Feature Gates Guide

Some Claude Code features exist in the software but are turned off by default. They require an environment variable, a Statsig flag (enabled by Anthropic on your account), or a specific plan tier to activate. These are called feature gates.

This guide covers the gates most likely to affect you as you explore Claude Code. You don’t need to understand all of them — most are for enterprise deployments or experimental research features.

Types of feature gates

Gate type	Example	How it’s activated	Can you set it?
Environment variable	`CLAUDE_CODE_ENABLE_TASKS=1`	Set it in your shell	Yes
Statsig flag	`tengu_session_memory`	Enabled by Anthropic on your account	No
Plan tier	Max plan required	Upgrade your subscription	Yes, via your account
Dual-gate	`/team-onboarding`	Requires both an env var AND a Statsig flag	Partially — you set the env var, Anthropic sets the flag

Feature gates relevant to junior developers

Persistent tasks

CLAUDE_CODE_ENABLE_TASKS=1

Enables a task coordination system that persists across sessions.

Agent teams

CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Enables TeamCreate/TeamDelete tools for coordinating groups of agents.

Coordinator mode

CLAUDE_CODE_COORDINATOR_MODE=1

Enables full multi-agent orchestration mode.

Session memory

tengu_session_memory  (Statsig — Anthropic enables this, not you)

A background subagent extracts key notes from your session into ~/.claude/sessionMemory.md.

Scratchpad

tengu_scratch  (Statsig — Anthropic enables this, not you)

Enables a .claude/scratchpad/ shared directory between agents.

Team onboarding

CLAUDE_CODE_TEAM_ONBOARDING=1
# AND tengu_flint_harbor Statsig flag (Anthropic enables)

Enables the /team-onboarding command.

Loop skill

CLAUDE_CODE_ENABLE_LOOP=1

Enables the /loop command for recurring scheduled prompts.

CFC

CLAUDE_CODE_ENABLE_CFC=1

Purpose unknown — this env var exists in the source but its command surface is undocumented.

How to set environment variables

You can set these in your shell session, or add them to a .env file if your setup supports it:

# In your shell (lasts for this terminal session)
export CLAUDE_CODE_ENABLE_TASKS=1
claude

# Or when starting Claude directly
CLAUDE_CODE_ENABLE_TASKS=1 claude

For persistent configuration, add to your shell profile (~/.bashrc, ~/.zshrc):

export CLAUDE_CODE_ENABLE_TASKS=1

Feature disable flags

Some features are on by default and can be turned off:

Flag	What it disables
`CLAUDE_CODE_DISABLE_AUTO_MEMORY=1`	Turns off automatic memory extraction (privacy)
`CLAUDE_CODE_DISABLE_TELEMETRY=1`	Disables usage telemetry
`DISABLE_<COMMAND>_COMMAND=1`	Disables a specific slash command (e.g., `DISABLE_FEEDBACK_COMMAND=1`)

Before enabling experimental features

Read the discrepancies doc first — Commands/discrepancies-and-gaps.md documents known gaps between env-var-gated features and their public docs. Item 10 (the /tasks system) is especially important.
Experimental features can change behavior significantly — CLAUDE_CODE_COORDINATOR_MODE doesn’t just add new tools; it changes how Claude orchestrates work entirely. Don’t enable it during a critical session without understanding what it does.
Some gates require multiple conditions — the /team-onboarding command requires both an env var and a Statsig flag. If only one is set, the command won’t appear or won’t work.
Statsig flags are not documented in advance — tengu_* flags may appear in docs or source before they’re rolled out to your account. If a feature sounds like it should work but doesn’t, the Statsig flag may not be enabled for you.