Claude Code 2.1.302 Starts Headless Runs on Third-Party Providers in Default Permission Mode

Our take: never leave the permission mode of an unattended job to a default. Claude Code 2.1.302 changes that default for headless and Agent SDK runs on third-party providers. Set the mode explicitly in every script now and the next default change will not touch you. Claude Code stays Conditional in our directory: this is a configuration change to absorb, not a reason to rethink the tool.

What happened

Anthropic shipped Claude Code 2.1.302 on 16 October 2026. Two changes in the release alter how scripted runs behave.

The permission default

From the changelog: "Changed claude -p and Agent SDK sessions with no permission mode to start in default mode on third-party providers, with telemetry off or on a fresh install; --permission-mode auto keeps auto".

Fork subagents everywhere

"Changed fork subagents (subagent_type: \"fork\") to be 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 same release makes the desktop app and VS Code always run the main conversation's subagents in the background, as the terminal does. A subagent's own Agent calls in claude -p, the Agent SDK, the desktop app and VS Code now run in the foreground by default.

Also in 2.1.302

• Prompt and agent hooks: "at most 12 of one event run at a time; the rest wait their turn".
• CLAUDE_CODE_AUTO_MODE_MODEL chooses the model for the client-side auto mode classifier. It does not change the model for server-side auto mode.
• An Idle compaction row in /config turns idle compaction off or back on.
• /design-sync and /design-login now warn that standalone Claude Design (claude.ai/design) "closes by December 14, 2026".

Why it matters

The permission change hits one specific population: jobs on a third-party provider that never set a mode and run with telemetry off or from a fresh install. Anthropic's permission modes docs explain why: before 2.1.302, runs that did not fetch feature flags (on a third-party provider, or with DISABLE_TELEMETRY) could start in auto mode. Now they start in Manual mode, which appears as default in SDK options and settings. A headless run in that mode does not prompt: calls needing approval are denied unless an allow rule, hook, canUseTool callback or --permission-prompt-tool approves them. An unattended pipeline that leaned on the old auto behavior can fail without any change to its own code.

Fork subagents turning on by default in claude -p and the Agent SDK means a headless run can now spawn forks it did not spawn by default before. If your job budgets tokens or time tightly, that is a behavior change worth checking.

What changes for you

• Headless or SDK on a third-party provider: pass --permission-mode in every job (--permission-mode auto keeps auto), or approve specific calls with allow rules such as --allowedTools "Bash(npm test)" (the SDK has an equivalent allowed-tools option).
• You do not want forks outside the terminal: set CLAUDE_CODE_FORK_SUBAGENT=0.
• Heavy hook setups: expect prompt and agent hooks beyond 12 per event to queue.
• Standalone Claude Design users: export what you keep there before 14 December 2026.

FAQ

Does this affect runs against Anthropic's own API?

The 2.1.302 changelog scopes the change to third-party providers, with telemetry off or on a fresh install. Anthropic's permission modes docs say claude -p and Agent SDK runs with no mode set start in default on 2.1.302 or later, so set the mode explicitly wherever you run.

How do I keep auto mode in headless jobs?

Pass --permission-mode auto explicitly; the changelog states it keeps auto.

Can I turn fork subagents off?

Yes. CLAUDE_CODE_FORK_SUBAGENT=0 turns them off, and claude mcp serve never offers them.