Opt-in vs Always-On Mode in i-have-ADHD: Complete Activation Guide

The i-have-ADHD skill supports opt-in activation via session-specific commands and always-on mode through flag files or extensions, differing primarily in persistence, automation, and the underlying trigger mechanisms.

The ayghri/i-have-adhd repository implements a specialized AI interaction style designed for ADHD workflows through two distinct activation methods. Understanding the difference between opt-in and always-on mode in i-have-adhd allows users to choose between temporary, manually-triggered assistance and persistent, automatic rule application across every AI session.

What Are the Two Activation Modes?

The skill operates through fundamentally different pathways:

• Opt-in (command) invocation: Manual activation per session using slash commands that load the skill from skills/i-have-adhd/
• Always-on (extension) mode: Automatic injection at session start via the hooks/always-on.sh script when it detects the flag file ~/.claude/.i-have-adhd-always

Both methods ultimately source their behavior from skills/i-have-adhd/SKILL.md, but differ significantly in timing, persistence, and setup requirements.

How Opt-In Mode Works

Opt-in mode treats the ADHD ruleset as a temporary skill that users explicitly request when needed.

Activating Opt-In Mode

To trigger the skill manually in Claude Code, type the command:

/i-have-adhd

For Codex users, the trigger uses a dollar prefix:

$i-have-adhd

Once activated, the rules remain active only for the current session.

Technical Implementation

The opt-in mechanism resides in skills/i-have-adhd/ and relies on front-matter configuration to prevent automatic loading. According to the source code, the skill declares disable-model-invocation: true in its front-matter, ensuring nothing executes until the user types the explicit trigger command.

Deactivating Opt-In Mode

To end the skill in an active session, type:

stop adhd mode

Alternatively, normal mode also terminates the ADHD-specific formatting for that session.

How Always-On Mode Works

Always-on mode automatically prepends the ADHD ruleset to every conversation without manual intervention.

Setting Up Always-On Mode

Users can enable automatic activation through two methods. The first involves creating a flag file in your Claude configuration directory:

touch ~/.claude/.i-have-adhd-always

The second method installs the skill as a Gemini extension:

gemini extensions install https://github.com/ayghri/i-have-adhd

The Session-Start Hook Mechanism

Under the hood, always-on mode relies on hooks/always-on.sh, a session-start hook registered in hooks/hooks.json. At the start of every Claude Code session, this script checks for the presence of $CLAUDE_CONFIG_DIR/.i-have-adhd-always (defaulting to ~/.claude/.i-have-adhd-always).

If the flag file exists, the hook reads skills/i-have-adhd/SKILL.md, strips its YAML front-matter, and prints the full ruleset before every assistant reply. This ensures the ADHD formatting guidelines are automatically active without user intervention.

Disabling Always-On Mode

To disable the automatic behavior, remove the flag file:

rm ~/.claude/.i-have-adhd-always

For extension-based installations, uninstall via:

gemini extensions uninstall i-have-adhd

Alternatively, editing hooks/hooks.json to remove the "command": "sh \"${CLAUDE_PLUGIN_ROOT}/hooks/always-on.sh\"" entry prevents automatic injection while preserving the skill files for opt-in use.

Key Differences at a Glance

Activation Method: Opt-in requires typing /i-have-adhd or $i-have-adhd, while always-on checks for ~/.claude/.i-have-adhd-always or uses a Gemini extension.

Session Persistence: Opt-in lasts only for the current conversation, whereas always-on applies to every new session automatically via hooks/always-on.sh.

Deactivation: Type stop adhd mode to end opt-in sessions; remove the flag file or uninstall the extension to disable always-on.

Implementation Detail: Opt-in uses disable-model-invocation: true in the skill front-matter to prevent auto-loading; always-on relies on hooks/hooks.json to execute hooks/always-on.sh at session start.

Summary

• Opt-in mode requires explicit command invocation per session and uses disable-model-invocation: true to remain dormant until triggered, storing the skill in skills/i-have-adhd/.
• Always-on mode leverages hooks/always-on.sh and hooks/hooks.json to automatically inject rules from skills/i-have-adhd/SKILL.md when ~/.claude/.i-have-adhd-always exists.
• Both modes access the identical ruleset defined in skills/i-have-adhd/SKILL.md, differing only in automation level and session persistence.
• Users can switch between modes by creating or deleting the flag file, or by toggling the hook registration in the configuration.

Frequently Asked Questions

What happens if I use the opt-in command while always-on mode is enabled?

If you trigger /i-have-adhd while the flag file exists, the ruleset remains active as expected, but you will have redundant rule injection since hooks/always-on.sh already prepends the guidelines automatically. To avoid confusion, choose one activation method based on your workflow needs.

Can I use always-on mode with Codex?

The flag file mechanism (~/.claude/.i-have-adhd-always) specifically targets Claude Code's hook system via hooks/always-on.sh. Codex users should rely on the opt-in command $i-have-adhd for session-specific activation, as the automatic injection relies on Claude-specific hook infrastructure defined in hooks/hooks.json.

Why does the always-on hook strip YAML front-matter?

The hooks/always-on.sh script removes the YAML front-matter from skills/i-have-adhd/SKILL.md before injection because the front-matter contains metadata flags like disable-model-invocation: true that control skill loading behavior. Stripping these headers ensures only the actual formatting rules are injected into the conversation context, not the configuration metadata.

Where can I find detailed installation instructions?

Comprehensive setup guides for both activation modes are documented in INSTALL.md at the repository root, while README.md provides quick-start examples. The core logic resides in skills/i-have-adhd/SKILL.md, with automation handled by hooks/always-on.sh and configured through hooks/hooks.json.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}