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_FORK_SUBAGENT and --permission-mode auto for v2.1.302 Summary: v2.1.302 turns fork subagents on by default in claude -p, the Agent SDK, desktop, and VS Code, starts third-party sessions with no permission mode in default, adds CLAUDE_CODE_AUTO_MODE_MODEL for the client-side auto classifier, and adds an Idle compaction row in /config. Primary source: https://github.com/anthropics/claude-code/releases/tag/v2.1.302

Why it matters

Claude Code v2.1.302 changes defaults that apply outside the terminal unless you pin them. Fork subagents (subagent_type: "fork") are now on by default in claude -p, the Agent SDK, the desktop app, and VS Code, as they already are in the terminal; CLAUDE_CODE_FORK_SUBAGENT=0 turns them off, and claude mcp serve never offers them. The desktop app and VS Code always run the main conversation's subagents in the background; a subagent's own Agent calls in claude -p, the Agent SDK, desktop, and VS Code now run in the foreground by default. claude -p and Agent SDK sessions with no permission mode start in default on third-party providers (telemetry off or a fresh install); --permission-mode auto keeps auto. Teams that still want auto permissions or the previous subagent layout will get different behavior until those settings are explicit in env, CI, and agent instructions. Separately, CLAUDE_CODE_AUTO_MODE_MODEL chooses only the client-side auto mode classifier model (not server-side auto mode), and /config now has an Idle compaction row to turn idle compaction off or back on; notices after an idle compaction say how to turn it off. Server-side auto mode now compactes a conversation that is too long for the safety check instead of denying every checked tool call.

Suggested actions

1. Upgrade to Claude Code v2.1.302, then in each Agent SDK, desktop, VS Code, and claude -p environment either leave fork subagents on or set CLAUDE_CODE_FORK_SUBAGENT=0; do not expect claude mcp serve to offer subagent_type: "fork".
2. For claude -p and Agent SDK sessions on third-party providers that still need auto permissions, always pass --permission-mode auto; sessions with no permission mode now start in default (telemetry off or a fresh install).
3. export CLAUDE_CODE_AUTO_MODE_MODEL=<client-side-classifier-model> so the client-side auto mode classifier uses that model; this does not change the model for server-side auto mode.
4. In Claude Code, run /config and set the Idle compaction row to off or back on for this project; after an idle compaction, follow the notice that now explains how to turn it off.
5. In the desktop app and VS Code, treat the main conversation's subagents as always backgrounded; a subagent's own Agent calls now run in the foreground by default, matching claude -p and the Agent SDK.

Config surfaces this release may change

• skills (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
• subagent definitions (high confidence) — check your repo before applying
• context and compaction settings — 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-v2-1-302-set-claude-code-fork-subagent-and-permission-mode-a and mark Applied, Skipped, or Failed. Proposal id: 35962cde-6f10-4812-b9b9-87c03b9811b4

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.