Daily work changes in concrete ways. Unattended runs no longer freeze on an approval prompt - with denyInsteadOfPrompting the policy decides, and the run proceeds. Slow MCP tool calls no longer serialize a session - background execution keeps the plan moving while tasks complete. And latency tuning stops being a manual review exercise - the /claude-api latency-optimize command turns it into a repeatable step with eval-verified results. Gateway operators should check that auto mode behavior still matches their expectations after the update. Sources: github.com Continue the conversation: https://t.me/AIAutomationArchitecture · Bluesky