Harness for humans with ADHD
Обзор
human-harness a focus harness for a human with ADHD Before / After · Install · How it works · Usage Humans with ADHD are just LLMs with a terrible context window and no system prompt. So I gave myself one. human-harness is a Claude Code skill that keeps you on one task. Agent harnesses feed an LLM one instruction at a time so it doesn't wander off; this does the same for you. It casts a system prompt, breaks your task into the single next action, and shows you one thing at a time. Drift, and it re-injects your system prompt at you. 1. /human-harness ship the auth PR, reply to Sarah 2. Picked from the task. Coding makes you a senior engineer, outreach makes you a salesperson. Fixed for the session. 3. The task becomes one concrete next action, small enough to actually start. Figuring out the next step is the hard part with ADHD, so the harness does it for you. 4. Never the full list. One action, a time estimate, and what is off-limits while you do it. 5. Done moves on.
README
human-harness a focus harness for a human with ADHD
Before / After · Install · How it works · Usage
Humans with ADHD are just LLMs with a terrible context window and no system prompt. So I gave myself one.
human-harness is a Claude Code skill that keeps you on one task. Agent harnesses feed an LLM one instruction at a time so it doesn’t wander off; this does the same for you. It casts a system prompt, breaks your task into the single next action, and shows you one thing at a time. Drift, and it re-injects your system prompt at you.
What it looks like
Before and after
How it works
- You dump a task.
/human-harness ship the auth PR, reply to Sarah - It casts a system prompt. Picked from the task. Coding makes you a senior engineer, outreach makes you a salesperson. Fixed for the session.
- It atomizes. The task becomes one concrete next action, small enough to actually start. Figuring out the next step is the hard part with ADHD, so the harness does it for you.
- It shows you one thing. Never the full list. One action, a time estimate, and what is off-limits while you do it.
- You answer. Done moves on. Stuck breaks the step into something smaller. Not now sends it to the back, no guilt.
- You drift, it pulls you back. Go off-task and it re-injects your system prompt instead of following the tangent.
Install (30 seconds)
One line. Auto-detects your agents (Claude Code, Codex, Gemini, OpenClaw) and installs into each:
curl -fsSL https://raw.githubusercontent.com/dhasson04/human-harness/main/install.sh | sh
Or drop the one file in by hand:
git clone https://github.com/dhasson04/human-harness
mkdir -p ~/.claude/skills/human-harness
cp human-harness/SKILL.md ~/.claude/skills/human-harness/SKILL.md
No key, no config, no account, no dependencies — the agent is the engine. Per-agent paths and the --only / --all / --uninstall flags are in INSTALL.md.
Usage
| command | what it does |
|---|---|
/human-harness |
break a task down and run it, one action at a time |
/human-harness |
no task? it asks what you want done |
/human-harness about |
your model card |
The persona adapts to the work. do my outreach makes you a salesperson; write the post makes you a writer.
FAQ
Do I need an API key? No. It runs inside Claude Code, which is already the model.
Does it store my tasks? No. Nothing is saved. The task lives in the session and that is it.
Is it only for developers? No. The system prompt adapts to whatever you hand it.
License
MIT. Do whatever, just ship something.
Рекомендуемые инструменты
Попробуйте другой запрос или уберите фильтр.
Установка
npx skillfish add dhasson04/human-harness