DevAgentRadar → Claude Code

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

Context

Assistant: Claude Code Proposal: Set CLAUDE_CODE_AUTO_MODE_SERVER after v2.1.319 removes classifier fallback Summary: v2.1.319 asks about or denies auto-mode calls the server cannot check (telemetry/flags off or LLM gateway) and no longer falls back to the built-in classifier. Set CLAUDE_CODE_AUTO_MODE_SERVER=0 to keep the old path, or =1 behind a gateway even with experimental betas disabled. Bedrock, Vertex, and Foundry are unchanged. Primary source: https://github.com/anthropics/claude-code/releases/tag/v2.1.319

Why it matters

Claude Code 2.1.319 stops falling back to the built-in auto-mode classifier when the server cannot check a call because telemetry or feature flags are off, including behind LLM gateways. Those calls are now asked about or denied. Behind a gateway, CLAUDE_CODE_AUTO_MODE_SERVER=1 turns the server-side check on even if CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is set; managed settings that set the latter must set 1 too. Teams that still want the classifier should set CLAUDE_CODE_AUTO_MODE_SERVER=0 (Bedrock, Vertex, and Foundry are unchanged). Pair the upgrade with permissions.denyInsteadOfPrompting so unattended terminal, subagent, and teammate turns deny instead of blocking on a permission prompt. MCP servers that run a tool call as a task are moved to the background immediately unless CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND is set. CI that used claude -p in auto mode must handle exit 1 when actions are refused because no safety check was available. After rollout, /status no longer reports Auto mode server Enabled when the review never reached the session.

Suggested actions

1. In managed settings and every developer shell behind an LLM gateway, set CLAUDE_CODE_AUTO_MODE_SERVER=1 whenever CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is already set, so v2.1.319 still runs the server-side auto-mode check.
2. If you need the pre-2.1.319 built-in classifier when telemetry or feature flags are off, export CLAUDE_CODE_AUTO_MODE_SERVER=0. Skip this on Bedrock, Vertex, and Foundry (unchanged).
3. Add "permissions.denyInsteadOfPrompting": true to Claude Code settings so a tool call that would wait on a permission prompt is denied and the turn continues in terminal sessions, subagents, and teammates.
4. Update CI that runs claude -p in auto mode: v2.1.319 now exits 1 for both text and JSON output when actions are refused because no safety check was available (previously exited 0). Fail the job on non-zero.
5. Leave MCP task tool calls on the new default (Claude Code backgrounds them immediately and tells Claude the result when they finish), or export CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND=1 to wait inside the tool call. After upgrade, run /status and confirm Auto mode server matches whether the server-side review actually reaches the session.

Config surfaces this release may change

• hooks (high confidence) — check your repo before applying
• MCP servers (high confidence) — check your repo before applying
• permission rules (high confidence) — check your repo before applying
• background and headless runs — check your repo before applying
• settings files — check your repo before applying
• skills — check your repo before applying
• subagent definitions — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/claude-code-v2-1-319-set-claude-code-auto-mode-server-after-v2-1-319-rem and mark Applied, Skipped, or Failed. Proposal id: 97579178-ce45-4518-9658-af5b7aa75a83

Your job

1. Restate the change in one sentence.
2. Propose a minimal plan for my repo (or a throwaway pilot).
3. Implement only what I approve; prefer small diffs and tests.
4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.