How to Configure i-have-adhd Always-On for Different AI Agents

Enable always-on mode by creating a flag file for Claude Code, installing the Gemini extension, or appending the markdown rules block to your agent's persistent configuration file.

The i-have-adhd skill enforces ADHD-friendly output rules—concise answers, numbered steps, concrete time estimates, and no preamble. By default, you must invoke it manually with /i-have-adhd. This guide shows how to configure always-on mode so these rules apply automatically to every session across Claude Code, Gemini CLI, Zed, Cursor, and other agents.

---

Claude Code: Flag File Activation

Claude Code uses a session hook that checks for a flag file at startup.

How It Works

The SessionStart hook in hooks/hooks.json triggers hooks/always-on.sh on every new session. The script looks for $CLAUDE_CONFIG_DIR/.i-have-adhd-always (default ~/.claude/.i-have-adhd-always). When present, it reads skills/i-have-adhd/SKILL.md, strips YAML frontmatter, and injects the rules as a system message.

Enable Always-On

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

Disable Always-On

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

Key Files

• hooks/hooks.json – Registers the SessionStart hook
• hooks/always-on.sh – POSIX script that loads SKILL.md when the flag exists
• skills/i-have-adhd/SKILL.md – Core ADHD-friendly rules

---

Gemini CLI: Extension Installation

Gemini CLI offers two routes: a command (manual opt-in) and an extension (always-on).

Install the Always-On Extension

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

The extension's GEMINI.md loads the full skill at startup, making rules active from message 1.

Key File

• gemini-extension.json – Defines the extension metadata and entry point

---

Zed: Personal AGENTS.md

Zed reads ~/.config/zed/AGENTS.md (or project-level AGENTS.md) at session start.

Add the Rules Block

cat >> ~/.config/zed/AGENTS.md <<'EOF'

## Output style

The reader has ADHD. Shape every response so it can be acted on:
1. Lead with the answer or next action: command, path, or snippet first.
2. Number multi-step work; one bounded action per step.
3. End with one next action doable in under two minutes.
4. Finish the current issue before raising a new one.
5. Restate progress each turn ("step 3 of 5 done").
6. Give time estimates in concrete units, never "a bit".
7. After a change, show what now works.
8. Errors: state location, cause, and fix. No drama.
9. Cap lists at 5 items.
10. No preamble, no recaps, no closers.
EOF

---

Hermes: AGENTS.md or SOUL.md

Hermes loads rules from project-level AGENTS.md or your persona file SOUL.md.

Append the same markdown block (shown in Zed section above) to either location. Hermes reads these files each session without requiring manual invocation.

---

Pi: Project AGENTS.md

Pi automatically loads AGENTS.md from your project root. Place the rules block there for always-on behavior across all sessions in that project.

---

Codex: User AGENTS.md

Codex reads ~/.codex/AGENTS.md for every chat. Add the rules block to this file for permanent, global application.

mkdir -p ~/.codex
cat >> ~/.codex/AGENTS.md <<'EOF'

## Output style

The reader has ADHD. Shape every response so it can be acted on:
[... rules 1-10 ...]
EOF

---

Cursor, OpenCode, and Other Harnesses

For agents with persistent rules files, paste the markdown block into the appropriate location:

Agent	Configuration Path
Cursor	~/.cursor/rules/user.rules or ~/.cursor/rules/
OpenCode	~/.config/opencode/AGENTS.md
Antigravity (agy)	~/.gemini/GEMINI.md

Cursor Example

mkdir -p ~/.cursor/rules
cat >> ~/.cursor/rules/user.rules <<'EOF'

## Output style

The reader has ADHD. Shape every response so it can be acted on:
1. Lead with the answer or next action: command, path, or snippet first.
2. Number multi-step work; one bounded action per step.
3. End with one next action doable in under two minutes.
4. Finish the current issue before raising a new one.
5. Restate progress each turn ("step 3 of 5 done").
6. Give time estimates in concrete units, never "a bit".
7. After a change, show what now works.
8. Errors: state location, cause, and fix. No drama.
9. Cap lists at 5 items.
10. No preamble, no recaps, no closers.
EOF

---

Summary

• Claude Code: Touch or remove ~/.claude/.i-have-adhd-always to toggle; hook scripts in hooks/ drive the behavior
• Gemini CLI: Run gemini extensions install for the always-on extension route
• Zed, Hermes, Pi, Codex: Append rules to AGENTS.md (location varies by agent)
• Cursor/OpenCode/others: Paste rules into the harness-specific persistent rules file
• All methods use the same 10-rule block from skills/i-have-adhd/SKILL.md

---

Frequently Asked Questions

What is the i-have-adhd skill?

The i-have-adhd skill is a set of 10 output rules designed for ADHD cognitive styles: lead with answers, number steps, cap lists at 5 items, use concrete time estimates, and eliminate preamble. It is implemented in the ayghri/i-have-adhd repository.

Does always-on mode slow down my agent?

No. The rules are static markdown injected once at session start. For Claude Code, hooks/always-on.sh simply reads a local file and prints its contents—a sub-millisecond operation.

Can I use always-on for some projects but not others?

Yes. Claude Code uses a global flag file, so it's all-or-nothing. For Zed, Pi, and Hermes, place AGENTS.md in specific project directories to enable rules only for those projects. Cursor and OpenCode typically use global rules files.

Where can I find the complete installation instructions?

The INSTALL.md file in the ayghri/i-have-adhd repository contains step-by-step setup for every supported agent, including both command-mode and always-on configurations.

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